EP 108 | Philip Joris | HemoVision, 3D Bloodstain Pattern Analysis

HemoVision | EP108 | Philip Joris | Jun 5 2025 | 2 PM Eastern

Join us for a deep dive into the intersection of forensic science, Bloodstain Pattern Analysis and Machine Learning with Philip Joris, co-founder of Forentrics and developer of HemoVision, an advanced tool for bloodstain pattern analysis.

In this episode, we’ll explore how Philip’s background in electronics, AI, and medical imaging led to the creation of powerful computational tools for crime scene investigation. Learn how HemoVision supports investigators in reconstructing bloodstain patterns, enhancing objectivity, and visualizing complex scenes with cutting-edge technology.

Don’t miss this conversation on the future of digital forensics, the evolution of BPA, and the role of AI in modern crime scene reconstruction.

Originally Aired on: June 5, 2025

Transcript

00;00;00;00 - 00;00;29;29
Unknown
No.

00;00;29;29 - 00;00;31;16
Unknown
welcome to Forensics Talks.

00;00;31;16 - 00;00;42;19
Unknown
This is episode 108, and my guest today is Phillip. Yours. And Phillip is from Belgium, and he's the co-founder of forensics. And it's a startup,

00;00;42;19 - 00;00;53;00
Unknown
developing advanced software for crime scene investigation. He holds a master's degree in areas like electronics and artificial intelligence. And he's earned a PhD in electrical engineering.

00;00;53;03 - 00;00;53;20
Unknown
During his,

00;00;53;20 - 00;01;12;17
Unknown
doctoral research, he developed some new techniques for bloodstain pattern analysis and also virtual autopsies. I definitely want to ask him about that. And all of this work led to the creation of HemoVision. Okay, so the company is forensics, and this is their flagship tool. And it's something that I'm quite familiar with and something that we're going to be focusing on today.

00;01;12;17 - 00;01;29;17
Unknown
So let me go ahead and bring him in here. There he is Philip right. Hygiene. All right. So full disclosure for everybody. So I've known Philip for at least I don't know. It's got to be like seven, seven, eight years. So I'm not sure what we're going. Yeah. Do. Yeah. Okay.

00;01;29;17 - 00;01;32;20
Unknown
So and we've worked quite a bit together, so,

00;01;32;20 - 00;01;34;08
Unknown
you know, sometimes when I bring,

00;01;34;08 - 00;01;36;12
Unknown
people in here, I don't really know them very well.

00;01;36;13 - 00;01;45;07
Unknown
But I know you fairly well, and I think you know me very well. We've done workshops together, we've done a bunch of stuff, and I keep I keep asking myself, when did I actually first meet you?

00;01;45;07 - 00;01;47;14
Unknown
Well, in the IABPA in Canada.

00;01;47;14 - 00;01;52;14
Unknown
In Ottawa. So that was a what was that, 2017 or something like that? I think I just under 18.

00;01;52;14 - 00;02;11;01
Unknown
Yeah. Well, 1818 was Paris wasn't it wasn't or was it 2019 was Paris. Was the IBM conference I don't know, I remember we went for a walk in Paris and or at the, at the conference there. We had a good chat at lunch. So anyway, I'll get it. All good. All right, well, let's get started here, I want to ask you about your background and,

00;02;11;01 - 00;02;12;16
Unknown
how this all got started.

00;02;12;16 - 00;02;15;05
Unknown
And obviously, you know, electrical engineering.

00;02;15;05 - 00;02;19;18
Unknown
Hey, just so you know, I did two years of electrical engineering, and then I couldn't stand it, so I switched over to,

00;02;19;18 - 00;02;22;27
Unknown
aerospace. And that was a lot more interesting for me. So,

00;02;22;27 - 00;02;25;24
Unknown
but I know what you have to go through in electrical.

00;02;25;24 - 00;02;28;25
Unknown
But can you kind of walk me through, you know what?

00;02;28;27 - 00;02;29;08
Unknown
Which

00;02;29;08 - 00;02;34;01
Unknown
maybe what you thought you were going to be doing at first and then, you know, kind of how you ended up, you know,

00;02;34;01 - 00;02;37;07
Unknown
coming out of this with, with HemoVision, let's say. Yeah,

00;02;37;07 - 00;02;38;19
Unknown
a long journey. Long journey.

00;02;38;19 - 00;02;39;07
Unknown
As you said,

00;02;39;07 - 00;02;47;05
Unknown
I'm indeed an engineer in electronics and ICT, and I think it was during those study that that I actually, kind of developed a love for,

00;02;47;05 - 00;02;55;05
Unknown
programing and more specifically like computer vision and teaching computers how to understand stuff and, and make decisions.

00;02;55;07 - 00;02;56;24
Unknown
And after that, Masters, I,

00;02;56;24 - 00;03;08;17
Unknown
I decided to do another master's specializing in artificial intelligence to deepen my knowledge in that field. And at the time, while I was studying, I was actually watching a very popular VPA series called Dexter.

00;03;08;17 - 00;03;11;29
Unknown
Pretty sure a lot of people will be familiar with that.

00;03;11;29 - 00;03;18;24
Unknown
At the beginning of the year, all of the students, they get a list of a lot of thesis topics that they have to select a thesis topic from.

00;03;18;27 - 00;03;30;01
Unknown
And one of the topics in the list said, hey, we want to do something with bloodstain pattern analysis and artificial intelligence. We're not sure what we want to do, but we want to do something with it. And I was like,

00;03;30;01 - 00;03;33;14
Unknown
I'm in love with the series. Why not go for it?

00;03;33;14 - 00;03;35;22
Unknown
And yeah, that's basically my first,

00;03;35;22 - 00;03;42;03
Unknown
steps in in AI and combining it to, with, with forensics.

00;03;42;06 - 00;03;45;11
Unknown
And yeah, things just grew from there. I,

00;03;45;11 - 00;03;55;12
Unknown
I finished my, my masters and my, my promoters at the time, they were actually like, hey, are you interested in doing a PhD where you work on the overlap between forensics and,

00;03;55;12 - 00;03;57;20
Unknown
engineering and I was like, yeah, hell yeah.

00;03;57;20 - 00;03;59;07
Unknown
And that's that's how it all got started.

00;03;59;07 - 00;04;02;07
Unknown
Yeah. Excellent. So and I saw that,

00;04;02;07 - 00;04;09;11
Unknown
you know, you had some work with this virtual autopsy stuff, so what was that about? How did that get started? Yeah. So part of the PhD was on,

00;04;09;11 - 00;04;10;11
Unknown
bloodstain patterns.

00;04;10;11 - 00;04;12;23
Unknown
And the other part was indeed on virtual autopsies.

00;04;12;23 - 00;04;13;29
Unknown
So the goal was to,

00;04;13;29 - 00;04;19;17
Unknown
instead of using the traditional invasive autopsy where you put a body on a table and actually just cut it open and look for stuff,

00;04;19;17 - 00;04;20;13
Unknown
maybe we can,

00;04;20;13 - 00;04;21;07
Unknown
do some,

00;04;21;07 - 00;04;23;22
Unknown
searching for anomalies on CT scans.

00;04;23;25 - 00;04;25;28
Unknown
So, in hospital where I,

00;04;25;28 - 00;04;26;20
Unknown
worked,

00;04;26;20 - 00;04;28;03
Unknown
it's quite common for,

00;04;28;07 - 00;04;38;18
Unknown
bodies to be scanned in the CT scanner. And I was developing an approach that was general enough to detect, like, any type of anomaly, not focusing on one specific anomaly, like,

00;04;38;18 - 00;04;42;09
Unknown
specific type of tumor or something. But I was looking at a more general approach.

00;04;42;09 - 00;04;42;29
Unknown
And,

00;04;42;29 - 00;04;51;16
Unknown
the goal there was to apply it to, gunshot trauma because it's very diverse. There's there's bone fragments, there's air bubbles, there's,

00;04;51;16 - 00;04;57;18
Unknown
pieces of metal from the actual bullet. So the appearance of the anomalies is very, very diverse.

00;04;57;18 - 00;05;08;28
Unknown
And and in the end, I developed a method that could detect all of these anomalies and then try to link them all together to determine the trajectory of the bullet inside of the body.

