基于视频监控的目标遮挡问题研究

发布时间：2019-05-23 11:02

【摘要】：基于视频监控的目标跟踪技术一直是计算机视觉领域的热点研究问题,在军事制导、安防建设和智能交通领域有着广阔的应用前景。智能视频监控系统包括:视频采集模块、图像预处理模块、目标检测模块、目标跟踪模块和智能分析模块,涉及到模式识别、视觉分析等多个领域。在视频监控系统中,由于背景的复杂多变,目标在运动过程中经常会出现部分或全部被遮挡的情况。本文针对目标跟踪过程中普遍存在的遮挡后目标易跟错、跟丢的问题展开研究,提出了基于“物体恒存性”的方法解决长时间遮挡后目标容易跟错和丢失的情况,有效的解决了单目标和多目标跟踪过程中长时间遮挡问题。主要研究内容包括:(1)在运动目标检测方面,针对视频监控中的背景不稳定的情况,本文采用混合高斯背景建模方法提取前景运动目标,利用开、闭运算等方法对二值化的前景目标进行连通性处理,去除噪声点,填补细小的孔洞,获得较为完整的目标图像。(2)在运动目标跟踪方面,提取运动目标的多种类型特征,主要包括颜色、边缘、纹理、直方图等,并与目标运动轨迹特征相结合,重点解决目标跟踪过程中被短时间遮挡后,单一特征消失,目标容易跟错和跟丢的问题。针对长时间遮挡情况,本文提出了基于“物体恒存性”算法,分析遮挡关系,利用目标未遮挡时提取的多种类型的特征,在遮挡物的附近快速的查找消失目标,提高查找的效率。(3)对本文提出的“物体恒存性”算法,利用“Weizmann”、“KTH”、“CAVIAR”标准视频库以及自录的视频序列进行实验验证。针对跟踪过程中可能出现的光照突变、短时间遮挡、长时间遮挡以及循环遮挡等情况,给出了实验结果和分析,表明该算法能够有效地解决遮挡问题,达到很好的识别跟踪效果。
[Abstract]:Target tracking technology based on video surveillance has always been a hot research issue in the field of computer vision, and has a broad application prospect in the fields of military guidance, security construction and intelligent transportation. Intelligent video surveillance system includes: video capture module, image preprocessing module, target detection module, target tracking module and intelligent analysis module, involving pattern recognition, visual analysis and other fields. In video surveillance system, due to the complexity and variability of the background, the target is often partially or completely blocked in the process of motion. In this paper, the problem that the target is easy to follow and lose after occlusion is studied in the process of target tracking, and a method based on "object persistence" is proposed to solve the problem that the target is easy to follow and lose after long time occlusion. It effectively solves the problem of long time occlusion in the process of single target and multi-target tracking. The main research contents are as follows: (1) in the aspect of moving target detection, in view of the unstable background in video surveillance, this paper uses the hybrid Gao Si background modeling method to extract the foreground moving target, using the open, Closed operation and other methods are used to deal with binarization foreground targets, remove noise points, fill small holes, and obtain more complete target images. (2) in moving target tracking, various types of features of moving targets are extracted. It mainly includes color, edge, texture, histogram and so on, and combines with the moving trajectory feature of the target, which focuses on solving the problem that the single feature disappears after a short time occlusion in the process of target tracking, and the target is easy to follow up and lose. In order to solve the problem of long time occlusion, this paper proposes an algorithm based on "object persistence", which analyzes the occlusion relationship and makes use of various types of features extracted by the target when it is not occlusive to quickly find the disappeared target near the occlusive object. (3) the "object persistence" algorithm proposed in this paper is verified by "Weizmann", "KTH", "CAVIAR" standard video libraries and self-recorded video sequences. In view of the possible light mutation, short time occlusion, long time occlusion and cyclic occlusion in the tracking process, the experimental results and analysis show that the algorithm can effectively solve the occlusion problem. Achieve a good recognition and tracking effect.
【学位授予单位】：河北工业大学
【学位级别】：硕士
【学位授予年份】：2015
【分类号】：TN948.6;TP391.41

【参考文献】