基于视频角点特征匹配的车速检测方法  

Vehicle Speed Detection Method Based on Video Corner Feature Matching

作　　者：支晨蛟[1] 唐慧明[1] 

机构地区：[1]浙江大学信息与通信工程研究所,杭州310027

出　　处：《计算机工程》2013年第12期176-180,共5页Computer Engineering

基　　金：国家科技重大专项基金资助项目(2010ZX03004-003-01);中央高校基本科研业务费专项资金资助项目(2012FZA5008)

摘　　要：针对传统特征匹配车速检测方法实时性较差的问题,提出一种改进的角点特征匹配车速检测方法。基于视频图像,采用混合高斯模型检测方法提取运动车辆目标,利用Harris算法检测车辆目标的角点特征,将运动估计和NCC匹配相结合,优化匹配区域搜索方法,对车辆目标角点进行角点粗匹配,再通过RANSAC算法进行角点精匹配和单视测量坐标转换以实现车速检测。实验结果表明,与传统方法相比,该方法的角点粗匹配速度提高400%,角点精匹配速度提高200%,车速准确性达到90%以上,能有效提高车速检测的实时性和准确性,满足实际车速检测的要求。In order to improve the real-time performance of traditional feature matching vehicle speed detection, this paper presents an improving Harris comer matching method. Based on video, it uses moving object detection and Harris comer detection, combines motion estimation and NCC template matching, optimizes matching area search strategy for comer coarse matching, then uses random sample consensus for comer fine matching and single view metrology coordinate transformation, and finally achieves vehicle speed measurement. Experimental results show that, compared with the traditional method, comer coarse matching speed is increased by four hundred percent, comer fine matching speed is increased by two hundred percent, the vehicle speed accuracy reaches more than ninety percent, and the new method successfully improved the real-time of the algorithm and the accuracy of comer matching.

关 键 词：车速检测 运动目标检测 角点检测 特征匹配 运动估计 单视测量 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]