Computer Engineering and Applications ›› 2018, Vol. 54 ›› Issue (6): 193-197.DOI: 10.3778/j.issn.1002-8331.1708-0313

Previous Articles     Next Articles

Fast binocular stereo matching algorithm based on strong similarity detection

MA Long, SUN Mingze, HUANG Chao, PEI Xin, ZHOU Hang, ZHANG Hongyan   

  1. Sino-European Institute of Aviation Engineering, Civil Aviation University of China, Tianjin 300300, China
  • Online:2018-03-15 Published:2018-04-03


马  龙,孙铭泽,黄  超,裴  昕,周  航,张鸿燕   

  1. 中国民航大学 中欧航空工程师学院,天津 300300

Abstract: Stereo matching is an important branch of binocular vision, which can reduce the three-dimensional information through the depth map, but because of its huge amount of computation, real-time is difficult to be guaranteed. In this paper, a fast stereo matching algorithm based on strong similarity is proposed. Firstly, the binocular image is processed by the pole, so that the matching region is fixed on the same horizontal line, and the matching region is reduced. Secondly, the image is gray, and the points in the search range which are close to the gray value of the matching points are defined as strong similarities. The matching cost of the strong similarity points is calculated and the optimal parallax is obtained. Point for normal parallax calculation, the final parallax correction and filtering, get the final parallax. The results show that the proposed method can improve the accuracy of SAD by 70% without compromising the accuracy rate, which provides a good foundation for the practical application of stereo matching algorithm. In the visual navigation, obstacle detection also has a good application prospects.

Key words: binocular vision, epipolar rectification, stereo matching, strong similarities, matching cost

摘要: 立体匹配是双目视觉的一个重要分支领域,能够通过深度图还原出三维信息,但由于其计算量庞大,实时性难以得到保障。为此,提出了一种基于强相似点的快速立体匹配算法。首先,将双目图像通过对极处理,使匹配区域固定在同一水平线上,减少匹配区域;其次,对图像进行灰度转化,并将搜索范围内与待匹配点灰度值接近的点定义为强相似点,对强相似点所在块进行匹配代价计算,并得出该点最优视差,对不存在强相似点的待匹配点进行正常视差计算;最后将进行视差修正与滤波,得到最终视差图。经Middlebury算法测试平台的提供数据进行验证,结果表明在不损失精确率的前提下,该方法相对于SAD速度提高70%左右,为立体匹配算法的实际应用奠定了良好基础,在视觉导航、障碍物检测方面也有着良好的应用前景。

关键词: 双目视觉, 极线校正, 立体匹配, 强相似点, 相似代价