基于三个特征点的人体跌倒检测

设计了视频监控跌倒检测系统,采用人体姿态分析,通过人体轮廓提取、星形骨架提取、人体部位识别等步骤对视频图像进行处理,首先检测出三个特征点。由特征点构成了跌倒判断的特征向量。根据三个特征点相对位置的变化和特征向量与水平地面的夹角,来区分跌倒与正常人体活动的差别。以3个志愿者的步行、坐下和蹲下三类最具代表性动作为实验样本,采用上述方法进行测试。实验结果表明系统能有效检测跌倒,总体成功率超过90%,对正常活动的误报率仅7.5%。

According to Computer vision technology, a video surveillance fall detection system designed which using human gesture analysis. Respectively for the body contour extraction, star skeleton extraction and body parts recognition technology processing the monitor image. Check out the three feature points, constitutes a fall determination eigenveetor by the feature point. According the relative positioning of the three feature points and feature vector angle with the horizontal ground, to distinguish the difference between the normal human activity and fails. Using the above method test 3 volunteers to walk, sit and squat which three most representative action for the experimental sample. Experimental results show that the system can effectively detect a fall, the overall success rate of over 90%, the normal activities of a false positive rate of only 7.5%.