00;05;09;01 - 00;05;17;05
Unknown
Yeah. Fun. We had a lot of fun experiments. Yeah. Super cool. And actually, you did, I think I remember hearing about this from Irwin.

00;05;17;05 - 00;05;20;04
Unknown
Yeah. We're just. Yeah, he's retired since, but great guy. And,

00;05;20;04 - 00;05;21;23
Unknown
so you had to shoot at all these,

00;05;21;23 - 00;05;23;13
Unknown
yeah. So we went to,

00;05;23;13 - 00;05;25;14
Unknown
to a local butcher in the morning at 5 a.m..

00;05;25;14 - 00;05;27;01
Unknown
We picked up some sheep heads.

00;05;27;01 - 00;05;28;29
Unknown
And then we put them through the CT scanner.

00;05;28;29 - 00;05;39;11
Unknown
We went to shoot at them with a low caliber rifle, and then we scanned them again. So we had a before and after. And that was basically my training data or my testing data on which I could develop and test my my algorithms.

00;05;39;11 - 00;05;46;14
Unknown
Yeah. The things we do for forensic science. Yeah, yeah. And there's always a butcher involved for some reason, I don't know.

00;05;46;14 - 00;05;48;15
Unknown
Yeah. And there were still nice and warm. It's,

00;05;48;15 - 00;05;49;13
Unknown
very special.

00;05;49;13 - 00;05;52;16
Unknown
Memory. Yeah. Right.

00;05;52;16 - 00;05;54;09
Unknown
Well, let's talk about,

00;05;54;09 - 00;05;57;02
Unknown
let's get into the bloodstain, pattern analysis and,

00;05;57;02 - 00;06;01;07
Unknown
like some of the things that you were working on initially.

00;06;01;09 - 00;06;15;15
Unknown
So before, like, just before. Well, let me ask you this, like, at what point did it sort of hit you that you said, you know what? I'm going to start? I'm going to start this project where I'm going to put together this bloodstain, you know, pattern software. And,

00;06;15;15 - 00;06;18;11
Unknown
so, you know, what was the trigger that said, hey, you know, we need something like this.

00;06;18;11 - 00;06;24;06
Unknown
Like, were you talking to people in the, in the area? And they said, hey, you know, we really could use a new tool like this or,

00;06;24;06 - 00;06;26;25
Unknown
how did that come to be? Yeah. So the,

00;06;26;25 - 00;06;35;13
Unknown
the starting point, as I said, was the the master's thesis. And this was in close collaboration with the forensic department of Liver in Belgium.

00;06;35;15 - 00;06;37;06
Unknown
And at the time there was else,

00;06;37;06 - 00;06;38;25
Unknown
she was at the time a bloodstain,

00;06;38;25 - 00;06;43;25
Unknown
expert. And she was actually doing. Yeah, crime scene investigations, doing the BPA,

00;06;43;25 - 00;06;51;11
Unknown
talking with her and and learning what it was all about. I try to develop methods that could actually help her do a better job,

00;06;51;11 - 00;06;55;01
Unknown
make her life easier as a BPA expert.

00;06;55;03 - 00;07;00;09
Unknown
And at the time, it was just like, hey, this is cool from a research perspective. Like, how can we,

00;07;00;09 - 00;07;07;26
Unknown
take certain steps of the whole BPA process and, and use AI to automate this or make it more accurate or more objective?

00;07;07;26 - 00;07;16;02
Unknown
That was from a research perspective, like really cool. But then we started interacting with like the federal police in Belgium.

00;07;16;05 - 00;07;17;11
Unknown
They also saw,

00;07;17;11 - 00;07;27;05
Unknown
a lot of potential in the software. I also did some conferences. So like my first IBM meeting was in Rome, 2014, if I'm correct.

00;07;27;05 - 00;07;34;02
Unknown
And as a newcomer, it was all very overwhelming to see all of the people doing doing their amazing BPA work.

00;07;34;02 - 00;07;48;23
Unknown
But the feedback I got on, on even the very earliest steps of HemoVision was so overwhelmingly positive that I said, okay, maybe we should take another look at this and develop it a little bit further and make an actual product out of it.

00;07;48;25 - 00;07;54;23
Unknown
Yeah. And I think what's interesting is, you know, there I don't think there's very many electrical engineers that are, you know,

00;07;54;23 - 00;08;10;21
Unknown
people studying electrical engineering that are into the whole BPA world. So I think bringing in a fresh perspective like that is really helpful. So what were the whether any things at the beginning that you found out about, you know, BPA or whatever, that sort of open your eyes and said, oh my God.

00;08;10;21 - 00;08;15;12
Unknown
Like there's opportunities here or maybe some things that may have surprised you or,

00;08;15;12 - 00;08;32;17
Unknown
maybe something that you said, oh, man, I, you know, I can I can solve this problem, you know, I mean, like, I can do something here. Like what? What were some of those things maybe that hit you early on? Yeah. There's a there's quite a few things, but, the thing that struck me most was that a lot of the work is still,

00;08;32;17 - 00;08;35;16
Unknown
highly dependent on manual work, like there's a lot of manual steps.

00;08;35;16 - 00;08;38;21
Unknown
And when talking to else, like the expert from Nova,

00;08;38;21 - 00;08;46;10
Unknown
they had to do so much work just to get, for example, to the result from an area of origin analysis, they had to perform so many measurements.

00;08;46;10 - 00;08;52;04
Unknown
When they initially explained stringing to me, I was like, wow, this is this is weird. Why would you actually do this?

00;08;52;04 - 00;08;54;06
Unknown
I mean, there's got to be a better approach.

00;08;54;06 - 00;08;58;07
Unknown
Then I learned that people don't actually string anymore. They use tangent method, etc.

00;08;58;07 - 00;09;04;20
Unknown
but still still, it's it's a lot of work getting getting to a single result. And I thought, well,

00;09;04;20 - 00;09;11;07
Unknown
maybe we can automate certain parts of it and make life easier. Yeah. So and I think,

00;09;11;07 - 00;09;25;21
Unknown
or talk to me about the, the automated process or the simplification of the documentation process and HemoVision because you take a different approach than other softwares, which are, let's say, a lot more manual.

00;09;25;27 - 00;09;27;26
Unknown
So maybe can you describe what you're doing there?

00;09;27;26 - 00;09;30;07
Unknown
In emotion. Yeah, sure. So,

00;09;30;07 - 00;09;43;21
Unknown
whenever we're doing, for example, an area for analysis, we want to analyze the convergence of multiple trajectories from multiple blood stains. So we need to place them all in a you know, same frame of reference. They have to be in the same space.

00;09;43;23 - 00;09;52;19
Unknown
And what people typically do is they measure the coordinates, like they take a reference point somewhere, the corner of the room, and then they just measure what the horizontal and vertical offset is.

00;09;52;19 - 00;10;01;14
Unknown
But instead of doing all this measuring, I thought, let's just reconstruct the whole scene in 3D and get the measurements automatically in this 3D space.

00;10;01;14 - 00;10;04;18
Unknown
And that's exactly what we're doing with HemoVision. So,

00;10;04;18 - 00;10;16;25
Unknown
we just we rely on photogrammetry, which I think a lot of people are familiar with. So we take a few photographs from different perspectives. And from these photographs, we can then reconstruct parts of the crime scene in three dimensions.

00;10;16;25 - 00;10;21;06
Unknown
And we can take accurate or fairly accurate measurements in this space.

00;10;21;08 - 00;10;23;16
Unknown
Yeah. One of the aspects. Yeah, you have

00;10;23;16 - 00;10;27;25
Unknown
so, for example, you have these markers that you place on the wall like, so you have,

00;10;27;25 - 00;10;33;28
Unknown
like there's one larger one and then there's like these little ones. So can you talk to me about these markers and what their function is? Yeah. So

00;10;33;28 - 00;10;36;23
Unknown
when we initially started to develop methods,

00;10;36;23 - 00;10;41;25
Unknown
for the reconstruction, we said, well, photogrammetry is a great example of how we can achieve,

00;10;41;25 - 00;10;45;26
Unknown
a good degree of accuracy regarding 3D reconstructions.

