基于骨架平衡的3D人体异常行为识别方法仿真

Simulation of 3D Human Abnormal Behavior Recognition Method Based on Skeleton Balance

为提高对人体异常行为的识别精准度,提出基于3D骨架的人体异常行为识别算法优化方法。根据3D骨架结构信息,建立人体3D骨架模型,依据关节点的运动速度及坐标变化,提取各动作的相关运动特征。通过搭建概率协作分类器,完成对3D骨架对应行为的分类。针对动作行为识别表征向量建立对应分类识别模型,设定人体骨架平衡参数,在先验信息的基础上训练最优系数向量,建立异常行为模型库。根据任意运动特征分布特点,排除异常干扰数据,建立隶属度函数集合确定动作主次结构,计算出每个运动特征权重值,识别异常行为。实验结果表明,所提方法具备良好的识别性能,识别准确度高,延迟小。

In order to improve the accuracy of recognizing abnormal human behavior,a method of optimizing abnormal human behavior recognition was proposed based on 3D skeleton.According to the information of the 3D skeleton structure,we built a 3D skeleton model of the human body,and then extracted relevant motion features of action by the movement speed and coordinate changes of joint points.After constructing a probability cooperative classifier,we completed the classification of the corresponding behaviors of the 3D skeleton.Correspondingly,we constructed a classification recognition model for the behavioral recognition representation vector and established the balance parameters of the human skeleton.On the basis of prior information,we trained the optimal coefficient vector and built an abnormal behavior model library.According to the distribution characteristics of any motion feature,we excluded abnormal data,and established a membership function set to determine the main and secondary structure of actions,and thus to calculate the weight of each motion feature.Finally,we determined the recognition of abnormal behavior.Experimental results show that the proposed method had good recognition performance,high accuracy,and small delay.