Hypothesis A: Computer vision can recognize patterns of facial motion and eye movement that match contemporaneous mental states and also predict mental disorders, through traditional statistical analysis as well as a convolutional neural network.
Hypothesis B: Viewing a person's face with a hyperspectral camera during a clinical interview will produce identifiable patterns of heartbeat and blood flow that correlate with mental states identified by a computer vision system.
|Concept||✅||Concept plan completed and posted.|
|Literature review||✅||Literature review completed.|
|Case||✅||Computer installed in rackmount case.|
|A100 subsystem||❌||A100 GPU subsystem designed, still have to purchase.|
|Aerospace connectors||❌||Quote received, still have to purchase.|
|Cables and wiring||❌||Lining up a builder of mil-spec cables and wiring harnesses.|
|Industrial cameras||✅||Industrial cameras delivered and test images made.|
|FLIR camera||❌||Waiting to place order for FLIR science kit.|
|Software engineering||❌||Software engineer tentatively selected.|
|Algorithms||❌||Algorithms for pattern recognition in development in Mathematica 13.1.|
|Research||❌||Inquiries starting to be circulated for validating hypothesis.|
|Publication||❌||Inquiries of interest in this research made to journals.|
|Product ready||❌||Product must be validated by research, which is expected in mid-2023.|
October 1, 2022: Product web page launched!
September 25, 2022: Playing around in Mathematica, I have succeeded in generating a 3D model from two camera images using entirely built-in functions. This is proof that the theory is workable even in a prototyping environment, well before live video is possible.
September 25, 2022: I am delighted with my industrial cameras from The Imaging Source, which produce great test video [HEVC] or [H.264] at 60 Hz. I built a custom stereo camera rig using parts from Samy's Camera pending delivery of my actual stereo rig. The build quality of these adorable cameras is exceptional at this price.
September 17, 2022: I re-purposed a pair of ultra high quality aircraft vents to be air inlets in my case. These things are beastly, and basically the highest quality "eyeball" vents ever made. You push in and turn to open/close, and you can reposition them to any angle. Perfect to go in the top lid here, which I will do today. It so happens that the M5 rack screws and cage nuts fit perfectly, which works for me.
September 5, 2022: I got a no-contact temperature gauge at an auto parts store just to make sure everything is within Supermicro thermal specs. The exhaust from the case fans is about 110 degrees, which is close to what it was putting out in a tower configuration in an empty room. So I'm fine with this...next step is putting the IP67 fans in the case along with vents in the top lid, and seeing how that goes. The steel chassis is going to be a great heat sink; the only question is whether it's sustainable to run this closed when the ambient temperature is around 75 degrees. It's intended to be used indoors in a laboratory environment, but it would be nice to have some flexibility. There are "wide temp" memory modules and other chips available, but the issue is the CPU and GPU. Under full load, the fans almost never go above normal speed. This is an air-cooled system, and I'm not sure how much flexibility I have with that because a water-cooled setup would be less durable. Work continues!
September 3, 2022: I finally downloaded my Mathematica 13.1 purchase. This is the greatest technical computation program of all time, which I have been using at an introductory level since 2013. Can't wait to get the pattern recognition algorithms developed in this, which will be game-changing. Separately, I found a C++ software engineer for the desktop application and we'll get a demo going as soon as possible showing a simple wire frame/3D model from the stereoscopic camera inputs.
September 3, 2022: I got the workstation upstairs at home for now, and I also picked up the product labels. They are laser-engraved brass and turned out amazing. Although I injured myself carrying this upstairs (it's about 140 pounds), it was worth it.
With the noise-deadening acoustic foam, this is very tolerable in a home office environment.