00;10;45;28 - 00;10;54;19
Unknown
But there are certain cases where photogrammetry runs into difficulties. And this is the case, for example, when there's very little texture on the walls,

00;10;54;19 - 00;11;04;22
Unknown
when you have plain white walls. And that's typically what we have in a lot of homes, people just paint your walls in a single color. And the bloodstains are too small to to give you sufficient,

00;11;04;22 - 00;11;07;09
Unknown
texture and key points to reconstruct.

00;11;07;11 - 00;11;16;07
Unknown
And then we said, let's, let's make the approach more robust, because we want to make it so that it's applicable in each and every case, and not just 50% of the cases.

00;11;16;07 - 00;11;22;14
Unknown
So we have this little checkerboard pattern state resemble somewhat somewhat what you have in the back in your background.

00;11;22;14 - 00;11;23;25
Unknown
And we place a few,

00;11;23;25 - 00;11;30;24
Unknown
in and around the spatter pattern, and then the software can accurately detect those and use those to reconstruct,

00;11;30;24 - 00;11;33;00
Unknown
the scene in a more robust way.

00;11;33;02 - 00;11;36;23
Unknown
They have different functions as well. They provide scaling information.

00;11;36;23 - 00;11;38;10
Unknown
You can also take photographs,

00;11;38;10 - 00;11;40;21
Unknown
freehand. So you don't have to be,

00;11;40;21 - 00;11;43;25
Unknown
perfectly orthogonal to the plane, like, frontal parallel.

00;11;43;25 - 00;11;52;26
Unknown
Any perspective distortions are corrected using these markers. So they have a lot of functions and they help us a lot. Yeah. To achieve this automated automation.

00;11;52;28 - 00;12;08;07
Unknown
Yeah. So, so the, the user or the person that's setting up the crime scene, the basic process is going to be to place those markers, in areas of importance like where the blood scenes are. And then now in terms of measurements, how many measurements do they need to take then?

00;12;08;07 - 00;12;10;12
Unknown
Typically we we only require one.

00;12;10;12 - 00;12;11;26
Unknown
So we have one larger,

00;12;11;26 - 00;12;13;15
Unknown
overview marker as we call it.

00;12;13;15 - 00;12;14;11
Unknown
And it's,

00;12;14;11 - 00;12;17;22
Unknown
used to position the patron as a whole in 3D space.

00;12;17;22 - 00;12;19;14
Unknown
And you just take one,

00;12;19;14 - 00;12;21;26
Unknown
reference measurements to this overview marker,

00;12;21;26 - 00;12;34;17
Unknown
from your chosen reference point, for example, the corner of the room. And this is just to position the patron as a whole in, in 3D space. So all the other than that measurement, there's no manual measurement involved.

00;12;34;19 - 00;12;35;29
Unknown
Okay.

00;12;35;29 - 00;12;38;01
Unknown
Part of the process that you have, though is,

00;12;38;01 - 00;12;49;17
Unknown
like you have to take photos for the overall pattern like that. The like there's like an overview marker that you have or you want to get all those markers in, but then you also allow individual photos for like those smaller,

00;12;49;17 - 00;12;52;15
Unknown
detail markers. So you can see the smaller scenes or whatever.

00;12;52;15 - 00;12;53;06
Unknown
So

00;12;53;06 - 00;13;02;04
Unknown
like can you give any recommendation on the like the number of photographs or the, the way that you would photograph for the overview versus the detail and like,

00;13;02;04 - 00;13;05;04
Unknown
how many photos are we talking about usually? Yeah, sure.

00;13;05;04 - 00;13;13;02
Unknown
So for a fairly normal, like a normal scene, just a few corners, the corner of a room or a few surfaces.

00;13;13;05 - 00;13;28;02
Unknown
Because this is, this is one of the new aspects of HemoVision. So you can place these markers on all kinds of surfaces. They can be arbitrarily angled. And he my vision will reconstruct them all at in the same time. We recommend taking like 15 to 20 photographs maybe.

00;13;28;02 - 00;13;32;19
Unknown
And you don't need to capture all markers in all photographs.

00;13;32;21 - 00;13;36;01
Unknown
They just need to be each marker needs to be in a few photographs.

00;13;36;01 - 00;13;42;09
Unknown
So let's say 15 to 20 overview images. And then you just take as many detail images as you need.

00;13;42;09 - 00;13;43;21
Unknown
I typically say that,

00;13;43;21 - 00;13;50;16
Unknown
detail image covers roughly about a 50 by 50 centimeter region. So half a meter by half a meter,

00;13;50;16 - 00;13;53;16
Unknown
depending on the resolution of your camera, of course.

00;13;53;19 - 00;14;01;12
Unknown
And then you can take 1 or 2, detail images per detail marker. And I would say for the most scenes that I've processed,

00;14;01;12 - 00;14;16;01
Unknown
between like ten detail markers, maybe 15, if it's a very, very complex language, very where a lot of different surfaces. So all in all, maybe 30 or 40 photographs. Okay. So again, between taking the measurements,

00;14;16;01 - 00;14;23;20
Unknown
and taking 30, let's say 30 photographs or something like that, then, you know, it might take you five minutes, right, like 5 minutes or 10 minutes or something.

00;14;23;20 - 00;14;31;26
Unknown
So it's a pretty okay. So that's I mean, that's significantly reduced time to document this pattern, which is, which is pretty great. So that's awesome. Yeah. It's

00;14;31;26 - 00;14;35;11
Unknown
it's a quite fast process. And it also has the advantage that,

00;14;35;11 - 00;14;43;25
Unknown
the person who takes the photographs doesn't necessarily need to have like the most expertise about obpa. So it can be somebody else that takes the photographs.

00;14;43;25 - 00;15;01;08
Unknown
They just need to know how to place the markers, take the photographs, and then somebody else can do the analysis in my vision. So all the information that is needed for the analysis is in the photographs because of the markers, because of the way we take the photographs. Right. And I guess you should be clear, like this is you're talking about strictly for the the pattern analysis.

00;15;01;08 - 00;15;06;04
Unknown
It's not a replacement for the regular crime scene photo. Of course, anything with the air course. Okay.

00;15;06;04 - 00;15;10;18
Unknown
And then you take all this, you take the images, you bring it into the software, and,

00;15;10;18 - 00;15;16;02
Unknown
the process is, is just. I mean, you're setting up a project. I mean, how long does it take to process something like that on a computer?

00;15;16;06 - 00;15;17;21
Unknown
On a decent computer?

00;15;17;21 - 00;15;19;27
Unknown
Well, it kind of depends on how experienced you are.

00;15;19;27 - 00;15;26;02
Unknown
I, I a few months ago, I made a video where I processed the scene in, like, two, four minutes, I believe.

00;15;26;02 - 00;15;29;03
Unknown
It's a bit more complex. Let's say 15 minutes. 20 minutes.

00;15;29;03 - 00;15;31;03
Unknown
Depending on how many blood stains you select.

00;15;31;03 - 00;15;32;27
Unknown
It's a fairly fast process.

00;15;32;27 - 00;15;34;21
Unknown
Yeah. For. Yeah. And the. But the image,

00;15;34;21 - 00;15;55;12
Unknown
on the, the structure for motion, the photogrammetry part there, I mean that can be done usually in a few minutes. Yeah, sure. Yeah. That's pretty quick. Okay. Cool. Let me ask you about bloodstain marking. Because part of the process traditionally has been that the analyst is going to take an ellipse and they're going to take this ellipse shape, and they're going to mark it over a blood stain, you know,

00;15;55;12 - 00;15;59;05
Unknown
fit it to the, you know, get the proper angle and all that sort of thing.

00;15;59;08 - 00;16;00;02
Unknown
So,

00;16;00;02 - 00;16;04;27
Unknown
there's something that you did in a section on your website too, and I should probably bring it back up here, but,

00;16;04;27 - 00;16;07;28
Unknown
it's a little bit of a different approach.

00;16;07;28 - 00;16;10;00
Unknown
To marking. And some of it is,

00;16;10;00 - 00;16;15;09
Unknown
automated and some of it is like semi-automated. So like it's under here under news at the bottom, but here.

00;16;15;12 - 00;16;34;07
Unknown
Yeah, here it is. So it's a statistical impact stain analysis. So this this has something called a shape model. And you've written a paper on this as well. So you've published the paper. So yeah. Explain to me how this is different or explain to everybody how this is different than, you know, the regular old, you know, marking an ellipse as best as you can get it.

