Application of Spiking Neural Networks for Action Recognition...

Application of Spiking Neural Networks for Action Recognition from Radar Data" by Arijit Mukherjee.   
Application of Spiking Neural Networks for ActionRecognition from Radar Data
Dighanchal Banerjee
∗
, Smriti Rani
†
, Arun M. George
‡
, Arijit Chowdhury
§
,Sounak Dey
¶
, Arijit Mukherjee

, Tapas Chakravarty
∗∗
, Arpan Pal
††
TCS Research & Innovation, Kolkata, IndiaEmail:
∗
[email protected],
†
[email protected],
‡
[email protected],
§
[email protected],
¶
[email protected],

[email protected],
∗∗
[email protected],
††
[email protected]   
Abstract
—In the past two decades, radar-based human sensinghas become a topic of intense research. Unlike vision-based tech-niques which require the use of camera, radars are unobtrusiveand privacy preserving in nature. Further, radars are agnosticof the lighting conditions and can be used for through-the-wallimaging thereby making them hugely effective in many situations.Compact, affordable radars have been designed that can beeasily integrated with remote monitoring systems. However, theclassical machine learning techniques currently used for learningand inferring human actions from radar images are computeintensive, and require large volume of training data, making themunsuitable for deployment on the network edge. In this paper,we propose to use the concepts of neuromorphic computing andSpiking Neural Networks (SNN) to learn human actions fromdata captured by the radar. To the best our knowledge, this isthe first attempt of using SNNs on micro-Doppler data fromradars. Our SNN model is capable of learning spatial as wellas temporal features from the data and our experiments haveresulted in 85% accuracy which is comparable with the classicalmachine learning approaches that are typically used on similardata. Further, the use of neuromorphic and SNN concepts makeour model deployable over evolving neuromorphic edge devicesthereby making the entire approach more efficient in terms of data, computation and energy consumption