Computer Engineering and Applications ›› 2021, Vol. 57 ›› Issue (23): 219-225.DOI: 10.3778/j.issn.1002-8331.2104-0353

• Graphics and Image Processing • Previous Articles     Next Articles

Research on Stereo Matching Algorithm Based on Multi-feature Fusion and Adaptive Aggregation

CHANG Yawen, ZHAO Dongqing, SHAN Yanhu   

  1. School of Instruments and Electronics, North University of China, Taiyuan 030051, China
  • Online:2021-12-01 Published:2021-12-02

多特征融合和自适应聚合的立体匹配算法研究

畅雅雯,赵冬青,单彦虎   

  1. 中北大学 仪器与电子学院,太原 030051

Abstract:

Aiming at the low matching accuracy of local stereo matching in illumination distortion and weak texture areas, a stereo matching algorithm based on cost calculation of multi-feature fusion and adaptive cross window aggregation is proposed. Firstly, the HSV color space component is introduced, combined with the improved Census transform and gradient information to calculate the matching cost, which eliminates the influence of the parallax boundary abnormal value, and enhances the algorithm’s robustness to illumination distortion. Secondly, an adaptive window cost aggregation method based on gradient information and variable color threshold is proposed to improve the matching accuracy of weak texture areas. Finally, the final disparity result is obtained by multi-step disparity refinement. Experimental results show that, compared with AD-Census algorithm, the mismatching rate of proposed algorithm is reduced by 3.24% under the same illumination exposure conditions. It can effectively solve the problem of mismatching parallax boundary and weak texture region, and has good robustness to illumination distortion and can effectively suppress noise interference.

Key words: machine vision, stereo matching, multi-feature fusion, adaptive window, radiometric distortion

摘要:

针对局部立体匹配在光照失真和弱纹理区域匹配精度低的问题,提出了一种多特征融合的代价计算和自适应十字窗口聚合的立体匹配算法。引入HSV颜色空间分量,结合改进后的Census变换和梯度信息作为匹配代价计算方法,排除了视差边界异常值的影响,增强了算法对光照失真的稳健性;提出了基于梯度信息和可变颜色阈值的自适应窗口代价聚合方法,提高了在弱纹理区域的匹配精度;通过视差计算和多步骤的视差精细得到了最终的视差结果。实验结果表明,所提算法较AD-Census算法在无光照失真条件下误匹配减少了3.24%,能有效解决视差边界和弱纹理区域错误匹配的问题,对光照失真稳健性好且能有效抑制噪声干扰。

关键词: 机器视觉, 立体匹配, 多特征融合, 自适应窗口, 光照失真