USC professor is crowdfunding his open-source machine vision camera for Arduino, Raspberry Pi, and PC projects.
Laurent Itti’s goal in life is to make machines that can see. Itti was frustrated with STEM education and courses that taught programming and robotics well but neglected machine vision. After he developed a theory that machine vision wasn’t taught enough because no simple machine vision devices existed, he decided to create his own. Itti is currently running a Kickstarter for JeVois, an open-source machine vision camera for maker projects.
JeVois consists of a video sensor, a quad core CPU, a USB video port, and a serial port in a 28 cubic centimeter package. Three operational phases are currently programmed into JeVois. A development mode shows a demo display and communicates results. Text-only mode gives text strings from the smart camera without the USB output. The Pre-Processing mode sends video for machines to use – edge maps or image crops around interesting objects in frame. In the campaign video Itti says that he wanted the system to have high frames per second, sacrificing picture resolution to have great applications in machine vision.
The quad-core ARM-V7 processor runs at 1.34 GigaHertz, with 256 MB DDR3-1600 RAM and a dual-core Mali-400 GPU. The 1.3 megapixel camera can run at speeds up to 120 frames per second. The campaign page discusses Linux (including Raspberry Pi), Windows, Mac OS X, and Android video streaming applications – Linux is working out of the box, the others require some tinkering to work properly. The open-source aspect of the project is very interesting here because the current limitations are laid out while fully expecting users to pick this up and run with it on several platforms where it doesn’t currently exist.
JeVois is an ambitious project, intended to be the first part of an outreach program that brings machine vision to the education and maker worlds. The campaign page imagines several applications, including visual attention, recognizing different places, reading digits, detecting QR codes, finding moving things in a frame, and tracking objects by color. This is a part of what Itti envisions as curriculum that can work from kindergarten up to PhD levels of education.
There’s obviously been a lot of work done on this system, and a lot of thought put into its applications over the last decade. If anything, this is the first time I’ve felt there was too much technical information on a Kickstarter page. The campaign ends on January 23 and if funded units will ship in February and March 2017.