基于立体视觉与纹理特征相结合的障碍物探测

Obstacle Detection Based on Combined Stereo Vision and Texture Feature

针对智能汽车行驶过程中障碍物检测问题,提出一种使用双目立体视觉的距离信息与障碍物纹理信息相结合的障碍物检测方法。首先,将双目相机的左图和右图校正后进行匹配得到视差图,根据相机内部参数和外部参数将视差图投影为具有距离信息的鸟瞰图,去除地面及高空背景目标,在鸟瞰图上将相近的点云聚类为一个目标,得到障碍物的粗略位置信息;然后,将得到的位置信息映射到原图,并形成多个候选区域,在候选区域查找障碍物边缘信息,得到障碍物具体位置。该算法结合了物体距离的三维信息和物体形状的二维信息,使障碍物的分割结果更加精准。针对典型交通场景进行了实验,结果表明,该算法能够有效弥补单独使用距离信息或者纹理信息带来的障碍物分割不准确的问题。

To aim at detecting obstacles in the process of intelligent vehicle driving,a method based on the distance information and the obstacle texture information is proposed.First,the left camera and right camera are corrected to obtain a parallaximage,and the parallaximage is projected into a bird＇s-eye view with distance information according to the camera＇s internal and external parameters.In the bird＇s eye view,the target information is mapped to the original image so that a plurality of candidate regions are formed,and the edge information of the obstacle is searched in the candidate region to obtain the specific position of the obstacle.The algorithm proposed in this paper combines the three-dimensional information of the object distance and the two-dimensional information of the object shape,therefore the segmentation result of the obstacle is more accurate.The experiment is carried out for the typical traffic scene.The result shows that the algorithm can effectively improve the segmentation inaccuracy caused by the application of distance information or texture information.