基于降采样聚类的双目图像测距算法研究

Binocular Image Ranging Algorithm Baseol on Downsampling Clustering

提出了一种针对道路交通图像等大尺寸场景图像的测距算法,主要解决道路图像中存在的光照变化频繁,场景空间尺寸较大导致处理时间较长的问题,以在保证精确度的同时获得比普通双目立体匹配算法更高的速度.创新的使用了一种降采样Census算法来求出初始匹配代价,然后通过聚类算法将局部匹配代价聚合进行视差图优化,获得尺度较小的视差图,以便于后续进行障碍物检测目标识别等相关操作.实验表明,在道路交通场景,该算法能够在保证准确率的基础上极大的提高算法效率.

This paper proposes a ranging algorithm for large-scale scene images such as road traffic images,which mainly solves the problem that the illumination changes frequently in the road image and the large size of the scene space leads to a long processing time,so as to ensure accuracy. Get a higher speed than the normal binocular stereo matching algorithm. In this paper,a downsampling Census algorithm is used to find the initial matching cost,and then the local matching cost is aggregated by the clustering algorithm to optimize the disparity map to obtain the disparity map with smaller scale,so as to facilitate the subsequent obstacle detection target recognition,and other related operations. Experiments show that in the road traffic scenario, the algorithm can greatly improve the efficiency of the algorithm on the basis of ensuring the accuracy.