人群监控图像异常轨迹数据聚类识别仿真
|
摘要
实现人群监控图像异常轨迹数据的聚类识别,对紧急突发事件发生时及时报警,方便安保人员及时采取应对措施,保证人们生命财产安全具有重要意义。针对当前方法存在识别准确率较低的问题,提出一种基于运动特征的异常轨迹数据聚类识别方法,将人群监控图像中任意行人的轨迹数据描述为一个流向量序列,提取被监测行人的运动特征,对人群监控图像中被监测行人所有轨迹的起点集合和终点集合进行计算,并采用Hausdorff距离对两个集合中长度不等的行人轨迹进行相似度计算,实现人群监控图像行人轨迹数据预处理。采用人虚拟的最小外接矩形的中心点替代人群监控图像中被监测行人的重心,通过最小外接矩形中心的变化描述行人的跳跃、下蹲、爬行、跑、徘徊五种异常轨迹。根据人群监控图像中被监测行人运动轨迹连接线的波峰波谷存在性,实现异常轨迹数据聚类识别。仿真测试结果证明,所提方法能够区分人群监控图像中行人的正常行为轨迹和异常行为轨迹,且识别准确率较高。
A method of data clustering recognition of abnormal trajectory based on motion feature was proposed. Firstly, the trajectory data of any pedestrians in crowd monitoring image were described as sequences of flow vectors to extract the motion feature of monitored pedestrian. Secondly, the set of starting points and the set of endpoints of all the tracks of monitored pedestrians in crowd monitoring image were calculated, and Hausdorff distance was used to calculate the similarity degree of pedestrian trajectories with different lengths in two sets, so as to preprocess the data of pedestrian trajectory in crowd monitoring image. Moreover, the virtual center point of the minimum enclosing rectangle was used to replace the barycentre of monitored pedestrian in crowd monitoring image. According to the change of the minimum enclosing rectangle, five kinds of abnormal trajectories such as jumping, squatting, creeping, running and hovering were described. Based on the existence of wave peaks and wave valleys of connecting line of monitored pedestrian trajectory in crowd monitoring image, the clustering recognition of abnormal trajectory data was realized. From simulation results, we can see that the proposed method can distinguish the normal behavior trajectory and abnormal behavior trajectory in crowd monitoring image. Meanwhile, the recognition accuracy rate is high.
A method of data clustering recognition of abnormal trajectory based on motion feature was proposed. Firstly, the trajectory data of any pedestrians in crowd monitoring image were described as sequences of flow vectors to extract the motion feature of monitored pedestrian. Secondly, the set of starting points and the set of endpoints of all the tracks of monitored pedestrians in crowd monitoring image were calculated, and Hausdorff distance was used to calculate the similarity degree of pedestrian trajectories with different lengths in two sets, so as to preprocess the data of pedestrian trajectory in crowd monitoring image. Moreover, the virtual center point of the minimum enclosing rectangle was used to replace the barycentre of monitored pedestrian in crowd monitoring image. According to the change of the minimum enclosing rectangle, five kinds of abnormal trajectories such as jumping, squatting, creeping, running and hovering were described. Based on the existence of wave peaks and wave valleys of connecting line of monitored pedestrian trajectory in crowd monitoring image, the clustering recognition of abnormal trajectory data was realized. From simulation results, we can see that the proposed method can distinguish the normal behavior trajectory and abnormal behavior trajectory in crowd monitoring image. Meanwhile, the recognition accuracy rate is high.
引文
[1] 胡瑗,夏利民,王嘉. 基于轨迹分析的行人异常行为识别[J]. 计算机工程与科学, 2017,39(11):2054-2059.
[2] 郑杰,李建平.物联网传感网络路由改进设计算法研究[J].科技通报,2017,33(3):92-95.
[3] 赵俊松,武建亮,刘依. 基于事件字典的行人异常事件检测[J]. 电视技术, 2017,41(7): 164-169.
[4] 王恬,等. 利用姿势估计实现人体异常行为识别[J]. 仪器仪表学报, 2016,37(10): 2366-2372.
[5] 何传阳,等. 基于智能监控的中小人群异常行为检测[J]. 计算机应用, 2016,36(6): 1724-1729.
[6] 潘新龙,等. 基于多维航迹特征的异常行为检测方法[J]. 航空学报, 2017,38(4): 249-258.
[7] 周培培,等. 基于DBSCAN聚类算法的异常轨迹检测[J]. 红外与激光工程, 2017,46(5): 230-237.
[8] 潘磊. 基于图像熵的密集人群异常事件实时检测方法[J]. 计算机科学与探索, 2016,10(7):1044-1050.
[9] 孙炜,等. 基于支持向量机优化的行人跟踪学习检测方法[J]. 湖南大学学报(自然科学版), 2016,43(10):102-109.
[10] 姜佰辰,等. 海上交通的船舶异常行为挖掘识别分析[J]. 计算机仿真, 2017,34(6): 329-334.
[2] 郑杰,李建平.物联网传感网络路由改进设计算法研究[J].科技通报,2017,33(3):92-95.
[3] 赵俊松,武建亮,刘依. 基于事件字典的行人异常事件检测[J]. 电视技术, 2017,41(7): 164-169.
[4] 王恬,等. 利用姿势估计实现人体异常行为识别[J]. 仪器仪表学报, 2016,37(10): 2366-2372.
[5] 何传阳,等. 基于智能监控的中小人群异常行为检测[J]. 计算机应用, 2016,36(6): 1724-1729.
[6] 潘新龙,等. 基于多维航迹特征的异常行为检测方法[J]. 航空学报, 2017,38(4): 249-258.
[7] 周培培,等. 基于DBSCAN聚类算法的异常轨迹检测[J]. 红外与激光工程, 2017,46(5): 230-237.
[8] 潘磊. 基于图像熵的密集人群异常事件实时检测方法[J]. 计算机科学与探索, 2016,10(7):1044-1050.
[9] 孙炜,等. 基于支持向量机优化的行人跟踪学习检测方法[J]. 湖南大学学报(自然科学版), 2016,43(10):102-109.
[10] 姜佰辰,等. 海上交通的船舶异常行为挖掘识别分析[J]. 计算机仿真, 2017,34(6): 329-334.