基于背景差分和均值漂移的闯红灯车辆视频自动检测系统

AN AUTOMATIC DETECTION SYSTEM FOR RED LIGHT RUNNING VEHICLES BASED ON BACKGROUND SUBTRACTION AND MEAN-SHIFT

在视频序列中使用均值漂移算法进行运动目标跟踪是一种可行的办法,但这种方法需要手动指定初始跟踪目标,无法做到全自动跟踪。在十字路口这一静态背景下,通过缩小感兴趣区域,使用背景差分的方法,结合图像形态学处理,完成对运动车辆目标的自动获取,从而实现闯红灯目标的抓取。自动检测系统和人工检测的对比实验表明,该系统的正确检出率可达到95.4%以上。

Using mean-shift algorithm to track moving target in video series is a feasible way. However, the method needs to specify initial target manually, and is not an entirely automatic tracking approach. In static background of crossroads, by zooming out the interested region, using the method of background subtraction and in combination with image morphological processing, the automatic capture of moving vehicle targets is implemented, thus the snatch of red light running object is achieved as well. Comparative experiment on the automatic detection system and the manual inspection shows that the correct detection rate for this system reaches up to 95.4%.