Pedestrian Gait Recognition Method Based on Micro-Doppler and Power Spectrum
Benteng Lyu,Aijun Liu,Changjun Yu,Shuai Xia
DOI: https://doi.org/10.1109/piers62282.2024.10618870
2024-01-01
Abstract:Pedestrian gait recognition has been widely utilized in intelligent transportation, personnel monitoring, health monitoring, and other fields. Radar-based gait recognition technology has gradually been applied in various domains. However, traditional radar-based point cloud gait recognition technique has encountered issues such as sparse point clouds, a high false alarm recognition rate, and low accuracy. In contrast, gait recognition methods based on pedestrian micro-Doppler features offer advantages of stability, accuracy, and difficulty in camouflage. To address the issues of existing pedestrian gait recognition methods, which rely on a large number of training samples and exhibit low recognition accuracy, this thesis proposes a deep learning-based pedestrian gait recognition method that leverages micro-Doppler and power spectrum analysis. Firstly, based on the Boulic human walking model, four gait models are established: normal walking, lame walking, walking without swinging arms, and turning down. The radar echo data of human walking processes are obtained through simulation. Secondly, power spectrum estimation and time-frequency analysis are conducted on the radar echoes of different gaits. The smoothed pseudo Wigner-Ville distribution is employed to obtain time-frequency graphs for each gait. Subsequently, an envelope extraction method based on energy distribution is applied to process the time-frequency graphs and extract the micro-Doppler envelope of different gaits. Finally, the thesis utilizes a fully connected neural network based on power spectrum, a convolutional neural network based on time-frequency graph and another fully connected neural network based on envelope graph for gait recognition. The gait recognition rate is improved by the decision fusion technique of adaptive class weights. Experimental simulation results demonstrate that the proposed algorithm achieves higher recognition accuracy compared to other algorithms under the condition of small samples. This research primarily relies on simulated data to explore the potential of deep learning algorithms based on micro-Doppler and power spectrum analysis in gait recognition. It provides a theoretical foundation for future experiments utilizing actual measured data.