基于突变点检测的智能空间行为识别方法

Intelligent Spatial Behavior Recognition Method Based on Mutation Detection

基于非嵌入式传感器数据的行为识别对家居设备控制、异常行为监测非常重要,是智能空间环境下行为识别的研究热点,不仅利于隐私保护而且能长期积累数据满足个体行为偏好。针对传感器数据序列中行为边界标识,并依次改善在线行为识别效果的问题,基于行为突变点检测思想识别连续行为的相似度程度,使用KL散度实现突变点检测,针对突变点检测阈值的选择问题,使用遗传算法对其进行自动设置。使用RF、QSVM、加权K近邻(Weighted KNN,wKNN)、DT算法实验验证突变点时域特征能够有效提高在线行为识别能力,证明了本文方法的有效性。

Behavior recognition based on non embedded sensor data is very important for home equipment control and abnormal behavior monitoring.It is a research hotspot of behavior recognition in intelligent space environment.It is not only conducive to privacy protection but also can accumulate data for a long time to meet individual behavior preferences.Aiming at the problem of identifying behavior boundary in sensor data sequence and improving the effect of online behavior recognition in turn,based on the idea of behavior mutation detection to identify the similarity degree of continuous behavior,KL diver-gence is used to realize mutation detection,and genetic algorithm is used to automatically set the threshold of mutation detection.Using RF,QSVM,weighted KNN(wKNN)and DT algorithm to verify the time-domain feature of mutation point can effectively improve the ability of online behavior recognition,which proves the effectiveness of this method.