基于静电信号的人体动作识别

A human motion recognition method by detecting electrostatic signals generated by human behaviors is proposed. Based on the analysis of the charge characteristics of human body, a static electricity detection system is designed to collect the electrostatic induction signals of 5 typical actions of the tested persons, i.e. walking, stepping, sitting down, taking the goods, and waving hand. The characteristic parameters of the collected 5 kinds of human body electrostatic signals are extracted, their significant differences are analyzed, and the characteristic parameters for classification are optimized. 3 kinds of classification algorithms including support vector machine, decision tree-C4.5 and random forest, are used based on Weka platform to classify the 250 collected signal samples by 10-fold cross-validation. The results show that the random forest algorithm obtains the best recognition effect with the accuracy of 99.6%. The research shows that the proposed action classification method based on human electrostatic signals for single environment can effectively identify typical human actions.