基于双目视觉的障碍物检测算法

Algorithm of Obstacle Detection Based on Binocular Vision

针对在前方障碍物占图像宽度比例相对较大的情况下,利用V-视差提取道路再检测障碍物的各类方法误检测率较高的问题,提出一种障碍物检测新算法。首先计算U-视差并采用双阈值算法将道路和障碍物分类,然后结合原视差图移除道路部分后形成新视差图,再计算V-视差和U-视差并分别提取障碍物相关线,最终实现障碍物的检测。实验结果表明,该方法检测效果良好,特别在前述情况下,较同类方法误检测率低。

A new detection method for obstacle is presented because of the lower wrong detection rate by using V - disparity to extract obstacles to be detected on the road. Firstly, U -disparity is calculated and double thresholds algorithm is used to classify the roads and obstacles. And then, the new disparity map is formed by combining with the original disparity map which is removed the parts of the roads. Finally, the V -disparity and U - dispari-ty are calculated and the correlation lines of obstacles are extracted for obstacles detection. The experimental results shows that the method has a better detection effect and lower false positive rate especially in the mentioned cases.