针对PatchMatchStereo立体匹配算法在实现倾斜平面时，因使用随机函数生成平面参数而导致算法计算量大且误匹配率高的问题，提出一种基于局部一致性约束的立体匹配算法。首先，通过对图像中的像素进行稀疏匹配获得视差置信度高的支撑点；其次，利用三角剖分为图像内各像素点确定一个三角平面，计算平面参数并分配给该平面内的点；然后，通过迭代传播为每个像素点找到更加准确的平面参数，构建出局部一致性平行窗口模型；最后，通过平面参数计算视差值并通过视差后处理优化视差。本文算法在Middlebury评估平台第三版标准测试数据集上进行实验，实验结果表明，处理后的平均误匹配率比PMS算法降低了4.39%，其中对单个图像的误匹配率最高降低15.42%。本文算法在降低误匹配率的同时提高了图像处理的效率，相较于其他算法具有显著的优越性。

Aiming at the problem of large computational complexity and high mismatch rate caused by the use of random functions to generate planar parameters in the PatchMatchStereo stereo matching algorithm when implementing inclined planes, a stereo matching algorithm based on local consistency constraints is proposed. First, the algorithm obtains support points with high parallax confidence by sparse matching of pixels in the image. Second, the algorithm uses triangulation to determine a triangular plane for each pixel in the image, calculates the plane parameters and assigns them to the points in the plane. Then, a more accurate plane parameter is found for each pixel through iterative propagation, and a local consistent parallel window model is constructed. Finally, the disparity is calculated through the plane parameter and optimized through the disparity post-processing. The algorithm in this paper is tested on the standard test data set of the third edition of the Middlebury evaluation platform. The experimental results show that the average error matching rate after processing is 4.39% lower than that of the PMS algorithm, and the maximum error matching rate of a single image is reduced by 15.42%. The algorithm in this paper improves the efficiency of image processing while reducing the false matching rate, and has significant advantages over other algorithms.