基于车辆跟踪的两种算法研究

发布时间：2018-11-03 20:52

【摘要】：车辆跟踪广泛用于军事及交通系统。车辆跟踪的关键在于能实时准确地跟踪。在有部分遮挡,车辆突然改变轨迹,光照变化的影响下,都能保证跟踪的准确性。本文主要讨论基于车牌识别的车辆跟踪和有部分遮挡的车辆跟踪。在跟踪车辆时,车牌具有唯一性,能快速确定车辆的身份。若被跟踪车辆的车牌被遮挡,此时跟踪车牌的方法失效。用特征点匹配方法跟踪目标车辆。为了满足实时跟踪的需要,提出改进的ORB(Oriented FAST and Rotated BRIEF)算法跟踪目标车辆,实验表明在有部分遮挡时能实时准确地跟踪目标车辆。基于车牌识别的车辆跟踪首先识别目标车辆的车牌。对采集的图像去噪,然后灰度化,再对其边缘检测。形态学结合边缘检测初步确定连通域,再根据车牌的固有特征精确定位车牌区域。定位到车牌后,对其二值化,然后用垂直投影法结合车牌的先验知识对其分割,将车牌分割成单个字符。对得到的字符用基于轮廓的模板匹配法将其识别出来。用卡尔曼滤波跟踪车牌从而跟踪目标车辆。在跟踪车牌的过程中,车牌面积较小,容易被其它车辆遮挡。此时,车辆庞大的外形特征可以帮助我们跟踪车辆,然而跟踪车辆时,目标车辆也易被其它车辆遮挡。为了解决部分遮挡情况下车辆实时跟踪不丢失的问题,提出了基于特征点匹配的改进的ORB算法。其具有平移,旋转,缩放不变性。改进的ORB算法在FAST检测特征点后用拉普拉斯极值去除虚假角点,相比ORB算法提高了匹配的正确率,也提高了检测速度。其中改进的FAST检测特征点速度快,BRIEF(Binary Robust Independent Elementary Features)描述子缩短了建立描述符的时间,且减少了存储空间。由此提高了特征点匹配的速度,满足实时跟踪的需要。实验表明,在有光照变化和噪声干扰的情况下,改进的ORB算法依然能够快速准确地跟踪有部分遮挡的车辆。
[Abstract]:Vehicle tracking is widely used in military and transportation systems. The key of vehicle tracking is to track the vehicle accurately and in real time. Under the influence of partial occlusion, sudden change of vehicle trajectory and light change, the tracking accuracy can be guaranteed. This paper mainly discusses vehicle tracking based on license plate recognition and vehicle tracking with partial occlusion. When tracking the vehicle, the license plate is unique and can quickly determine the identity of the vehicle. If the license plate of the tracked vehicle is blocked, the method of tracking the vehicle license plate is invalid. The feature point matching method is used to track the target vehicle. In order to meet the need of real-time tracking, an improved ORB (Oriented FAST and Rotated BRIEF) algorithm is proposed to track the target vehicles. The experimental results show that the target vehicles can be tracked in real time and accurately when there is partial occlusion. Vehicle tracking based on license plate recognition first recognizes the vehicle license plate of the target vehicle. The image is de-noised, then grayscale, and then the edge is detected. The connected region is preliminarily determined by morphology combined with edge detection, and then the license plate area is accurately located according to the inherent characteristics of the license plate. After locating the license plate, the license plate is binary, then the license plate is segmented into a single character by the vertical projection method combined with the prior knowledge of the license plate. The resulting characters are recognized by contour-based template matching. The target vehicle is tracked by Kalman filter. In the process of tracking the license plate, the license plate area is small, easy to be blocked by other vehicles. At this point, the huge shape of the vehicle can help us track the vehicle, but when tracking the vehicle, the target vehicle is easily blocked by other vehicles. An improved ORB algorithm based on feature point matching is proposed to solve the problem of vehicle real-time tracking without losing in partial occlusion. It has translation, rotation, zoom invariance. The improved ORB algorithm uses Laplace extremum to remove false corner points after FAST detection. Compared with ORB algorithm, it improves the accuracy of matching and the speed of detection. The improved FAST detection feature point fast, BRIEF (Binary Robust Independent Elementary Features) descriptor shortens the time of establishing descriptor and reduces the storage space. Therefore, the speed of feature point matching is improved and the need of real-time tracking is satisfied. The experimental results show that the improved ORB algorithm can track partially occluded vehicles quickly and accurately in the presence of illumination changes and noise interference.
【学位授予单位】：辽宁科技大学
【学位级别】：硕士
【学位授予年份】：2015
【分类号】：U495

【相似文献】