00;16;34;09 - 00;16;34;26
Unknown
Yeah. So,

00;16;34;26 - 00;16;36;09
Unknown
this this was actually part of the,

00;16;36;09 - 00;16;39;29
Unknown
the early works that we did on HemoVision. So, like, the core algorithms.

00;16;39;29 - 00;16;51;14
Unknown
So as you, as you said, traditionally people just assume that a droplet is perfectly spherical. And when you projected onto a flat surface, it becomes a perfect ellipse. Obviously, we know that there's different factors involved.

00;16;51;17 - 00;17;00;23
Unknown
When the blood droplet oscillates in the air and contact comes in contact with the surface, etc., and we said like, hey, instead of assuming the shape of,

00;17;00;23 - 00;17;13;18
Unknown
blood stained to be elliptical, why don't we learn the shape of a blood stain and then link it to its impact angle? So what we did is we created a database of, I think, a little over 400 blood stains.

00;17;13;20 - 00;17;29;01
Unknown
And for each of these stains, we knew the exact angle at which it impacted the surface. So we know that this shape of this blood stain corresponds to this angle. And then we use something called principal component analysis, which is basically just,

00;17;29;01 - 00;17;31;16
Unknown
dimensional reduction technique that,

00;17;31;16 - 00;17;35;09
Unknown
can be used to learn how the shape varies across your data set.

00;17;35;09 - 00;17;59;19
Unknown
So we sampled the outline of each of the blood stains was just a contour of the blood stain. And we learned how this contour changes with impact angle and the approach, the advantages that we do not need to assume anything. The data will tell us what a blood stain looks like. And then we can apply this blood stain and transform it and deform it so it fits onto a new blood stain in a new image.

00;17;59;19 - 00;18;10;10
Unknown
And then we can get its impact angle from it. How do you handle blood stains, which sometimes are malformed on the back end. So you may have a really long tail. You get all these little, you know, defects sometimes, you know, you hard to get,

00;18;10;10 - 00;18;13;12
Unknown
you know, the front end is really nice. And then the back end is all a mess.

00;18;13;12 - 00;18;27;18
Unknown
So is there some how did you approach that? Yeah. So that's one of the difficulties that that we have with ellipses. So typically when people mark the front and the back to the front it's very clear. It's the tip of the blood stain. But the back is not clear.

00;18;27;18 - 00;18;30;18
Unknown
There's some kind of. Yeah. Region,

00;18;30;18 - 00;18;33;01
Unknown
where it's not very clear where the blood stain should end.

00;18;33;03 - 00;18;39;21
Unknown
So our model is not really a closed contour. It's like an open ended contour or like an inverted U-shape.

00;18;39;21 - 00;18;58;03
Unknown
So it basically fits around the blood stain by primarily focusing on the front of the blood stain and somewhat automatically ignoring the tail information. Because the tail for the impact angle has no relevant information, it can be used for the direction, but for the impact angle it holds no information.

00;18;58;03 - 00;19;04;29
Unknown
So the model automatically ignores details. And that's just because it's open ended. Okay, so were there any

00;19;04;29 - 00;19;11;07
Unknown
I'm just curious now because we're talking about, you know, machine learning and I don't know, somehow I think I'm know enough to slip in the AI word, but,

00;19;11;07 - 00;19;11;24
Unknown
yeah.

00;19;11;24 - 00;19;17;08
Unknown
But were there any I just, you know, for out of curiosity, were there any challenges with working with data sets like that?

00;19;17;11 - 00;19;19;01
Unknown
With teaching it? I mean,

00;19;19;01 - 00;19;23;28
Unknown
in reading here, I mean, you have to do 2.5 degree increments, right? You need a lot of images.

00;19;23;28 - 00;19;28;10
Unknown
I mean, is this something that you can continually update? Do you feel like you can keep throwing more data to it, like,

00;19;28;10 - 00;19;32;17
Unknown
what were some of the challenges with, with using this type of technique?

00;19;32;19 - 00;19;34;24
Unknown
Yeah. So I think the technique we used,

00;19;34;24 - 00;19;35;19
Unknown
is more like,

00;19;35;19 - 00;19;43;00
Unknown
it's more like statistics. So with only, let's say of 400 examples, we could get really accurate results.

00;19;43;00 - 00;19;52;05
Unknown
In hindsight, we probably could have used a little less data because the technique we use is so powerful and it's just basically pure statistics. I think if you're,

00;19;52;05 - 00;19;58;21
Unknown
going to move to different models, like if you're throwing in VA AI word, if we're talking about deep learning, for example,

00;19;58;21 - 00;20;00;14
Unknown
we've done some experience with that as well.

00;20;00;14 - 00;20;03;09
Unknown
400 examples is not going to get you anywhere.

00;20;03;09 - 00;20;18;14
Unknown
So you would need a lot more and you would need to apply some data augmentation. And I think there's other challenges that there as well. I think one of the the key values that we have is trying to have explainable models that we can explain how they get to a result and why they get to a result.

00;20;18;14 - 00;20;22;24
Unknown
And that's something that's very more difficult with deep learning.

00;20;22;24 - 00;20;27;13
Unknown
So in terms of data collection, it was actually quite straightforward.

00;20;27;13 - 00;20;30;08
Unknown
And one of the difficulties we actually had was,

00;20;30;08 - 00;20;36;04
Unknown
trying to convince people that it actually worked. Oh, really? Yeah. One of the I think one of the first comments,

00;20;36;04 - 00;20;43;04
Unknown
the very first presentation I ever gave about this model, I think, was at the IBM, where one of the first comments was, why would you do this?

00;20;43;04 - 00;20;59;26
Unknown
We've been doing ellipses for over 60 years. There's no real answer. I said, well, I just want to try something different. That was my answer, I think. But anyway, yeah, that was because that was before GPT. That's why it was after GPT. People would have been more accepting of it. I think. So,

00;20;59;26 - 00;21;03;21
Unknown
well, one of the I mean, one of the functions and it's not,

00;21;03;21 - 00;21;10;09
Unknown
something that we typically see in, in software right now, but one of the main functions of, of, let's say pattern analyst is pattern classification.

00;21;10;09 - 00;21;14;14
Unknown
And since we're talking about, you know, this, this sort of thing, have you thought about, you know,

00;21;14;14 - 00;21;23;24
Unknown
tools that could assist a, you know, an investigator on classifying different types of patterns? How would that work? Well,

00;21;23;24 - 00;21;27;05
Unknown
funny you mentioned, but we're currently doing some research on that, obviously.

00;21;27;05 - 00;21;34;05
Unknown
So we've we're looking at how can we support the BPA, in classifying,

00;21;34;05 - 00;21;40;16
Unknown
for as long as I've been going through these BPA conferences, there's always been discussion about classification because everybody does it differently.

00;21;40;22 - 00;21;41;11
Unknown
Everybody,

00;21;41;11 - 00;21;42;14
Unknown
documents their,

00;21;42;14 - 00;21;49;19
Unknown
patterns differently. I think everybody has their own standard operating procedures. And we want to see if there's some way,

00;21;49;19 - 00;22;03;24
Unknown
we can support the forensic expert. And I think if you're dreaming about AI, you could just throw in a picture of a random bloodstain pattern and it's going to say, hey, this is the type of bloodstain pattern that you're talking about.

00;22;03;24 - 00;22;11;01
Unknown
This is probably the mechanism for all kinds of information, but I think maybe that's that's going a bit too fast. I think there's there's more

00;22;11;01 - 00;22;13;19
Unknown
I think there's intermediate steps we can take.

00;22;13;19 - 00;22;17;04
Unknown
One of the things for, for example, looking at Ozark has like this, this perfect,

00;22;17;04 - 00;22;28;02
Unknown
decision tree that can help you to, yeah, go through all of the characteristics of, let's name patterns and come to a conclusion based on a standardized document that's accepted by a lot of people.

00;22;28;05 - 00;22;50;00
Unknown
I think if we can develop tools that include these kinds of documents for experts, I think that's that's a pretty good first step. Okay. So like workflows and guidelines and sort of walking through a process of of going through. And so there's this. Yeah. So at first it would be it's not like hands off the you know the machine does everything and the computer is doing everything.

