Defense Department Taps U of A to Develop Video Tracking Tool

The University of Arkansas joins a nationwide team developing next-generation video object tracking technology. The work is funded by a highly competitively contract from the Defense Advanced Research Projects Agency. If the team successfully completes all phases of the project, it could bring $6 million in federal research funding to the U of A. 

Jeff Dix, an assistant professor in the Department of Electrical Engineering and Computer Science, leads the work at the U of A. 

Dix will focus on designing small, power efficient analog circuits that can analyze a video signal. 

By avoiding the conversion of the analog video signal to digital, the circuit will consume less power. DARPA's goal is an analog neural network that would be 2,000 times more power efficient than current technology. 

"They want it to be under a watt of power. It's crazy," Dix said. 

The proposed circuit must also be dramatically smaller and lighter than current solutions. 

"You could do this neural network now with a server, but it's physically large," he said. 

Dix assumes the goal is a mobile video object tracking tool. While a DARPA contract challenges researchers to meet specific benchmarks, it often does not reveal the ultimate application of the technology. 

The proposed final circuit will fit on a silicon die that measures 100 millimeters squared, roughly the size of a small button on a dress shirt. 

"In terms of our world, it's huge," Dix said. "That's not something that universities typically do." 

The integrated circuits typically created at the U of A are more often 1 millimeter squared. Dix will work with Green Mountain Semiconductor, a Vermont-based semiconductor design firm, and another team member, to design and fabricate this large, complex circuit for the DARPA project. 

The other team members are SRI, an independent, non-profit institution that began at Stanford University, and AI Sensation, which develops artificial intelligence solutions. 

SRI and AI Sensation will create AI algorithms both to process the pixels in an image and to compensate for noise and variance in Dix's circuit. 

The project is divided into three phases, and the researchers must meet concrete benchmarks to advance onto the next phase. DARPA sets ambitious goals for its projects. Most teams do not advance through all the stages, but even approaching the goals can lead to major advances. 

"It could be something unobtainable. This is high risk, high reward," Dix said.

NOTE: This material is based upon work supported by the Army Contracting Command–APG under Contract No. W912CG-25-C-A004. Any opinions, findings and conclusions or recommendations expressed in this material are those of the author(s) and do not necessarily reflect the views of the Army Contracting Command–APG.