基于视频的电梯乘客异常状态检测算法研发

发布时间：2018-06-01 23:30

本文选题：视频分析 + 目标检测和跟踪　；参考：《浙江大学》2017年硕士论文

【摘要】：随着城市化的快速发展,涌现出大量的高层建筑,极大依赖了电梯系统,如何提高电梯的安全性和运行效率已成为迫切需求。本文分析了当前电梯存在的安全问题,主要包括人员超载、打斗、放置危险品等乘客异常状态,在不需要人为干预的情况下,使用监控视频信息分析电梯乘客状态,检测出异常后进行报警,具有较好的研究意义和工程应用价值。本文研发了基于视频的电梯人数统计算法、电梯人员异常行为检测算法以及电梯遗留物检测算法。采用基于VIBE背景建模算法,实现目标检测;级联Adaboost和SVM分类器,并辅以Hough圆变换进行人头识别;同时使用基于MedianFlow算法进行跟踪计数,实现了电梯实时人数统计算法。采用运动估计提取运动信息,本文在搜索策略和匹配函数上改进了经典的三步运动搜索法;然后计算运动区域的熵值,并根据运动信息线索判断是否出现异常行为,最后实现电梯人员异常行为检测算法。采用基于VIBE的双背景模型提取目标前景,并对输出图像进行形态学滤波和连通域分析;判断物体短暂静止,并在其主体离开后进行计时,最后实现遗留物检测算法。本文对基于视频的电梯乘客异常状态的三个检测算法分别进行了测试,并分析了测试结果,可准确实时地统计乘客人数、检测乘客异常行为以及检测遗留物。
[Abstract]:With the rapid development of urbanization, a large number of high-rise buildings have emerged, greatly dependent on the elevator system, how to improve the security and operational efficiency of elevators has become an urgent need. This paper analyzes the current elevator safety problems, including personnel overload, fighting, placing dangerous goods and other passengers abnormal state, without the need for human intervention, the use of monitoring video information to analyze elevator passenger status, It has good research significance and engineering application value to alarm after detecting anomaly. In this paper, a video based algorithm of elevator population statistics, elevator personnel abnormal behavior detection algorithm and elevator residue detection algorithm is developed. The algorithm based on VIBE background modeling is used to realize target detection; cascaded Adaboost and SVM classifier, supplemented by Hough circle transform, are used to recognize human head; at the same time, based on MedianFlow algorithm to track and count, the elevator real-time number statistics algorithm is realized. Using motion estimation to extract motion information, this paper improves the classical three-step motion search method in search strategy and matching function, then calculates the entropy of motion region, and determines whether abnormal behavior occurs according to the clues of motion information. Finally, the algorithm of detecting the abnormal behavior of elevator personnel is realized. The dual-background model based on VIBE is used to extract the target foreground, and the output image is analyzed by morphological filtering and connected domain analysis. In this paper, three algorithms for detecting abnormal status of elevator passengers based on video are tested, and the test results are analyzed, which can accurately and real-time count the number of passengers, detect the abnormal behavior of passengers and detect the remnants.
【学位授予单位】：浙江大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：TP391.41

【参考文献】