August 28, 2022: It took several hours this weekend to install my workstation in the Pelican Super-V rackmount case. No joke, that was the most difficult rack install ever due to the awkwardness of moving a heavy computer around on a garage floor, and otherwise. The computer is in a 4U heavy duty Supermicro case, which I'm happy to have, but this is a beast of around 130 pounds with the rolling case, and that's before even adding the A100 GPU "backpack." I'm going to have to transport this only via air cargo with special handling and "do not drop" stickers on it (or driving it-- it actually fits perfectly in the back seat of my car). The Pelican Super-V is a great product; in retrospect, I'm not sure much if any weight would be saved by going with carbon fiber as in my RFP below. Shout-out to Pam and Mike with Engineered Packaging Services, which got me this fantastic case as a special order from Pelican. I specified square rack holes and a 30" depth, and it came out perfectly. Here it is before the computer went in:
Check out those lid hangers on the case-- super cool. My intention is to have the aerospace connectors on the left side and leave the bottom lid on, with IP67 fans in it. Although I will lose waterproofing, it's still going to be water-resistant and I don't intend to use it in standing water anyway (lol). The case also has shock absorbers that attach the rack to the shell, which is a very cool feature. All I need is drop resistance from a very small height, just in case I can't manage to smoothly extract it from my car, haha.
With the computer installed.
Still have room for a 1U device in the top.
Look at those fans! Insane.
There you have it, a computer that costs as much as a car.
August 13, 2022: Request for Proposal -- custom C++ software engineering with the OpenCV library:
Hi, I am a part-time inventor and systems integrator with an interesting project that intersects computer vision and psychology. The concept plan provides background information on the project: https://www.raellic.com/files/Raellic.pdf I'm beginner-level with Python implementations of OpenCV and I might get pretty far in Python, but that's not what I'm after here-- I need maximum performance. The goal is real-time 3D point cloud/mesh generation and measurement from stereoscopic medium-high resolution (2,048 x 1,536) industrial cameras at 60 Hz, with a synchronized overlay of low resolution (640x512) radiometric thermal video at 60 Hz, with all the data recorded. For real-time performance at 60 frames per second, even with a C++ application I estimate I will need an extreme performance workstation (already built), with an add-on Nvidia A100 GPU "backpack" running Linux so it can run the Riva A.I. suite (not built yet). This thing is going to be in a rolling rackmount case with aerospace-grade connectors that weighs around 100 pounds. Let me worry about that part. Your job is to develop a desktop application in C++ that is the fastest possible implementation of the appropriate OpenCV libraries needed to accomplish the goal: I want to see a 3D model on screen with various measurements and telemetry appearing in real-time while the subject is talking, and the ability to non-destructively add comments and annotations with timestamps-- with all the data recorded in an open format for later use and further analysis. The A100 subsystem will process the audio using Riva's automatic speech recognition and run macros on the resulting transcript stream. I'm currently planning the A100 subsystem as a 1U rackmount module that goes in the rolling case. However, I am also considering putting the A100 in my data center cabinet in case I want to downsize the package into a smaller form factor and rely on having internet connectivity for the speech processing. Not sure which option I will go with yet. The product will be sold commercially if validated by the research discussed in the concept. I am not doing open source due to the magnitude of the investment required, as well as the fact that I have come up with my own inventions, algorithms, and valuable I.P. here-- at least in the psychology portion of the project. This is expected to be journal-level work in the area of correlating facial motion with mental processes, and it's going to be submitted for publication when it's completed. The product is very likely to become export-controlled and/or subject to ITAR due to the inclusion of export-controlled components from Teledyne FLIR. I think there are lots of overseas candidates who would be interested, but because of export controls and ITAR, I believe I currently have to limit this to U.S. candidates only. Sorry! Send c.v., highlight any relevant experience, and indicate an hourly rate of at least $50/hr. Opportunity is remote, but on the off chance any candidates are in the San Francisco Bay Area, I do have an actual office available for up to full-time use during the project. Thank you for your consideration. Edit: in case it wasn't obvious, GPU acceleration is a key aspect of this, so plan on using the CUDA library. I have a RTX 3090 in the workstation, so maybe that is fast enough without the A100...not really sure.
August 10, 2022: Request for Proposal -- carbon fiber 5U rackmount case at 30" depth, in the style of the Pelican Super-V 5U rackmount case. The intention is to save 15 pounds in order to keep the case under 100 pounds. In all likelihood, the high value of the system requires air freight-type special handling, however, the option of short-notice air travel in checked luggage would be highly desirable.
Inquire via email to my R&D address.
Dynamic Facial Expression of Emotion and Observer Inference
Eye Movement in Unipolar and Bipolar Depression: A Systematic Review of the Literature
Eye movement characteristics in schizophrenia: A recent update with clinical implications