00;22;50;00 - 00;22;51;28
Unknown
It's interactive and

00;22;51;28 - 00;22;58;13
Unknown
really assisting more than taking over. And I think this can then be combined with other techniques like,

00;22;58;13 - 00;23;00;07
Unknown
a lot of people, I think just,

00;23;00;07 - 00;23;10;14
Unknown
they, they dump all their photographs in a report, like a plain PDF document. And I think that's fine that there's they are typically very descriptive in, in how to describe the patterns.

00;23;10;17 - 00;23;14;04
Unknown
But if you can combine everything into a, the 3D environment where you locate,

00;23;14;04 - 00;23;27;19
Unknown
all of these patterns in their correct location and you can do fly through and maybe use virtual reality to bring this to court. I think there's a lot of different options there as well. What kinds of patterns,

00;23;27;19 - 00;23;30;27
Unknown
are you able to do with HemoVision?

00;23;30;29 - 00;23;34;20
Unknown
So right now, I think, I think we tested,

00;23;34;20 - 00;23;36;16
Unknown
our methods on pretty much all,

00;23;36;16 - 00;23;40;01
Unknown
projected bloodstain patterns, I mean, airborne ones.

00;23;40;01 - 00;23;44;16
Unknown
So we initially started with spatter patterns resulting from an impact,

00;23;44;16 - 00;23;50;25
Unknown
to do area of origin analysis. We also tested these techniques on cessation patterns,

00;23;50;25 - 00;23;54;17
Unknown
accelerated patterns, because they're all, projected blood with an origin.

00;23;54;17 - 00;23;56;01
Unknown
You could say,

00;23;56;01 - 00;23;59;03
Unknown
and then we're also recently,

00;23;59;03 - 00;24;06;28
Unknown
we've been developing the cast off module. So whenever you have a bloody object and you swing it, blood,

00;24;06;28 - 00;24;20;16
Unknown
will typically be released from this object, and it deposits like a linear pattern. And we're now developing a method that from this pattern can actually reconstruct a swing trajectory, not just the swing plane, but also the trajectory.

00;24;20;19 - 00;24;23;20
Unknown
So we've implemented this as a beta,

00;24;23;20 - 00;24;38;04
Unknown
featuring HemoVision, and we're still doing some validation to make sure that everything is working correctly and that we have known error rates and those kinds of stuff. Okay, I got a sample project here. So I want to bring this up on screen so people can see here,

00;24;38;04 - 00;24;39;14
Unknown
what I'm what I'm talking about.

00;24;39;14 - 00;24;40;07
Unknown
This one was a,

00;24;40;07 - 00;24;44;22
Unknown
you can see the images. So folks, if you're looking at my screen, actually, let me let me shrink this down a bit.

00;24;44;22 - 00;24;54;15
Unknown
You can see that there's some images here so I can flip through these images. So this is interesting because this is not a typical, you know, flat wall and, you know, easy to measure from.

00;24;54;15 - 00;25;09;13
Unknown
We have, you know, standard objects here, things on different planes and such. So if you had to measure that by hand becomes very difficult. So yeah, this is yeah, this is like a would you call it a multi-surface project kind of thing. Yeah. Okay. Yeah. It's really funny that you mentioned that because,

00;25;09;13 - 00;25;11;09
Unknown
when I was reading all the literature,

00;25;11;09 - 00;25;15;22
Unknown
most of the experiments are people creating these kinds of patterns on a single vertical wall.

00;25;15;29 - 00;25;22;09
Unknown
And that's how we developed HemoVision. And we were like, we've got it all down, we've got it developed, we've got a nice interface,

00;25;22;09 - 00;25;27;17
Unknown
let's get it to the world. And then I had the opportunity to attend to crime scenes,

00;25;27;17 - 00;25;32;27
Unknown
with the federal police in Brussels. And then I get to my first crime scene and there's just blood everywhere.

00;25;32;29 - 00;25;47;03
Unknown
It's on. All different kinds of surfaces, tables, walls. And I was like, well, this is going to be difficult. And then we really started developing and and this is, this is the result from. Yeah, from that experience. Yeah. Excellent. And

00;25;47;03 - 00;25;52;15
Unknown
like you can flip through, I can see you, you know, you can flip to a 2D to a 3D view and that sort of thing.

00;25;52;20 - 00;25;53;06
Unknown
So,

00;25;53;06 - 00;25;58;29
Unknown
let me, let me bring up another one here because this is, this is a cast off. Let me go to the 3D view here.

00;25;58;29 - 00;26;01;17
Unknown
But this is one that we actually did,

00;26;01;17 - 00;26;02;16
Unknown
together at,

00;26;02;16 - 00;26;04;14
Unknown
that actually was in Kansas City and,

00;26;04;14 - 00;26;08;12
Unknown
yeah, yeah. IBP in Kansas City. So this was actually a cast off pattern.

00;26;08;12 - 00;26;10;10
Unknown
Right? So and,

00;26;10;10 - 00;26;29;24
Unknown
well, what can you tell me about what's going on? You explain it to to what people are seeing here. Yeah. So the red lines are and the estimated linear trajectories of the bloodstains and then the tube as you. Well, I think it's been coined noodle at the conference. So let's call it the noodle for now.

00;26;29;27 - 00;26;35;25
Unknown
The noodle actually represents part of the, the flight path that was made by the, the swinging object.

00;26;35;25 - 00;26;46;06
Unknown
I think if you do some more stains on top, the the tube will expand. But there's this is definitely was definitely a very, very nice example. I think we, we had just developed this method at

00;26;46;06 - 00;26;47;25
Unknown
at Kansas and it was a very nice,

00;26;47;25 - 00;26;51;04
Unknown
environment to, to test this new approach was really fun.

00;26;51;07 - 00;26;52;02
Unknown
Yeah.

00;26;52;02 - 00;26;57;29
Unknown
Yeah. I love the fact I like how you can visualize all these trajectories sort of going through this, this little volume, this tube,

00;26;57;29 - 00;27;06;07
Unknown
the, the noodle. And I'll probably have to give Irv Albrecht credit for that because I remember him from Toronto. I remember him saying I it might have been I think he said it was a noodle.

00;27;06;09 - 00;27;12;12
Unknown
And, I didn't think about it like that, but, hey, I give him credit for it. Yeah. Yeah.

00;27;12;12 - 00;27;13;18
Unknown
But so this is,

00;27;13;18 - 00;27;17;13
Unknown
if you had to define this. I mean, it's not like, area of origin. So if we like,

00;27;17;13 - 00;27;24;02
Unknown
maybe, maybe let's do that. Let's describe. Or if you can describe for me what does the area of origin tell us?

00;27;24;05 - 00;27;43;26
Unknown
So if I flip back to this project, you know, you got this egg looking thing here, right, versus what this noodle is telling us here, right? This tube and all these lines that are going through. So maybe we'll start here with the air of origin, you know, how is this being defined or how is this classified. Yeah. So this one,

00;27;43;26 - 00;27;49;23
Unknown
when we're doing area foraging analysis, we're assuming that the bloodstains are coming from a common area of origin.

00;27;49;26 - 00;27;54;03
Unknown
We're not talking about a point, but rather an area, because it's always a dynamic event. And,

00;27;54;03 - 00;27;57;01
Unknown
so this could be the result of a beating, for example.

00;27;57;01 - 00;27;59;14
Unknown
Obviously, this was created by using an impact,

00;27;59;14 - 00;28;07;00
Unknown
device that we made. So there's an impact into liquid blood, and the blood is then dispersed through the air to surrounding objects,

00;28;07;00 - 00;28;08;04
Unknown
walls.

00;28;08;06 - 00;28;10;20
Unknown
And we can calculate,

00;28;10;20 - 00;28;14;20
Unknown
backwards from these trajectories. So the trajectories are also in red here.

00;28;14;20 - 00;28;18;09
Unknown
And the area of origin indicates some kind of statistical,

00;28;18;09 - 00;28;25;17
Unknown
statistical region wherein the area of origin should be. Okay. And let's switch. Let's go to here. Let's go to the cast,

00;28;25;17 - 00;28;28;16
Unknown
cast off example here. So how how is this different now?

