Abstract: A multi-lanes detection method is proposed by extracting the edge of the background image automatically for intelligent applications of traffic surveillance video processing, which requires stationary scene and lanes information in advance. It focuses on static image processing: after background extraction from traffic video, filters and mathematical morphology are used for pretreatment. Hough transformation locates the lanes by gaining its skull from the edge of the background. With geometrical restriction, improper lines are excluded from lanes, and road area is confirmed. Several scence tests have been done to insure the method is effective. The proposed method turns out to be practically valuable, and has the robustness to detect multi-lanes from different scenes of traffic surveillance video under variant illumination conditions.

Key words: traffic surveillance video, multi-lanes detection, lane snatch, skull

摘要： 为自动有效地获取交通监控场景中的多车道信息，提出一种利用骨架化边缘的多车道检测算法，以克服视频处理对固定场景和明确的先验车道位置信息的依赖。算法主要针对静态的交通背景图处理，采用背景提取、滤波和数字形态学预处理等，由Hough变换确定车道位置的骨架线；由行车方向约束车道线角度，利用车道线几何成像特性检测出准车道线，获取车道线和车道区域。实验表明，对不同的交通场景和不同光照条件，该方法能有效检测多车道，鲁棒性强，具有较高的工程应用价值。

关键词: 交通监控视频, 多车道检测, 车道线提取, 骨架化