论文部分内容阅读
文章针对交通治安卡口应用,提出了一种基于视频的全天候智能车辆检测方法.该方法主要特点是不考虑车辆跟踪,提高了计算速度并避免了跟踪误差,适于嵌入到监控摄像机中;而且自动区分白天/夜间场景.对于白天场景,通过背景剪除、阴影抑制、形态学计算等手段获得运动信息,然后根据车辆的尺寸、对比度与纹理特征实现车辆检测;对于夜间场景,利用车灯的高亮度与对称性特征得到车辆检测结果.该方法快速有效,在现场采集的实际路况视频数据上,白天与夜间车辆检测准确率分别为96.42%和95.96%.“,”This paper proposes a video-based 24-hour vehicle detection method for traffic security-access monitoring.The method is suitable for being embedded into surveillance cameras,and since vehicle tracking is not considered,the efficiency of vehicle detection is improved and tracking errors are avoided.Two vehicle detection strategies for daytime and nighttime are switched between each other automatically.Under day illumination,the background subtraction,shadow suppression,morphological operator and connected component computation are performed sequentially to extract moving objects in the video frame.Then the vehicles passing the traffic security-access are detected according to size,luminance contrast and texture features of vehicles.Under night illumination,the high-luminance and pairing characteristics of headlights in the vehicle are explored to detect the vehicles.The proposed method is efficient and effective,which achieves detection rates of 96.42% and 95.96% on realistic traffic videos for daytime and nighttime,respectively.