00;28;28;16 - 00;28;39;05
Unknown
How is it how was it being calculated? Yeah. So this is different in the sense that the object that's releasing the blood is actually in motion. So I think you used I don't know, was it a knife or a hammer or something or. No. Just

00;28;39;05 - 00;28;43;16
Unknown
like a plastic stick, wooden or something. Yeah. It was some kind of a rod or something like that.

00;28;43;16 - 00;28;53;28
Unknown
I can't even remember it now. Yeah, it was drenched in blood. And you started swinging it out a wall and, I've got videos of it if you want to. Yeah, yeah, I think we do. Somewhere.

00;28;53;28 - 00;28;56;02
Unknown
We've got some different things beyond,

00;28;56;02 - 00;29;01;27
Unknown
as you swing the object, the blood is actually released in a somewhat circular motion or radial motion.

00;29;01;29 - 00;29;10;28
Unknown
And in this case, we're not calculating back to a single area of origin, but more towards the whole trajectory of where the the blood has been released from.

00;29;10;28 - 00;29;14;22
Unknown
We're doing some, some nice validation or you doing some nice validation on it.

00;29;14;22 - 00;29;21;05
Unknown
I think it's, it's looking promising. It is. It is. In fact, I don't know why I have this,

00;29;21;05 - 00;29;33;02
Unknown
interesting cast off, but yeah, it's really interesting what we can do, and I think I think it's, you know, the fact that it's just another way of defining, you know, another this other mechanism, you know, and, and saying something about it.

00;29;33;05 - 00;29;36;19
Unknown
Yeah. If you, if you go to the right on the property view, you can turn on the,

00;29;36;19 - 00;29;42;00
Unknown
the plane. I think there's should be a checkbox to enable the plane.

00;29;42;00 - 00;29;43;05
Unknown
I think I might, I think I

00;29;43;05 - 00;29;43;19
Unknown
have a,

00;29;43;19 - 00;29;52;12
Unknown
I got a, like a special version. So it's like, okay, no worries. So typically you could also visualize the plane, which is the 2D plane in which the object was swung.

00;29;52;12 - 00;29;54;14
Unknown
So somewhere within this plane,

00;29;54;14 - 00;29;56;25
Unknown
an arm was swinging or an object was swing.

00;29;56;25 - 00;30;04;15
Unknown
So there's different kinds of ways to visualize this. And I think the tube is one of it. And again, the width of the tube just represents some kind of,

00;30;04;15 - 00;30;11;22
Unknown
uncertainty similar to the area of origin. Okay. You have this image. Can you explain what the image is doing here.

00;30;11;25 - 00;30;12;12
Unknown
Yeah. So,

00;30;12;12 - 00;30;14;07
Unknown
the augmented reality we have,

00;30;14;07 - 00;30;19;26
Unknown
is one of the things we focus on to make, evidence more easy to present in court.

00;30;19;26 - 00;30;28;13
Unknown
So if you have a 3D scanner and you scan your crime scene in 3D, you can import this into HemoVision and show your results in the full context of the crime scene.

00;30;28;13 - 00;30;54;19
Unknown
And this is very good for presentation purposes, but not everybody has a 3D scanner. So we thought we still want to be able to include like the context, the full context of the crime scene. So let's develop some kind of augmented reality. And what it does basically is we have just a plain standard photograph, and we're going to estimate the position of the camera with respect to our scene that we have, and we're going to overlay our results on top of this photograph.

00;30;54;26 - 00;31;04;23
Unknown
And this is basically as if you would have used strings in real life. As this allows you to get some kind of context, crime scene context, or most of it,

00;31;04;23 - 00;31;13;24
Unknown
but without having a 3D scanner. Yeah, yeah. No, I think it's looks super cool because you're using the original photograph, so it looks realistic, you know, I mean, and you get these things coming out so really, really nice.

00;31;14;01 - 00;31;22;12
Unknown
Good job there. Are there other things you could do with this? Like, somehow, could you extend this to something that's, like, animated or moving or. I'm not sure. Yeah. So this is,

00;31;22;12 - 00;31;26;10
Unknown
this is on a on a still image. And basically we could,

00;31;26;10 - 00;31;31;29
Unknown
extend this method to video. We have done some experiments with it, so it's always more for augmented reality.

00;31;31;29 - 00;31;33;19
Unknown
It still remains a still image.

00;31;33;19 - 00;31;40;01
Unknown
It gives you a good idea, but it's still a still image. And it's difficult or maybe difficult to interpret 3D. So,

00;31;40;01 - 00;31;53;12
Unknown
if you're taking a video of your crime scene, you could basically do this for every frame of your video, and then you would actually you would be able to move around, for example, the body and see all of the see all of the bloodstains in the areas of origin.

00;31;53;14 - 00;31;56;03
Unknown
We do have a little bit of work there to make it,

00;31;56;03 - 00;31;58;07
Unknown
process in,

00;31;58;07 - 00;31;59;13
Unknown
acceptable time frame.

00;31;59;13 - 00;32;04;00
Unknown
You don't want to or we don't want it to process for half an hour. We want it to be quicker.

00;32;04;00 - 00;32;08;07
Unknown
But we're definitely looking into that as well. Yeah. So you can just do this with videos as well.

00;32;08;09 - 00;32;09;26
Unknown
Amazing. Okay.

00;32;09;26 - 00;32;20;02
Unknown
So tell me about the use of human vision right now. Have there been cases that you know of maybe like where human vision has been used, or do you know, different types of cases where it's been used? Yeah. So,

00;32;20;02 - 00;32;22;05
Unknown
I think one of our longest, our,

00;32;22;05 - 00;32;25;01
Unknown
longest lasting partnership is with the federal police in Brussels.

00;32;25;01 - 00;32;28;13
Unknown
So they've been using it for two and a half years now, I believe.

00;32;28;13 - 00;32;28;25
Unknown
They've,

00;32;28;25 - 00;32;29;20
Unknown
done,

00;32;29;20 - 00;32;34;16
Unknown
a lot of cases with it. And I think most recently 1 or 2 of them also went to, to trial.

00;32;34;16 - 00;32;39;23
Unknown
I'm not sure exactly when or how, but, yeah. So they are using it operationally.

00;32;39;23 - 00;32;42;17
Unknown
There's obviously other countries as well.

00;32;42;19 - 00;32;43;22
Unknown
Yeah.

00;32;43;22 - 00;32;44;07
Unknown
So

00;32;44;07 - 00;32;46;19
Unknown
let me ask you, like, just, you know, based on where you are today,

00;32;46;19 - 00;32;48;18
Unknown
like, where do you see,

00;32;48;18 - 00;32;56;17
Unknown
bloodstain pattern analysis using software like HemoVision, like moving forward to or like, are they going to be integrating other types of things like,

00;32;56;17 - 00;33;04;19
Unknown
what kind of features or, you know, evolving technology might be implemented in human vision in the future?

00;33;04;22 - 00;33;05;16
Unknown
Yeah. So there's

00;33;05;16 - 00;33;07;26
Unknown
a lot of different things on our list.

00;33;07;26 - 00;33;09;18
Unknown
We have, we've got a lot of ideas.

00;33;09;18 - 00;33;15;28
Unknown
I think one of the main things is growing HemoVision by, by actively listening to our, to our current customers,

00;33;15;28 - 00;33;19;00
Unknown
we get a lot of feedback from them. And that's really nice.

00;33;19;00 - 00;33;24;03
Unknown
And cast off as one of the things that was asked for us to hey, could you research that a little bit?

00;33;24;03 - 00;33;26;25
Unknown
And so we're doing that now.

00;33;26;25 - 00;33;31;05
Unknown
But also the augmented reality, video is something we're working on.

00;33;31;05 - 00;33;43;16
Unknown
We're also getting some questions about non-linear trajectories. So for those who don't know. So it's very fairly common to use linear trajectories to represent the flight path of bloodstains.

00;33;43;16 - 00;33;50;27
Unknown
But in reality the bloodstains don't travel in a linear way to travel in a non-linear way due to drag and gravity and that kind of stuff.

00;33;50;27 - 00;33;54;18
Unknown
But we typically ignore it under certain rules. You,

00;33;54;18 - 00;33;58;16
Unknown
apply it to select blood stains to limit the effects of drag and gravity.

