基于MEMS惯性传感器时域特征的人体行为识别

提出了一种基于微机电系统(MEMS)惯性传感器组合系统的高精度实时人体行为识别算法。算法选取一个2 s的滑动时间窗作为特征提取窗口,提取惯性传感器组合系统输出的时域特征作为特征参量,采用基于平衡决策树的支持向量机对人体不同行为模式进行分类识别。在实验室自主研发的可穿戴智能终端平台上进行测试,结果表明,在识别时间缩短到2 s/次的条件下,对5种行走类行为模式和5种非行走类行为模式的识别率均可达88%。与现有算法相比,该算法的实时性和精度得到明显提高,且拓展了模式识别的种类。

Human Behavior Recognition Based on Time domain Features of MEMS Inertial Sensors

A high precision real time human behavior recognition algorithm based on MEMS inertial sensor combination system is proposed in this paper. The time domain features of the inertial sensor combined system are extracted as feature parametersin this algorithm by selecting a sliding time window of 2 s as the feature extraction window, and the support vector machine based on a balanced decision tree is used to classify and recognize various human behavior patterns. The tests of the proposed algorithm are carried out on the wearable intelligent terminal platform independently developed by the laboratory. The resultsshow that the recognition rate of five walking behavior patterns and five non walking behavior patterns are up to 88% when the recognition time is reduced to 2 s/time. Compared with the existing algorithms, the real time performance and accuracy of the proposed algorithm are significantly improved, and the types of pattern recognition are expanded.