结合跟踪算法的交通标志检测与识别

Traffic sign detection and recogniton combined with tracking algorithm

针对基于视频的交通标志检测与识别算法,提高其实时性、降低其误检率以及改善目标检测算法在检测过程中遇到掉帧或者目标物体被遮挡情况下表现不佳。由于视频多帧的冗余信息,对连续10帧进行检测以及识别,之后对当前结果进行投票确定当前的类别,当确定类别后,停止继续检测与识别,使用DeepSORT对当前目标进行跟踪。有效提高了识别准确率、提高识别的实时性,并且对遮挡以及掉帧情况有很好的效果。通过实验可知,方法有一定的实用性,整体的性能优于单纯使用检测与识别算法。

For video-based traffic sign detection and recognition algorithms, improve their real-time performance, reduce their false detection rate. Improve the shortcomings of target detection algorithm when frame loss or target object is occluded. Due to the redundant information of the video multi-frame, the continuous 10 frames are detected and identified, and then the current result is voted to determine the current category. When the category is determined, the recognition algorithm is stopped and the current target is tracked using Deep SORT. The algorithm effectively improves the recognition accuracy, improves the real-time performance of recognition, and has a good effect on occlusion and frame dropping. Through experiments, the method has certain practicability, and the overall performance is better than the simple detection and recognition algorithm.