00;33;58;16 - 00;34;07;07
Unknown
Nonetheless, in some cases, it could be interesting to have a method that incorporates gravity and drag and we're looking in into into that as well.

00;34;07;07 - 00;34;09;15
Unknown
We've done some early research on it.

00;34;09;15 - 00;34;10;17
Unknown
Other stuff involved.

00;34;10;17 - 00;34;12;07
Unknown
So we've recently,

00;34;12;07 - 00;34;13;13
Unknown
upgraded our,

00;34;13;13 - 00;34;19;20
Unknown
software to include all kinds of 3D scans and mesh models so you can all import these into HemoVision.

00;34;19;20 - 00;34;25;18
Unknown
Taking that a step further, maybe we can export it again into virtual or augmented reality glasses.

00;34;25;18 - 00;34;39;26
Unknown
Yeah. So a lot of ideas for innovation. Let me ask you about the non-linear trajectory, because I think that's an interesting area, because it's obviously trying to mimic the flight path of a bloodstain more realistically, more physically, realistically, because the

00;34;39;26 - 00;34;43;26
Unknown
yeah, the linear trajectory makes some pretty broad assumptions.

00;34;43;28 - 00;34;49;20
Unknown
So, normally if you're doing, you know, if you want to,

00;34;49;20 - 00;34;53;07
Unknown
calculate something going through the air, you have to know a lot about,

00;34;53;07 - 00;34;57;29
Unknown
mass and you have to know a lot about the conditions like that sort of thing. So,

00;34;57;29 - 00;35;05;13
Unknown
have you done any experimentation with that already? Like, if you, if you coded anything, maybe on the side to like, run through some scenarios and,

00;35;05;13 - 00;35;11;06
Unknown
how the how much can you get away with if you don't know a lot like, is it possible to, you know,

00;35;11;06 - 00;35;18;02
Unknown
document the scene conditions like people are doing right now with human vision and then feed that in and maybe use that as a starting point.

00;35;18;04 - 00;35;18;25
Unknown
Yeah. Well

00;35;18;25 - 00;35;19;27
Unknown
if we're talking about

00;35;19;27 - 00;35;24;10
Unknown
non-linear trajectories maybe we should shout out Daniel Ettinger. I think he's the main expert in

00;35;24;10 - 00;35;25;08
Unknown
in this field.

00;35;25;08 - 00;35;27;27
Unknown
So he has a method that incorporates quite a lot of,

00;35;27;27 - 00;35;30;17
Unknown
physics or a lot of physics in his methods.

00;35;30;17 - 00;35;33;07
Unknown
So I think he knows pretty much,

00;35;33;07 - 00;35;34;23
Unknown
all of it.

00;35;34;26 - 00;35;43;20
Unknown
We have done some experimenting and coding ourselves, but obviously there's certain things that are difficult to get from a crime scene. I mean,

00;35;43;20 - 00;35;53;19
Unknown
for example, the volume of blood, it's very difficult to determine based on just the appearance of a blood stain, because what's the how much does the the surface that the blood stain is deposited on absorb the blood?

00;35;53;19 - 00;36;02;06
Unknown
Is there any shrinkage, all that kind of stuff. So we did some experiments where we make certain assumptions.

00;36;02;06 - 00;36;03;17
Unknown
And we also,

00;36;03;17 - 00;36;13;28
Unknown
leave certain variables open. And we ask the software to optimize the whole system of constraints that we have. We have like 20 or 30 blood stains.

00;36;13;28 - 00;36;17;03
Unknown
We have, for example, an area of origin coming from linear trajectories.

00;36;17;03 - 00;36;23;15
Unknown
We can put this all into a system and we ask, okay, given everything that we can give you, what is the most plausible,

00;36;23;15 - 00;36;25;25
Unknown
solution for non-linear trajectories?

00;36;25;25 - 00;36;27;11
Unknown
And this is again more,

00;36;27;11 - 00;36;31;29
Unknown
numerical optimization and looking for the best possible parameters given a certain scenario.

00;36;31;29 - 00;36;39;03
Unknown
We have done, I think I did a talk at the I London IPA, like the, the virtual one we had during Covid.

00;36;39;05 - 00;36;45;13
Unknown
But I haven't picked up the material since then, but maybe I should pick it up again soon. I think you should, because,

00;36;45;13 - 00;36;49;14
Unknown
it's that. That would be super cool if we can start getting some, you know, some of that,

00;36;49;14 - 00;37;01;20
Unknown
implemented without a lot of effort. I mean, if the workflow would seem where, you know, the user, you know, the person is just documenting in the same way the photographs putting in a measurement and then, you know, making some assumption assumptions.

00;37;01;20 - 00;37;19;03
Unknown
I think that makes a lot of sense, at least. At least it would be closer, I think, to what's happening in reality. As opposed to these linear trajectories. Linear trajectories are useful, but they're also quite limited. Right? Yeah. If you're if your blood stains are still traveling upwards when they hit the wall,

00;37;19;03 - 00;37;22;14
Unknown
the overestimation of the height will be somewhat limited.

00;37;22;16 - 00;37;39;25
Unknown
But in some cases you only have blood stains on the floor, for example, that are elliptical, but they are on the floor. So it's very difficult to get an accurate estimate with linear trajectories in those cases. I think nonlinear ones would be very interesting to have. Okay, so we've talked about a lot of different things obviously, but I like what is the next thing for you.

00;37;39;25 - 00;37;52;25
Unknown
What are what are you primarily focusing on right now for HemoVision. What's the the next sort of thing you're going to implement. Or maybe there's more than one I don't know. Yeah. So all of that I think many of the things we, we discussed already, I think one of the,

00;37;52;25 - 00;37;54;27
Unknown
one of the things is now growing HemoVision,

00;37;54;27 - 00;37;58;05
Unknown
we're getting a lot of positive feedback from our customers, and we try to,

00;37;58;05 - 00;37;59;06
Unknown
also look at the small things.

00;37;59;06 - 00;38;12;05
Unknown
So we're looking at big things like cast off and non-linear trajectory and pattern classification. But there's also like small tweaks and improvements we can make to the software that our customers like. So that's something that we focus on.

00;38;12;05 - 00;38;20;07
Unknown
Yeah. And for HemoVision, obviously we've got tons of ideas, but we're also focusing on different areas of forensics.

00;38;20;09 - 00;38;22;22
Unknown
So as a company forensics, we try to,

00;38;22;22 - 00;38;25;27
Unknown
copy the vision that we had or have for HemoVision to other,

00;38;25;27 - 00;38;26;26
Unknown
fields.

00;38;26;26 - 00;38;38;23
Unknown
Vision 3D, as you mentioned, is, is is one of those. So we're looking at different areas as well. Yeah. Hey, I got a question from Jeff here. He says, you could ultimately give information of length of arm object.

00;38;38;23 - 00;38;43;00
Unknown
And this is going back to when we were talking about the cast off. So. Yeah. Yeah. So

00;38;43;00 - 00;38;49;13
Unknown
and actually, you can, you can say what, what you're doing in the software, maybe. Well,

00;38;49;13 - 00;38;50;21
Unknown
yeah. So,

00;38;50;21 - 00;38;51;25
Unknown
Hi, Jeff. I think you're,

00;38;51;25 - 00;39;00;13
Unknown
I think you're in the right direction. I think we could get some kind of estimate of arm length plus object.

00;39;00;15 - 00;39;04;08
Unknown
And if you didn't know which object object is used, you maybe could,

00;39;04;08 - 00;39;05;16
Unknown
find out the arm length.

00;39;05;16 - 00;39;12;27
Unknown
So what we do right now for to cast off is, as a result from how we I'm not going to go too deep into the method of how we actually,

00;39;12;27 - 00;39;13;13
Unknown
determined,

00;39;13;13 - 00;39;15;00
Unknown
the tube or the noodle.

00;39;15;00 - 00;39;20;25
Unknown
But what what we ultimately get from the analysis is like a center point and a radius.

00;39;20;28 - 00;39;28;07
Unknown
And so this would be the pivot point and the length of the arm plus object. Yes. Yeah. And,

00;39;28;07 - 00;39;29;23
Unknown
it is something that you currently,

00;39;29;23 - 00;39;35;01
Unknown
report right in the, in the software, but it's still under beta and I should give a shout out to, to,

00;39;35;01 - 00;39;39;03
Unknown
is the student from the University of Toronto. It's a severe.

