基于视频的人体异常行为检测与识别

摘要

A kind of the moving target detection algorithm was put forward.Human behavior picked up from video by frame difference method.Two kinds of behaviors including normal walking and falling were detected by frame difference method.The motion characteristics of the human body were picked up.A kind of identification algorithm about human abnormal be-havior was put forward based on improved Hu moment invariant.Two kinds of behaviors in-volving normal walking and falling were identified.Normal walking was normal behavior. Falling was abnormal behavior.The experiment proved that the recognition rate of the method was very high in the experimental environment.The method has great practical value.%提出了一种运动目标检测算法，通过三帧差分法，可以从视频图像中提取出运动目标的轮廓。通过三帧差分法分别对人体正常行走和人体摔倒两种行为进行检测，提取出人体在两种行为的运动过程中的特征。提出一种基于改进Hu矩不变性的人体异常行为识别算法，对人正常行走和摔倒两种行为进行识别。正常行走是正常行为，摔倒则为异常行为。实验证明，该方法对在本研究的实验环境下的正常行走和摔倒两种行为识别率很高，有一定实用价值。