Skip to main content

Allen School’s Aditya Kusupati earns Best Paper Runner-Up at BuildSys 2019 for new low-power, deep learning algorithm for radar classification

A team of researchers that includes Allen School Ph.D. student Aditya Kusupati has developed a new low-power real-time solution for mote-scale (tiny sensor with a weak microprocessor) radar-based intruder detection. Their work has enabled the first end-to-end deep learning solution for radar classification and won the “Best Paper Runner-Up Award” at the Association for Computing Machiner’s 6th International Conference on Systems for Energy-Efficient Buildings, Cities, and Transportation (BuildSys 2019) last month.

With the rapid growth of the Internet of Things sensors, there is an increased need for more sophisticated but efficient sensors. In their paper, Kusupati and the team uses a low cost, Arm Cortex M3 processor, which has only 96 KB of RAM for a more cost and energy-efficient solution.

“Imagine a situation inside a wildlife reserve, secluded from the modern world, with intermittent network connectivity and vast expanses to monitor,” said Kusupati. “You would ideally want to catch poachers, or intruders in a general setting, as soon as possible using minimal energy as the battery on your radar is limited since you don’t have electricity in the reserve. The technique we developed is the first deep learning based real-time solution that’s at least three times faster and more accurate than the existing state-of-the-art methods.”

The techniques proposed in the paper were also put to use to create a demonstration which was additionally presented at the conference. The solution proposed is hierarchical and so computationally very efficient while the models generated are tiny enough to fit and work in 96 kilobytes of RAM.

“The applications are varied and in the case of a more general use, it could be used as an inexpensive and non-intrusive intruder detection system for smart cities and even enable smart lights and more,” said Kusupati.  

Aditya worked on this while at Microsoft Research India before joining the Allen School. Collaborators include Ohio State University Ph.D. students Dhrubojyoti Roy, Sangeeta Srivastava and professor Anish AroraPranshu Jain, a Ph.D. student in IIT Delhi and Manik Varma, senior principal researcher at Microsoft Research India.

To learn more, read the research paper here.