Human Motion Recognition Based on Feature Fusion and Transfer Learning

In order to solve the problem that the recognition accuracy of human motion is not high when a single feature is used, a feature fusion human motion recognition method based on Frequency Modulated Continuous Wave (FMCW) radar is proposed. By preprocessing the FMCW radar echo data, the range and Doppler parameters of human motions are obtained, and the range-time feature map and Doppler-time feature map datasets are constructed. In order to fully extract and accurately identify the human motion features, the two features are fused, and then the two features maps and feature fusion spectrograms are put into the VGG16 network model based on transfer learning for identification and classification. Experimental results show that this method can effectively solve the problem of lack of information and recognition rate of single feature motion recognition, and the recognition accuracy is more than 1% higher than that of the single feature recognition method. © 2024, Electromagnetics Academy. All rights reserved.