Detection of close-proximity automotive targets using LSTM

Constant false alarm rate (CFAR) detector is widely used in automotive radar for target detection from the range-Doppler (RD) map. Due to the large target size relative to the range resolution of the radar, the target response is spread in the RD map. The presence of multiple closely-situated targets contributes to the scene's complexity.

Multiple detections provided by CFAR in such scenarios pose a challenge to the identification of the actual number of targets. We propose two LSTM-based networks to resolve the number of targets present in the scene. The performance of the proposed networks is presented in comparison with Resnet-50.