一种基于彩色路面区域分割的候选车辆区域视频检测方法  

A VEHICLE-CANDIDATES VIDEO DETECTION METHOD BASED ON SEGMENTATION OF COLOR ROAD AREA

作　　者：陈兆学[1] 郭鸣[2] 施鹏飞[3] 

机构地区：[1]上海理工大学医疗器械与食品学院,上海200093 [2]浙江大学城市学院计算分院,浙江杭州310015 [3]上海交通大学图像处理与模式识别研究所,上海200030

出　　处：《计算机应用与软件》2007年第10期31-33,36,共4页Computer Applications and Software

基　　金：国家"973"重点基础研究项目(G1998030408);上海市教委选拔培养优秀青年教师科研专项基金(358536)。

摘　　要：提出了一种基于彩色路面区域分割的候选车辆视频检测方法。该方法首先根据路面部分颜色的分布特点从单帧彩色图像中分割出路面部分并完成路面区域背景的初始化,然后通过对路面区域运用背景差方法和相关后处理过程进行动静态候选车辆的检测和分割。由于采用一种双背景策略能够在车辆检测的同时完成背景的训练和替换更新,该算法克服了传统背景差算法背景更新时容易存在误差累积以及对环境光线变化敏感的缺点,实现简单、稳健性好,可以满足交通视频监控系统中背景更新和车辆检测的实时性处理要求,实验结果证明了该方法的有效性。A vehicle-candldates video detection method based on segmentation of color road area is proposed. The road area is segmented at first from a single colorful video frame based on the feature of road area colors distribution. Either dynamic or static vehicle-candidates could be detected by background difference method and the related post-process course after the background initialization of road area. The background is trained and updated simultaneously just at the moment of vehicle detection by adoption of a double-background strategy, which overcomes the problems of error accumulation and sensitivity limitation over light conditions that occur during background updating of traditional methods. With simple and robust implementation, this method could satisfy the real-time requirements of background updating and vehicle detecting in traffic video surveillance system. Its validity is proved by the experimental result.

关 键 词：车辆检测 背景差 图像分割 HSI颜色模型 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术]