Research on SVM gait recognition based on single acceleration sensor

In this paper, a method of human motion recognition based on single position sensor with multiple eigenvalues is proposed. The data are analyzed in time domain and frequency domain respectively, and then nine related features are extracted by feature engineering. The experiment compares the pattern recognition methods of belt, pants pocket and coat pocket based on SVM, and classifies five kinds of motion patterns: walking, running, climbing stairs, descending stairs and relaxing. The results show that the comprehensive application of the six features has the best recognition effect, which can reach more than 98%.