电梯轿厢内乘客异常行为检测
发布时间：2018-08-18 13:26
【摘要】：电梯轿厢的内部空间狭小、密闭,是摔倒、侵害、抢劫和群体恐慌等事件的多发之地。通过视频监控,预防上述事件的发生对于维护民生安全具有非常重要的现实意义。本文对电梯轿厢内的异常行为检测问题进行了较深入的研究,取得了一些有价值的研究成果。论文的主要工作包括:1.设计并实现了一个结构上较为完整的实时的电梯轿厢内异常行为的图像检测系统。首先用减背景的方法提取图像中的人体目标,然后对检测出的人体目标区域的像素进行统计分析以实现对电梯轿厢内人数的估计。2.根据电梯轿厢内人数的多寡,构建了针对单人、两人及多人异常行为检测的模型和算法。当电梯轿厢内仅有一位乘客时,主要通过对目标区域的投影处理检测乘客是否有摔倒或蹲伏行为。当电梯轿厢内有两位乘客时,主要检测是否有暴力行为发生。通过计算乘客的运动历史图,获取运动的时空表征,并据此构造能量函数,通过能量的多少判断乘客之间是否有暴力行为发生。而当电梯轿厢内有多位乘客时,主要检测是否有群体恐慌行为发生。首先计算乘客的运动能量图,然后将运动能量图分别向水平和竖直两个方向投影,得到能量的空间分布直方图,最后根据直方图的空间分布特征,判定是否有群体性慌乱行为发生。3.为了实际验证本文所提出的电梯轿厢内乘客异常行为检测方法的有效性,采集和构建了不同情况下乘客在电梯轿厢内各种异常行为的数据集。该数据集包含单人视频34个,两人视频28个,多人视频26个。在该数据集上对本文所提出的上述算法的测试结果表明,本文所提出的异常行为检测算法简洁、实时、有效。
[Abstract]:The interior of the elevator car is small, closed and prone to falls, assaults, robberies and mass panic. Video surveillance to prevent the occurrence of these incidents for the maintenance of livelihood security has a very important practical significance. In this paper, the detection of abnormal behavior in elevator car is studied deeply, and some valuable research results are obtained. The main work of the thesis includes 1: 1. A complete real-time image detection system for abnormal behavior in elevator car is designed and implemented. Firstly, the human object in the image is extracted by subtraction method, and then the pixels of the detected human target area are statistically analyzed to estimate the number of people in the elevator car. According to the number of people in the elevator car, a model and algorithm for detecting abnormal behavior of single person, two person and many people is constructed. When there is only one passenger in the elevator car, it mainly detects whether the passenger falls or crouches by projecting the target area. When there are two passengers in the elevator car, violence is mainly detected. By calculating the movement history map of passengers, the space-time representation of motion is obtained, and the energy function is constructed accordingly, and the number of energy can be used to judge whether there is violence among passengers. When there are many passengers in the elevator car, the main detection is whether there is panic behavior. First, the motion energy map of passengers is calculated, then the motion energy map is projected horizontally and vertically, and the spatial distribution histogram of energy is obtained. Finally, according to the spatial distribution characteristics of the histogram, Determine whether group panic behavior occurred. 3. In order to verify the effectiveness of the method proposed in this paper for detecting the abnormal behavior of passengers in the elevator car, the data sets of passengers' abnormal behavior in the elevator car are collected and constructed. The dataset consists of 34 single-person videos, 28 two-person videos and 26 multi-person videos. The test results on the data set show that the algorithm proposed in this paper is simple, real-time and effective.
【学位授予单位】：中国科学技术大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：TU857;TP391.41
本文编号：2189607
[Abstract]:The interior of the elevator car is small, closed and prone to falls, assaults, robberies and mass panic. Video surveillance to prevent the occurrence of these incidents for the maintenance of livelihood security has a very important practical significance. In this paper, the detection of abnormal behavior in elevator car is studied deeply, and some valuable research results are obtained. The main work of the thesis includes 1: 1. A complete real-time image detection system for abnormal behavior in elevator car is designed and implemented. Firstly, the human object in the image is extracted by subtraction method, and then the pixels of the detected human target area are statistically analyzed to estimate the number of people in the elevator car. According to the number of people in the elevator car, a model and algorithm for detecting abnormal behavior of single person, two person and many people is constructed. When there is only one passenger in the elevator car, it mainly detects whether the passenger falls or crouches by projecting the target area. When there are two passengers in the elevator car, violence is mainly detected. By calculating the movement history map of passengers, the space-time representation of motion is obtained, and the energy function is constructed accordingly, and the number of energy can be used to judge whether there is violence among passengers. When there are many passengers in the elevator car, the main detection is whether there is panic behavior. First, the motion energy map of passengers is calculated, then the motion energy map is projected horizontally and vertically, and the spatial distribution histogram of energy is obtained. Finally, according to the spatial distribution characteristics of the histogram, Determine whether group panic behavior occurred. 3. In order to verify the effectiveness of the method proposed in this paper for detecting the abnormal behavior of passengers in the elevator car, the data sets of passengers' abnormal behavior in the elevator car are collected and constructed. The dataset consists of 34 single-person videos, 28 two-person videos and 26 multi-person videos. The test results on the data set show that the algorithm proposed in this paper is simple, real-time and effective.
【学位授予单位】：中国科学技术大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：TU857;TP391.41
【参考文献】
相关期刊论文 前1条
1 汤志伟;李万才;沈冬青;;智能高清视频监控在安防领域的应用与发展[J];中国安防;2010年07期
相关硕士学位论文 前3条
1 付高静;电梯轿厢内异常行为识别研究[D];哈尔滨理工大学;2015年
2 靳海燕;基于视频分析的电梯轿厢内异常行为检测研究[D];重庆大学;2012年
3 陆海峰;基于计算机视觉的电梯轿厢内异常行为检测[D];浙江工业大学;2009年
，本文编号：2189607
本文链接：https://www.wllwen.com/shoufeilunwen/xixikjs/2189607.html