00;39;39;05 - 00;39;44;10
Unknown
Yeah. Brickman. Maxwell. So she's been doing some work in this particular area quite a bit, actually. And,

00;39;44;10 - 00;39;45;08
Unknown
we're currently working on,

00;39;45;08 - 00;39;47;14
Unknown
a study, hopefully going to wrap that up soon and,

00;39;47;14 - 00;39;56;16
Unknown
get that published as well, show people what we've got. So hopefully that that will come through at some point. Sorry. I'm looking forward to the results.

00;39;56;22 - 00;39;57;27
Unknown
Yeah, yeah. Me too.

00;39;57;27 - 00;40;01;27
Unknown
We've got some blind testing that she's working on right now, so I went ahead and did some,

00;40;01;27 - 00;40;03;02
Unknown
some blind,

00;40;03;02 - 00;40;05;13
Unknown
castoff patterns. And now she's tasked with,

00;40;05;13 - 00;40;08;07
Unknown
trying to interpret those and analyze them and see what she gets.

00;40;08;07 - 00;40;11;05
Unknown
She doesn't have the advantage of knowing where it came from. Now. So,

00;40;11;05 - 00;40;13;01
Unknown
that's the way it works.

00;40;13;04 - 00;40;19;10
Unknown
I mean, you mentioned about conferences and such, so you guys get out every now and then to conferences. You're doing some training, like,

00;40;19;10 - 00;40;25;11
Unknown
you know what? What do you got lined up? You're doing anything online or are you planning anything to do online? Yeah. So conferences,

00;40;25;11 - 00;40;26;10
Unknown
definitely one of the,

00;40;26;10 - 00;40;28;27
Unknown
the IPAs, a yearly must for us, I think,

00;40;28;27 - 00;40;31;15
Unknown
last 4 or 5, six years, I've been,

00;40;31;15 - 00;40;31;29
Unknown
it's always,

00;40;31;29 - 00;40;33;19
Unknown
an amazing conference with amazing people.

00;40;33;19 - 00;40;37;14
Unknown
And I love coming back, so we'll definitely be there.

00;40;37;14 - 00;40;39;26
Unknown
We're doing other conferences as well, like NV,

00;40;39;26 - 00;40;40;16
Unknown
the European,

00;40;40;16 - 00;40;41;22
Unknown
conferences.

00;40;41;22 - 00;40;42;25
Unknown
We try to visit,

00;40;42;25 - 00;40;45;13
Unknown
some customers here and there if it's close by.

00;40;45;13 - 00;40;51;09
Unknown
So any any nice opportunity for us to interact with experts in the field for us, it's always nice to learn from.

00;40;51;09 - 00;40;52;26
Unknown
From people. Yeah.

00;40;52;26 - 00;40;54;27
Unknown
Online. Yeah, we do.

00;40;54;27 - 00;40;56;19
Unknown
On LinkedIn, we try to be active,

00;40;56;19 - 00;40;58;16
Unknown
and maybe in the future we'll do like,

00;40;58;16 - 00;41;07;24
Unknown
a HemoVision webinar, where people can learn more about HemoVision and provide us with questions and feedback and, yeah, let me,

00;41;07;24 - 00;41;11;21
Unknown
maybe this up here. So I just want to get this, but there is a there's a contact form on the,

00;41;11;21 - 00;41;12;11
Unknown
yes.

00;41;12;13 - 00;41;15;04
Unknown
On the human vision that the eager contact and,

00;41;15;04 - 00;41;28;24
Unknown
if you go over, just, they can always get Ahold of you here, right? If anyone wants to talk to you, right. Yeah. Okay. Perfect. Okay. Yeah. Send us an email. All right. Excellent. All right, so I think we covered a bunch of stuff here, and,

00;41;28;24 - 00;41;35;16
Unknown
I know that, you're going to be a busy guy in the near future because you've got some, some life changes happening.

00;41;35;19 - 00;41;36;09
Unknown
Yeah. So,

00;41;36;09 - 00;41;40;03
Unknown
that's true. Well, me and my wife are expecting our first,

00;41;40;03 - 00;41;42;16
Unknown
son, so it's our first child. So it's it's,

00;41;42;16 - 00;41;44;08
Unknown
exciting times. So,

00;41;44;08 - 00;41;46;17
Unknown
this year is actually going to be born at,

00;41;46;17 - 00;41;49;29
Unknown
during the IBP this year. So it'll be my,

00;41;49;29 - 00;41;52;10
Unknown
my colleague and co-founder Ruben, who will be,

00;41;52;10 - 00;41;53;17
Unknown
doing the honors of,

00;41;53;17 - 00;41;54;27
Unknown
of attending the IBP.

00;41;54;27 - 00;41;57;23
Unknown
So it's going to be a first for him, but it's going to be nice for him,

00;41;57;23 - 00;42;06;19
Unknown
to to meet all the lovely people I've been interacting with for years. Yeah. Ruben is kind of a quiet guy sometimes. I don't know how to take him, but he's. He's super smart. He's super nice.

00;42;06;19 - 00;42;07;16
Unknown
But he's,

00;42;07;16 - 00;42;08;12
Unknown
he's he's been with you.

00;42;08;17 - 00;42;11;19
Unknown
He's a co-founder, right? He's been with you for the beginning. Yes.

00;42;11;19 - 00;42;11;29
Unknown
He,

00;42;11;29 - 00;42;14;23
Unknown
he actually joined the project somewhere 2000.

00;42;14;23 - 00;42;17;01
Unknown
Let me think. Maybe 20, 21 somewhere.

00;42;17;01 - 00;42;17;06
Unknown
So,

00;42;17;06 - 00;42;31;22
Unknown
it's when we decided that we were going to build a company from the university, but we had to do some more research, transforming everything from the research phase to an actual intuitive, intuitive interface that is, yeah, easy to, to be used.

00;42;31;24 - 00;42;33;26
Unknown
And he joined the project then, and,

00;42;33;26 - 00;42;44;29
Unknown
he's been on it ever since, and he's a co-founder now. So, so he's from he was also from K11 and yes. Okay. Yeah. It was, it was the same program or a different program or is it. What was he studying? Something

00;42;44;29 - 00;42;47;00
Unknown
different program. We had some similar studies.

00;42;47;00 - 00;42;48;11
Unknown
Also the master of AI.

00;42;48;11 - 00;42;50;08
Unknown
But then he went to work on for a different,

00;42;50;08 - 00;42;56;02
Unknown
startup company. But I hired him to do on the valorization track again. Okay, cool. Ever.

00;42;56;02 - 00;42;59;19
Unknown
And things just grew from there, you know? Okay, well, he's a great guy, and,

00;42;59;19 - 00;43;06;04
Unknown
I might have to get him in here one day. Maybe next time I'm going to talk to him about some of the the heavy technical stuff he's been working on in the background.

00;43;06;04 - 00;43;10;13
Unknown
So. Yes. All right. Well, look, thanks so much for I appreciate it.

00;43;10;13 - 00;43;15;26
Unknown
We'll be in touch. We've got a lot to talk about, a lot of different projects we're working on together. But I really appreciate you being here. Thank you very much.

00;43;15;26 - 00;43;17;06
Unknown
Thank you very much, Eugene.

00;43;17;06 - 00;43;18;16
Unknown
Okay, folks. That's it.

00;43;18;16 - 00;43;23;22
Unknown
We are done with this particular episode talking about. But seeing pattern analysis, HemoVision.

00;43;23;22 - 00;43;26;16
Unknown
You know, I BPA that's going to be coming up. And so,

00;43;26;16 - 00;43;29;22
Unknown
if you have any questions for Phil, just go ahead and reach out.

00;43;29;22 - 00;43;33;21
Unknown
We will be back at some point in the near future with some more talks.

00;43;33;28 - 00;43;46;15
Unknown
Got some things on body cameras. I've got some things on forensic anthropology, a whole number of different areas. So. Hey, thanks, everyone. I really appreciate your time today. Have a great Thursday. We'll catch up with you soon. Bye bye.