基于视频监控的人流量统计系统设计

发布时间：2018-03-03 08:10

本文选题：视频监控　切入点：人流量统计　出处：《合肥工业大学》2015年硕士论文　论文类型：学位论文

【摘要】：针对目前世界各国对人流量统计的迫切需求,本文致力于开发一种基于视频监控的人流量统计系统。作为视频监控领域的重要应用和研究热点,人流量统计系统在车站安防、商场超市和交通管理等方面被广泛使用。系统主要建立在计算机视觉和模式识别技术的基础上,通过对运动目标检测和人体目标识别、跟踪算法的深入研究,最终实现了人数统计的目的。本文的主要工作如下：(1)本文在对比多种运动检测方法后,选择帧差法中的三帧差分法作为目标检测的核心算法,同时在此基础上融合了最大类间方差法,达到动态取阈值的目的。此外,为了降低改进后算法的时间复杂度,算法在阂值自适应前进行了预先判定,达到优先保证检测效果的同时最大化地提高算法的实时性。(2)人体的目标识别方面,系统采取SVM向量机分类HOG特征的识别方法。通过对HOG梯度信息的提取和SVM向量机的样本训练来判定检测目标是否属于人体。(3)对于目标跟踪部分,本文仔细分析了常用的Cam-Shift跟踪算法的缺陷,进行了相应的优化改进。对于复杂背景问题,系统加入了多特征融合实现最大程度的提取目标信息；对于遮挡干扰,系统增加了改进的Kalman滤波算法进行运动趋势预测。(4)基于对上述关键技术的研究,本文成功完成了系统软、硬件以及客户端平台的搭建。同时经实测证明,系统能够实时可靠地进行人流量统计,具有良好的实践价值。
[Abstract]:In view of the urgent demand of people flow statistics in the world at present, this paper is devoted to the development of a human flow statistics system based on video surveillance. As an important application and research hotspot in the field of video surveillance, the human flow statistics system is used in the station security protection. Shopping malls, supermarkets and traffic management are widely used. The system is mainly based on computer vision and pattern recognition technology. The main work of this paper is as follows: 1. After comparing many methods of motion detection, this paper chooses the three-frame difference method of frame difference method as the core algorithm of target detection. In addition, in order to reduce the time complexity of the improved algorithm, the threshold value is pre-determined before adaptive threshold. To achieve priority to ensure the detection effect while maximizing the real-time performance of the algorithm. The system adopts SVM vector machine to classify HOG features. By extracting the HOG gradient information and training the samples of SVM vector machine to determine whether the detection object belongs to human body. In this paper, the defects of common Cam-Shift tracking algorithms are analyzed carefully, and the corresponding optimization improvements are made. For complex background problems, the system adds multi-feature fusion to maximize the extraction of target information, and for occlusion interference, The system adds the improved Kalman filter algorithm to predict the motion trend. (4) based on the research of the above key technologies, the software, hardware and client platform of the system are successfully built. The system can count the flow of people in real time and reliably, and has good practical value.
【学位授予单位】：合肥工业大学
【学位级别】：硕士
【学位授予年份】：2015
【分类号】：TN948.6

【相似文献】