【摘要】 针对车辆违章行为中的违章变道行驶,给出一种基于视频图像处理的违章检测算法,该算法包括车辆检测、车辆跟踪和违章行为判别3个处理步骤。车辆检测采用基于支持向量机(SVM)和滑动窗口搜索的方法,以检测结果作为跟踪起点,跟踪算法在后续帧对车辆的位置进行预测。提出了一种新的车辆跟踪算法,该算法包括2个处理步骤,首先利用卡尔曼滤波器估计车辆的位置,然后依据检测环节中训练得到的SVM分类器模型,在估计位置的周围进行匹配,实现对车辆的精确跟踪。通过对车辆的跟踪,可以时刻获得车辆的位置,根据车辆的行驶状态与初始所属车道是否一致来判断是否有违章行为发生。试验结果表明,车辆违章检测算法的检测率高,误检率低,并且能够进行实时处理。更多还原

【Abstract】 Lane violation is one type of vehicle violation behaviors.A violation detection algorithm based on video image processing was given for lane violation detection.The algorithm is composed of three steps which are vehicle detection,vehicle tracking and violation judgment.Vehicle detection algorithm is based on support vector machine(SVM)and sliding-window search.Using the detection result as the starting point of tracking,vehicle tracking algorithm makes a prediction of vehicle’s location in the following frames.A new vehicle tracking algorithm which consists of two steps was proposed.First,it uses Kalman filter to estimate vehicle’s location.Second,based on the SVM classifier model trained in the detection process,it makes a matching around vehicle’s location estimated in the first step to realize accurate tracking.According to the result of vehicle tracking,vehicle’s location is available at any time.A judgment that whether lane violation has happened or can not be made based on comparison of vehicle’s current running state with its initial lane.The experimental result indicates that the detection rate of the vehicle violation detection algorithm is high and the false detection rate is low,besides,the total processing can be finished in real time.更多还原