【作者】 李璟；

【导师】 洪留荣；

【作者基本信息】 淮北师范大学 ， 计算机软件与理论， 2010， 硕士

【摘要】 随着经济的高速发展,我国汽车保有量急剧增长,很多城市出现了“停车难”问题。露天停车场的科学管理对减少道路拥堵,保证城市交通安全有重要意义,而对其进行视频监控是实现科学管理的有效手段,在这其中运动车辆的检测与跟踪是技术关键。本文的主要研究内容是基于视频的运动车辆检测和跟踪问题,其中涉及到视频环境下的背景实时更新与生成、运动车辆的检测、运动车辆的跟踪和重叠车辆的分割等内容。在运动车辆检测方面,本文对背景差分法进行了改进,提出一种新的背景生成和更新算法。露天停车场场景的变化频率较为缓慢,因此可以认为车辆监控视频图像中每个像素的灰度值在较长一段时间内会稳定分布在某一个确定的范围,本文就根据图像中各个像素点的灰度取值所属区间次数进行背景提取;在背景更新时,考虑当背景中含有长时间静止车辆时的背景重构问题,结合边缘检测判断是否进行背景更新,如果静止车辆的状态发生变化,则利用前景图中对应区域的场景进行背景重构。在运动车辆跟踪方面,将跟踪过程分成特征提取、运动预测和目标匹配三部分。特征提取主要是来描述目标的形状特征(如运动目标的质心,目标外接矩形的大小);目标运动估计模块利用Kalman滤波器预测被跟踪运动目标在下一帧中可能处于的位置,确定目标搜索范围;目标匹配模块是寻找运动目标在图象序列各帧中的对应关系,从而确定车辆的运动轨迹。车辆跟踪过程经常出现车辆遮挡重叠问题,它会使检测区域出现误检而导致跟踪目标丢失。针对以上问题,本文中首先使用运动区域的宽高比和占空比作为指标进行误检判断,然后用基于对应顶点车辆分割算法对含有多个车辆的区域进行切分。更多还原

【Abstract】 Along with the rapid development of economy, the amount of motor vehicles has a rapid growth in China. As a result, the difficulty of parking has perplexed many cities in recent years. The scientific management of open parking lots is very significant to reduce traffic jam and insure traffic safety. Video surveillance is effective technology means, in which moving vehicle auto-detecting and auto-tracking is the basic part.The thesis mainly discusses the detection of vehicle based on video. It involves in the following several topics, back ground update and extraction, moving vehicle detection, moving object tracking and moving vehicles segmentation.In first part of moving object detecting and retrieving,the paper propose a new algorithm of background extraction and updating. The interval pixel value belongs to in the video stream is calculated to extract background. The paper take advantage of edge detection method to realize background updating under the case that the current background image contains stagnant vehicle by analyzing area of continuous change. Testing results demonstrate that when the method is applied,the self-adaptive updating of background can be realized with a high accuracy, meanwhile it is simple to implement and suitable for real-time detection.For real-time tracking of moving target, this thesis present a method of tracking moving vehicle based on Kalman filtering, which accomplishes object tracking by feature extraction, motion estimation and object matching. Feature extraction involves describing the shape feature of present object such as the centriod and the size of its enclosing rectangle. Motion estimation is to predict the location of the tracked vehicle by means of Kalman filter in subsequent frames and ascertain the scope for searching the tracked object. Object matching is to establish one-to-one correspondence between moving objects over frames and the track of every moving object.The moving vehicles often are sheltered by each other, it causes the false detection in target matching process. To overcome the above problem, this paper first considers the width/height ratio and occupancy ratio to make false detection judgment. Then a new moving target segmentation algorithm based on corresponding points is presented.更多还原