tinyML Talks on June 6, 2022 “tinyRadar: mmWave Radar-based Human Activity Classification for Edge Computing” by Radha Agarwal

We held our next tinyML Talks webcast. Radha Agarwal from Indian Institute of Science, Bangalore presented tinyRadar: mmWave Radar-based Human Activity Classification for Edge Computing on June 6, 2022.

June 6 forum

Most of the current systems for patient monitoring, elderly, and child care are camera-based and often require cloud computing. But, camera-based systems pose a privacy risk, and cloud computing can lead to higher latency, data theft, and connectivity issues. Why face these challenges in the current era of intelligent sensing modalities with tiny and edge solutions?
This talk will give insights about a tinyML-based single-chip radar solution for on-edge sensing and detection of the environment. Thus, the hassle can be avoided by using the tiny radar, which protects privacy, and works in all weather and lighting conditions while sensing with a contactless interface. At the same time, edge computing on it gives a small form factor that makes it robust enough for remote deployments. This end-to-end pipeline from sensing to detection is demonstrated for real-time human activity classification using a Texas Instruments IWR6843 millimeter-wave radar board. The edge implementation of the 8-bit quantized inference engine is done on the radar’s Cortex-R4F MCU using the CMSIS-NN custom APIs.

Radha Agarwal is a master’s student in the Department of Electronic Systems Engineering, Indian Institute of Science (IISc), Bangalore, India. She is a Texas Instruments scholar working on a mmWave radar project in the NeuRonICS Lab, IISc under the guidance of Dr. Chetan Singh Thakur. She has done several projects involving machine learning, embedded systems, and digital design using FPGA. Her fields of interest include applications of machine learning and embedded systems. In 2020, she was awarded the prestigious Governor’s silver medal for her academic performance in B.Tech.

=========================

Watch on YouTube:
Radha Agarwal

Download presentation slides:
Radha Agarwal

Feel free to ask your questions on this thread and keep the conversation going!

Hi Olga / Radha Agarwal ,
I have gone through the paper , i have following questions. It would really help if you can throw some light on these questions.

  • As per the ARM forums / websites (Cortex-R support · Issue #314 · ARM-software/CMSIS_5 · GitHub) CMSIS-NN is not optimized for ARM Cortex R processors. So how did you guys manage to deploy ML Models on R4 using CMSIS-NN. Did you guys build without M-Profile Vector Extensions as mentioned in the github issue ?

  • As per the paper “The classification network utilized a total memory footprint of 11.07KB on Cortex®-R4F MCU”. Does 11.07kB include the CMSIS-NN SDK or is it just the Network ?.

Thanks.