Human Identification Based on Micro-Doppler Images Using Residual Networks

Importance of human identification is well known in surveillance systems. In this sense, radar is playing a key role in remote observation systems due to its ability in working in unsuitable weather and insufficient light. In this paper, human identification is investigated based on micro-Doppler signatures acquired from human walking and employing ResNet deep convolutional networks. Required micro-Doppler images are generated by simulation the backscattered signal of different persons using a Kinect sensor in absence of the radar. Then these images are fed to ResNet network to recognize person's identity. It is shown that this method can achieve more accuracy in larger group of people which are 95.53% in identifying 30 different persons and 97.9
^%
for 14 people.