Radar Based Pedestrian Classification using Deep Learning Approach

Developments in sensor fusion systems led to extensive research autonomous vehicles, and object detection is a crucial aspect of vehicle operation. Detecting obstacles can be difficult due the wide range potential obstructions, characteristics each sensor, influence surrounding environment. In this paper, authors use automotive radar data various neural networks classify objects (vehicles, single multiple people, bicycles). Combined with radar, proposed rapid algorithmic implementation for locating, monitoring extracting micro-Doppler. The evaluate three distinct network architectures five recorded classes targets: basic CNN, residual network, combined architecture convolution recurrent layers. Considerable accuracy 95.6% immediately before identification spectrogram from (~0.55 s produce 0.5 long spectrogram).