基于单应性约束协同追踪技术研究

发布时间：2018-01-12 14:23

本文关键词：基于单应性约束协同追踪技术研究　出处：《湖北工业大学》2017年硕士论文　论文类型：学位论文

【摘要】：随着社会的发展,社会安全成为人们日益关注的话题。基于机器视觉的监控追踪逐步在公共领域获得广泛应用。由于信息量的爆炸式增长,单靠人力现场监控需要成本很高,所以视频监控系统要能够实现主动协助监控,减少人力劳动。传统的单摄像机目标追踪精度低。在一些复杂的场景下,单摄像机目标追踪很难实现追踪要求。多摄像机比单摄像机获得的信息量多。通过单应性约束实现多视角的信息融合,多摄像机目标追踪可以克服一些单摄像机不能解决的问题。在目标检测模块采用了新的图像背景更新算法。这是一种结合帧间差分法、背景差分法以及图像分块分类新的目标检测算法,并且表现性能良好。在目标定位方面,本文采用单应性矩阵获取目标足部区域的定位。通过可容忍误差的单应性约束在视角平面画出矩形框,估计视角平面上的足部区域。接着统计足部估计区域的灰度直方图,通过比较灰度直方图的相似性来进行足部斑点分类,用这种方法进行足部斑点匹配具有较高的准确性。在目标追踪方面,本文采用足部斑点质心间欧氏距离比较算法,提高了跟踪实时性和准确性。通过仿真实验,本文在目标检测方面比传统方法获得的目标更加完整。目标定位和追踪结果都可靠。整体算法的实时性能更佳。
[Abstract]:With the development of society, people pay more and more attention to social security. Monitoring and tracking based on machine vision has gradually been widely used in the public domain. Because of the explosive growth of the amount of information. It is very expensive to rely solely on manpower on-site monitoring, so video surveillance system should be able to actively assist monitoring, reduce labor. Traditional single-camera target tracking accuracy is low. In some complex scenarios. Single camera target tracking is difficult to meet the requirements of tracking. The amount of information obtained by multi-camera is more than that of single camera. The information fusion of multi-view is realized through monoclinic constraints. Multi-camera target tracking can overcome some problems that can not be solved by single camera. A new image background updating algorithm is used in the target detection module. The background difference method and the new target detection algorithm based on image block classification have good performance and can be used in target location. In this paper, the location of the foot region of the target is obtained by using the homotropic matrix, and the rectangular frame is drawn in the view angle plane by the monotropic constraint of tolerance error. Then the grayscale histogram of the estimated area of foot is counted, and the classification of foot spots is carried out by comparing the similarity of gray histogram. This method has high accuracy in foot spot matching. In terms of target tracking, this paper uses Euclidean distance comparison algorithm to improve the real-time and accuracy of tracking. In this paper, the target detection is more complete than the traditional method. The results of target location and tracking are reliable, and the real-time performance of the whole algorithm is better.
【学位授予单位】：湖北工业大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：TP391.41

【参考文献】