摘要
道路人群拥挤行人异常行为智能检测方法的研究影响行人异常行为模式的变化,在图像、视频和生活领域具有较好的发展前景。针对当前方法存在识别率不均衡的问题,提出了一种基于投影近似子空间估计的异常行为检测方法。对异常行为样本的置信度进行取值,计算人群拥挤行人异常行为的距离函数,利用异常行为抽样来衡量行人异常行为样本的多样性,并对样本间的余弦角距离进行计算,分析异常行为样本的多样性和不确定性,在对道路人群拥挤行人异常行为抽样的基础上,利用抽样得到的数据对行人异常行为进行数据最小化重构,参考和估计行人异常行为投影近似子空间,通过计算得到第一个异常行为投影近似基,继续进行下一个异常行为投影近似基的求解,对行为向量的异常程度进行判断。仿真结果表明,提出方法具有较好的识别率,提高了行人异常行为检测的可行性,为后续实现道路人群拥挤行人异常行为的检测奠定了良好基础。
An abnormal behavior detection method based on approximate subspace estimation of projection was proposed. The confidence level of abnormal behavior samples was obtained to calculate the distance function of crowded pedestrian abnormal behavior. The abnormal behavior sample was used to measure the variety of pedestrian abnormal behavior samples,and the cosine angular distance between the samples was calculated. The variety and the uncertainty of abnormal behavior samples were analyzed. On the basis of sampling the abnormal behavior of pedestrians on roads,the data obtained from pedestrians were minimized by using the sampled data. An approximate projection subspace of pedestrian abnormal behavior was consulted and estimated,and the first abnormal behavior projection approximate base was obtained by calculating. We continued to solve the next approximate projection of the abnormal behavior and judged the abnormal degree of the action vector. The simulation results show that the proposed method has a better recognition rate and good feasibility of pedestrian abnormal behavior detection.
作者
逄焕利
李红岩
PANG Huan-li;LI Hong-yan(Changchun University of Technology,Changchtm Jilin 130012,China)
出处
《计算机仿真》
北大核心
2018年第11期405-408,共4页
Computer Simulation
基金
吉林省科技发展计划项目(20180201129GX)
关键词
拥挤行人
异常行为
智能
检测
Crowded Pedestrian
Abnormal Behavior
Intelligent
Detection
