Pedestrian and Bicyclist Identification Through Micro Doppler Signature with Different Approaching Aspect Angles

In realistic road scenarios, two main urban hazards, pedestrians and bicyclists, can approach the radar from any aspect angle which will affect the ability of a radar system to perform target identification. However, the identification of these two particular types of targets with different approaching directions has not been investigated thoroughly until now. In this paper, we explore the micro Doppler signature to accurately distinguish pedestrian and the bicyclist. One feature extraction method that is based on the sparse coding of a measured micro Doppler signature is proposed and validated. Further numerical analysis of the sparse matrices was performed to reduce the number of dimensions whilst keeping a sufficient number of features. As on comparison, the two feature extractors, i.e., singular value decomposition and robust principal component analysis are extensively studied and compared with the proposed method. Support vector machine (SVM) and extreme learning machine were adopted to classify the extracted features. Indoor experimental results indicate that the proposed methods can achieve higher identification accuracies than alternative approaches and the outdoor measurement results presented show good performance with realistic targets.