基于改进Census变换与梯度融合的立体匹配算法

双目立体匹配根据视差原理将平面视觉转化到三维立体视觉,是三维重建的核心步骤之一。针对局部立体匹配算法在深度不连续、弱纹理区域匹配精度低且易受光照、噪声等因素干扰的问题,提出了一种改进的立体匹配算法。首先,在代价计算阶段将改进的Census代价与梯度代价进行融合,并采用引导滤波算法对图像进行多尺度代价聚合;然后,采用赢家通吃算法计算初始视差;最后,采用左右一致性检测、中值滤波进行视差后处理,得到最终的视差图。实验结果表明,本算法在Middlebury2.0测试平台上的平均误匹配率为5.11%,且具有很好的稳健性和实用性。

Stereo Matching Algorithm Based on Improved Census Transform and Gradient Fusion

Binocular stereo matching transforms plane vision into three-dimensional stereo vision based on the parallax principle,which is one of the core steps of three-dimensional reconstruction.Aiming at the problems of local stereo matching algorithm in the depth discontinuity,low matching accuracy in weak texture areas,and easy to be interfered by factors such as light and noise,an improved stereo matching algorithm is proposed in this paper.First,in the cost calculation stage,the improved Census cost and the gradient cost are fused,and the guided filtering algorithm is used to perform multi-scale cost aggregation on the image;then,the winner-take-all algorithm is used to calculate the initial disparity;finally,the left-right consistency detection,middle value filtering performs disparity post-processing to obtain the final disparity image.Experimental results show that the average mistake match rate of the algorithm on the Middlebury2.0 test platform is 5.11%,and it has good robustness and practicability.