Human identification based on mmwave radar using deep convolutional neural network

Xuping Xie

Abstract:

With the development of computer vision
technology, camera-based face recognition technology provides
convenience in many aspects. However, large-scale application
of face recognition has brought up privacy concern. Compared
with the camera, millimeter-wave radar for identification is a
more secure device due to its no concern in t privacy leakage. In
this paper, millimeter-wave radar is employed to collect RangeDoppler heat map data in five experimenters of people walking.
Classic deep CNN models such as AlexNet, VGGNet,
GoogLeNet, and ResNet are used to extract features from the
heat maps. The features are extracted by the models and the
feature vector is subsequently sent to the classifier for identity
recognition. It is demonstrated that that it is feasible to use the
deep CNN model and Range-Doppler heat map for identity
recognition. The test accuracy of all models is over 96% and the
ResNet model has the highest accuracy of 97.9%