基于目标和运动信息的Mean-Shift算法在视觉车辆跟踪中的应用

发布时间：2018-12-11 20:47

【摘要】：基于计算机视觉的车辆跟踪多年来一直是热门的研究课题之一，它是智能交通的基础。视觉车辆跟踪技术涉及到诸多学科，它不仅与图像处理技术、计算机视觉息息相关，还与人工智能和模式识别等紧密联系。虽然目前有目标跟踪算法在车辆跟踪上的应用，但是目标车辆本身特征的可变性、背景的干扰、跟踪过程中目标的遮挡问题以及车辆的快速运动都成为了影响跟踪准确性的因素。因此，研究一种准确度高、鲁棒性强的车辆跟踪算法依然是目前迫切需要解决的问题。本文首先介绍了车辆跟踪在国内外的研究现状，接着研究了传统Mean-Shift算法在车辆跟踪中的应用。接着本文针对车辆跟踪中目标尺度变化、背景干扰、遮挡及目标快速运动等问题，在基于颜色特征的Mean-Shift算法基础上，结合目标信息和运动估计成功实现了车辆跟踪。由于目标车辆在移动的过程中尺度可能发生变化，或者被其它干扰物遮挡，这就使目标模型与候选模型之间的相似性系数降低，导致Mean-Shift算法陷入局部最优，从而造成定位失败。在本文中，在Mean-Shift算法基础上，结合了目标的信息，提高了Mean-Shift算法对目标尺度变化的适应性并优化了模型；当目标被严重遮挡时，结合运动估计，利用卡尔曼滤波预测目标的位置，从而弥补了Mean-Shift算法在处理遮挡问题时的不足。此外，，本文还针对Mean-Shift在跟踪快速移动的目标车辆容易陷入局部最优的问题，利用卡尔曼滤波器优化后的初始中心克服了基本Mean-Shift算法用泰勒级数估计当前帧初始窗口精度不高的缺陷。最后实验结果表明，改进的Mean-Shift算法能准确的对目标进行跟踪。
[Abstract]:Vehicle tracking based on computer vision has been one of the hot research topics for many years. It is the basis of intelligent transportation. Visual vehicle tracking involves many disciplines. It is not only closely related to image processing and computer vision, but also closely related to artificial intelligence and pattern recognition. Although there are some applications of the target tracking algorithm in vehicle tracking, the variability of the target vehicle's own characteristics and the interference of the background, The occlusion of the target and the rapid movement of the vehicle are the factors that affect the tracking accuracy. Therefore, it is still an urgent problem to study a vehicle tracking algorithm with high accuracy and robustness. This paper first introduces the research status of vehicle tracking at home and abroad, and then studies the application of traditional Mean-Shift algorithm in vehicle tracking. Then, aiming at the problems of target scale change, background interference, occlusion and fast moving of target in vehicle tracking, this paper successfully realizes vehicle tracking based on color feature based Mean-Shift algorithm, combined with target information and motion estimation. Because the scale of the target vehicle may change in the course of moving, or be blocked by other interference, the similarity coefficient between the target model and the candidate model will be reduced, and the Mean-Shift algorithm will fall into the local optimum. As a result, the location fails. In this paper, based on the Mean-Shift algorithm, combining the information of the target, the adaptability of the Mean-Shift algorithm to the change of the target scale is improved and the model is optimized. When the target is heavily occluded, the Kalman filter is used to predict the location of the target in combination with motion estimation, which makes up for the deficiency of the Mean-Shift algorithm in dealing with the occlusion problem. In addition, this paper also aims at the problem that Mean-Shift is prone to fall into local optimum in tracking fast moving target vehicles. Using the Kalman filter to optimize the initial center overcomes the defects of the basic Mean-Shift algorithm which uses Taylor series to estimate the current frame initial window with low accuracy. Finally, the experimental results show that the improved Mean-Shift algorithm can track the target accurately.
【学位授予单位】：重庆大学
【学位级别】：硕士
【学位授予年份】：2014
【分类号】：U495;TP391.41

【参考文献】