基于交通视频的车辆鲁棒检测和快速跟踪方法的研究

发布时间：2018-04-02 18:53

本文选题：车辆检测　切入点：暗通道　出处：《华南理工大学》2014年硕士论文

【摘要】：伴随着交通管理系统越来越自动化、智能化，智能交通管理系统已经成为研究的热点，车辆的检测和跟踪是智能交通管理系统的核心模块，，其涵盖了视频监控、计算机视觉技术、计算机图像处理、机器学习以及许多其他领域的技术。可靠又具有实时性的车辆检测和跟踪算法需要处理好背景变化、光线变化、阴影、目标遮挡等问题。关于车辆检测和跟踪算法，前人们已经提出了很多思路，但是要有效设计一个兼顾鲁棒性和实时性的车辆检测和跟踪算法依然是一项很具有挑战性的任务。基于此，本文对车辆检测和跟踪技术作了一定的探究和改进。本文在车辆检测部分，首先用混合高斯模型提取运动的目标，由于场景的复杂变化，使用混合高斯模型检测的结果不理想，阴影的存在严重影响着检测的效果，为此，本文根据阴影图的反相图具有类似半透明薄雾的性质，同时结合经典的去雾算法，提出了一种基于暗通道的阴影消除算法，该算法可以有效的消除车辆的阴影，并且对深色车辆的阴影也具有很好的鲁棒性，从而可以更加精准地提取车辆的轮廓区域。在车辆跟踪部分，本文提出了一种基于最大重叠面积和cam-shift的车辆快速跟踪算法，该算法能够有效地解决车辆粘连和分裂问题，同时还保证了车辆跟踪的正确性和实时性。最后，本文还依据车辆跟踪的结果，对一些相关的交通事件进行检测，研究了车辆检测和跟踪的实际应用价值。
[Abstract]:With the traffic management system becoming more and more automatic, intelligent, intelligent traffic management system has become a hot research, vehicle detection and tracking is the core module of the intelligent traffic management system, which covers the video surveillance.Computer vision technology, computer image processing, machine learning, and many other fields of technology.Reliable and real-time vehicle detection and tracking algorithms need to deal with background changes, light changes, shadows, target occlusion and other problems.As for vehicle detection and tracking algorithms, many ideas have been proposed, but it is still a challenging task to effectively design a vehicle detection and tracking algorithm that takes into account both robustness and real-time.Based on this, this paper makes some research and improvement on vehicle detection and tracking technology.In the part of vehicle detection, first of all, the mixed Gao Si model is used to extract the moving target. Because of the complex change of the scene, the result of the detection using the mixed Gao Si model is not ideal, and the existence of shadow seriously affects the effect of the detection.In this paper, a shadow cancellation algorithm based on dark channel is proposed according to the property that the inverse phase diagram of shadow map is similar to translucent mist, and combined with the classical de-fogging algorithm, which can effectively eliminate the shadow of vehicle.And it has good robustness to dark vehicle shadow, so it can extract the contour area of vehicle more accurately.In the part of vehicle tracking, a fast vehicle tracking algorithm based on maximum overlap area and cam-shift is proposed. The algorithm can effectively solve the problem of vehicle adhesion and splitting, and ensure the correctness and real-time of vehicle tracking.Finally, based on the results of vehicle tracking, some related traffic events are detected, and the practical application value of vehicle detection and tracking is studied.
【学位授予单位】：华南理工大学
【学位级别】：硕士
【学位授予年份】：2014
【分类号】：U495

【参考文献】