通道场景下人群统计系统的设计与实现

发布时间：2018-06-25 21:03

本文选题：人群统计 + 头肩轮廓特征　；参考：《电子科技大学》2014年硕士论文

【摘要】：目前,智能视频监控领域飞速发展,视频监控应用到日常生活中的方方面面。智能视频监控就是使用计算机视觉和图像处理的相关处理方法,将图像中的待检测目标检测出来,对图像中待检测目标的行为特征进行理解,而且行人目标的检测及跟踪是目前图像处理研究中的一大热点。本文研究的是“智能监控关键技术及其应用研究”中的子问题——“通道场景下的人群统计系统”,实现对道场景下的行人进行检测和流量统计。本文通过对基于视频监控的行人检测技术进行研究分析,在已有研究成果的基础之上,实现了通道场景下人群统计系统的系统原型,论文的主要工作包括:1、人体运动目标检测算法,综合快速三帧差分和头肩轮廓特征对人体运动目标进行快速检测,通过人体对头肩轮廓模型的检测,实现人体目标的快速检测,以满足实时性要求。2、头肩轮廓模型检测方法,首先提取头肩轮廓特征,将提取出来的头肩轮廓与样本库中的头肩轮廓进行匹配,若匹配成功,则判断该目标为人体目标,否则为非人体目标。3、人体运动目标跟踪算法,综合利用人体运动目标检测阶段得到人体头肩轮廓,用对人体头肩的跟踪来代替对人体运动目标的跟踪。采用人体头肩轮廓特征,结合卡尔曼滤波,通过将人体头肩轮廓与人体头肩样本库中的样本进行匹配,实现对人体头肩的跟踪,采用人体头肩的最小外接矩形来代替整个人体运动目标。4、多个人体运动目标跟踪算法,该算法对多个人体运动目标在运动过程中产生的遮挡、重叠、分离情况进行了处理,能够实现对多个人体运动目标的跟踪检测。5、人群流量统计算法,该算法通过设置扩展双向计数线,对通过计数线的行人目标进行计数,实现对人体目标的流量统计。本原型系统是在Windows操作系统下的VS2008平台上,运用OpenCV库进行开发的,系统开发语言是C++,硬件包含一个摄像头,一台笔记本电脑,其中CPU主频2.0GHZ,内存2G。
[Abstract]:At present, the field of intelligent video surveillance is developing rapidly, and video surveillance is applied to every aspect of daily life. Intelligent video surveillance is to use computer vision and image processing related processing methods to detect the target to be detected in the image and to understand the behavior characteristics of the target to be detected in the image. And pedestrian target detection and tracking is a hot spot in image processing. In this paper, the key technology of intelligent monitoring and its application is studied, which is called "crowd Statistics system in Channel scene", which realizes the detection and flow statistics of pedestrians in the road scene. Based on the research and analysis of pedestrian detection technology based on video surveillance, this paper realizes the prototype of the crowd statistics system in the channel scene based on the existing research results. The main work of this paper includes: 1, human body moving target detection algorithm, combining fast three frame difference and head-shoulder contour features to quickly detect human moving target, through the detection of human body head-shoulder contour model, to realize the rapid detection of human body target. In order to meet the requirement of real time, the head-shoulder contour model detection method firstly extracts the head-shoulder contour feature, and matches the head-shoulder contour with the head-shoulder contour in the sample database. If the matching is successful, the target is judged as the human body target. Otherwise, it is a non-human target. 3, the human body moving target tracking algorithm, using the human body moving target detection stage to get the human head and shoulder contour, using the human head and shoulder tracking to replace the human body moving target tracking. By using the human head-shoulder contour feature and Kalman filter, the head-shoulder contour of the human body is matched with the sample in the head-shoulder sample database to track the head-shoulder of the human body. The minimum outer rectangle of the head and shoulder of the human body is used to replace the whole human body moving object. 4, and the tracking algorithm of multiple human moving targets is adopted. The algorithm deals with the occlusion, overlap and separation of multiple human moving targets in the process of motion. The algorithm can realize the tracking detection of multiple human moving targets. 5. The algorithm of crowd flow statistics can count the pedestrian targets passing through the counting line by setting the extended two-way counting line and realize the flow statistics of human body targets. The prototype system is developed on the VS2008 platform of Windows operating system using OpenCV library. The system development language is C, the hardware includes a camera, a notebook computer, in which the CPU main frequency 2.0GHz, memory 2G.
【学位授予单位】：电子科技大学
【学位级别】：硕士
【学位授予年份】：2014
【分类号】：TP391.41;TN948.6

【参考文献】