摘要

自动车型识别（AVC）对解决车辆收费、道路监控、提高公路利用效率具有重要意义。传统的利用环形感应线圈或压电传感器等识别车辆类型，安装不方便，器件容易损坏，获取的交通信息比较少，因此未能广泛应用。而利用图像处理和模式识别技术识别车型，易于安装，可以长时间稳定工作，获取的信息丰富，适用范围广泛，是自动车型识别的发展方向。
采用图像处理和模式识别的方法识别车型，一般分为运动分割、多目标跟踪和车型判别三个步骤。本文首先从一段图像序列中恢复了不包含任何车辆的背景图像，利用背景和当前图像进行差异检测，从而分割出车辆目标，计算量较小。在分割的基础上，为了正确统计车辆数量，计算车辆速度，必须进行目标跟踪。本文利用运动的连续性，采用面积、形心、速度等区域特征进行区域跟踪，计算量小，可以在一定程度上解决因遮挡带来的分裂合并问题。在车型判别阶段，单摄像头摄像丢失了深度信息，很难从图像中获取车辆的特征参数，本文利用车辆边缘和车辆模型相匹配的方法识别车型，匹配的好坏程度用Hausdorff距离度量，这种方法对噪声和形变适应性较好。三维世界向二维图像平面投影的过程中，还需要定标摄像机，定标模型采用了方法简单、应用广泛的针孔模型。利用定标结果，可以在跟踪过程中计算车辆速度。算法在PC上实现之后，被移植到Trimedia TM1300 DSP上，进行了测试和优化。 计算机毕业论文----德维在线 www.devay.net


关键词：车型识别 运动分割 多目标跟踪 摄像机定标 模型匹配

ABSTRACT
AVC(automatic vehicle classification), the object of this paper, plays an important role in road design、traffic surveillance and highway charge. Because of inadequate information obtained、difficulty of mount and lowness of dependability, traditional vehicle detection and classification with induction loop or piezoelectricity sensor can not be widely used. On the contrary, AVC based image processing and pattern recognition, settled the above-mentioned deficiency, can be widely used in auto charge、park lot management and highway surveillance and become the direction of AVC development.
General speaking, AVC based image processing and pattern recognition can be decomposed into three steps such as moving segmentation、object tracking and vehicle classification. In this paper, background without any vehicle can be reconstructed from video sequences and object can be detected only that we subtract background from current image. Followed segmentation, multi-object tracking is necessary in order to count vehicles correctly and acquire vehicle speed. We performed object region tracking based on its characters such as area、location and velocity. Using this method, region merger and split can be resolved in some degree. In vehicle classification period, because deep information is absent, it’s a very difficult task to obtain vehicle characters from images. Vehicle recognition based model match method is adopted and we can match segmentation edges and model edges with Hausdorff distance. This method is not sensitive to deformation and noise. In order to get vehicle speed and perform model matching, perspective matrix is calculated based on pin-hole model. All above algorithm is realized and optimized on Trimedia DSP 1300. 计算机毕业论文----德维在线 www.devay.net

Key words: vehicle classification, moving segmentation, multi-object tracking, camera calibration, model match