交通环境下的车辆检测与跟踪算法研究与实现

发布时间：2018-05-01 11:24

本文选题：前景检测 + 光流检测　；参考：《电子科技大学》2015年硕士论文

【摘要】：交通环境下的车辆检测与跟踪是智能交通系统中的两个重要的组成部分,它们是实现智能交通系统的关键所在,车辆检测与跟踪的好坏决定了整个智能交通系统的好坏。传统的车辆检测一般存在检测速度不够快和检测结果不够准确等问题,而车辆的跟踪却很难平衡它的实时性与准确性。针对以上的问题,本文对交通环境下的车辆检测与跟踪做了各个方面的优化与调整,最终得到了一个检测结果好,跟踪鲁棒性强的车辆检测与跟踪系统,并能够在一般情况下实时处理的系统。本文主要从以下几个方面来优化交通环境下的车辆检测与跟踪系统。1、本文在车辆检测方面采用了基于haar特征的adaboost检测分类器,在分类器的训练方面选取了特定的haar特征进行训练。为了提高整个系统的运行时间和检测准确性,本文首先对视频中图像序列进行了运动目标建模,提取出运动团块,并只对运动团块所在区域进行检测,从而检测出运动车辆。其次,本文把视频中的车辆分为八个运动方向,分别对每一个方向进行训练,从而得到八个方向的检测分类器,在选择检测分类器时根据运动团块的光流方向来选择对应的分类器。2、本文在车辆跟踪方面,采用了基于检测的粒子滤波跟踪,并针对粒子滤波跟踪中存在的问题采用以下几个方面来进行了改进。首先采用了数学中的线性拟合来处理目标遮挡问题;其次充分利用检测结果来处理跟踪目标跟踪不准确问题;最后运用了多线程来提升整个跟踪系统的跟踪速度。对于整个的车辆检测与跟踪系统,本文从理论分析到实践操作,最后形成了一个车辆检测与跟踪系统。并通过实验测试了整个系统的各个模块的性能。对于整个智能交通系统做出了自己的贡献。
[Abstract]:Vehicle detection and tracking under traffic environment are two important parts of intelligent transportation system. They are the key to realize the intelligent traffic system. The quality of the whole intelligent traffic system is determined by the quality of the vehicle detection and tracking. The traditional vehicle detection generally exists the lack of fast detection speed and the inaccuracy of the detection results. It is difficult to balance the real time and accuracy of vehicle tracking. In view of the above problems, this paper optimizes and adjusts the vehicle detection and tracking under the traffic environment, and finally obtains a good detection and tracking system with robust detection results and strong robustness, and can be used in real time in real time. This paper mainly optimizes the vehicle detection and tracking system.1 under the traffic environment from the following aspects. In this paper, the AdaBoost classifier based on the Haar feature is used in the vehicle detection, and the specific Haar features are selected for training in the training of the classifier. In this paper, we first model the motion target of the image sequence in the video, extract the movement block, and detect only the area where the movement is located, and then detect the moving vehicle. Secondly, this paper divides the vehicles in the video into eight directions, and then trains each direction to get eight directions. In the selection of classifier, the corresponding classifier.2 is selected according to the direction of the optical flow of the movement mass. In this paper, the particle filter tracking based on detection is adopted in the vehicle tracking, and the following aspects are used to improve the problems in the particle filter tracking. First, the linear simulation in mathematics is adopted. In order to deal with the problem of target occlusion, then the detection results are used to deal with the inaccuracy of tracking target tracking. Finally, multithreading is used to improve the tracking speed of the whole tracking system. For the whole vehicle detection and tracking system, a vehicle detection and tracking system is formed in this paper from theoretical analysis to practice. The performance of each module of the whole system is tested through experiments, and has made its own contribution to the whole intelligent transportation system.

【学位授予单位】：电子科技大学
【学位级别】：硕士
【学位授予年份】：2015
【分类号】：U495;TP391.41

【共引文献】