基于透视点和区块特征的多车粘连分割方法

针对交通视频车辆检测与跟踪中经常出现的车辆前景粘连问题,提出了一种利用透视点在图像轮廓上搜索车辆分割点并通过区块特征识别车辆的粘连车辆分割方法.根据路面场景中车道线信息提取出透视点和车道区域,结合混合高斯模型与形态学梯度轮廓算法提取出车辆前景团块.利用透视点原理从前往后顺序分割粘连的车辆前景团块.对分割开的待检定区域,利用车辆区块特征进行检验识别,修正错误分割,将粘连的多辆车逐一分割.实验结果表明,该方法对直线道路上多辆车粘连的分割有较好的准确度和适应性.

Occluded vehicles segmentation based on perspective point and feature blocks

A new method is proposed based on the perspective point and characteristic blocks to segment the occluded vehicles on the freeway to resolve the multi-vehicle occlusion appearing in vehicle detecting and tracking.Perspective points and road regions are detected according to lane line in the highway scene.Combining the Gaussian mixture model(GMM) and the morphological gradient,the vehicle foreground is obtained.With the perspective point and contours of foreground blobs,occlude vehicles are detected and segmented in sequence from front to back.The feature blocks of vehicles are used to identify and amend the incorrect segments,finally the occluded vehicles are segmented one by one.Experiments on various typical highway scenes show the effectiveness of the proposed method in real time tracking.