基于目标检测的城市路口车辆加塞的行为识别研究

Research on Recognition of Vehicle Cut in Lion Behavior at Urban Intersections Based on Object Detection

针对城市交通路口常见的车辆加塞行为,建立一个基于目标检测算法的车辆加塞行为识别模型。首先对城市路口图像进行目标检测,并对检测结果进行目标类别筛选,保留加塞行为的目标类别;其次对目标中心点进行Hough变换,在图像中拟合多条直线;然后对Hough变换的结果进行非极大值抑制、非车道方向抑制等后处理,得到输入图像中车道方向直线;最后通过计算目标中心点与车道方向直线的距离判定目标是否具有加塞行为。本文采用Hough变换对基于深度学习的目标检测结果进行处理,通过拟合图像中车道方向直线,判断车辆与车道方向直线的距离,可有效检测路口车辆的加塞行为。

Aiming at the common vehicle cut in line behavior at urban traffic intersections, a vehicle cut in line behavior recognition model based on the object detection is established. First, perform object detection on the urban intersections image, and perform object category screening on the detection results to retain the traffic categories which may cut in line behavior. Secondly, perform Hough transform on the traffic object center points to fit multiple straight lines in the image. Then perform Non-Maximum Suppression, Non-Lane direction Suppression and other post processing on Hough transform result to obtain the lane direction lines in the image. Finally, by calculating the distance between the object center point and the lane direction line to determine whether the object has cut in line behavior. In this paper, Hough transform is used to process the object detection results based on deep learning. By fitting the lane direction lines in the urban intersections image, judging the distance between the vehicle and the lane direction line, it can effectively detect the traffic cut in line behavior of the intersection.