Abstract: Binocular stereo matching plays a vital role in applications such as autonomous driving and robotic navigation. To address the limitations of supervised methods, which rely heavily on large amounts of annotated data and perform poorly in texture-less regions, this paper proposes an unsupervised binocular stereo matching method based on monocular depth consistency. Firstly, monocular depth estimation is utilized to obtain an initial depth map of the scene, introducing global structural priors to enhance depth representation in low-texture areas. Secondly, a depth estimation consistency module is constructed to align the left and right views in 3D space through point cloud rigid body transformations, optimizing the depth results using Chamfer distance and color consistency constraints. Finally, a disparity reconstruction algorithm based on point cloud nearest neighbor search is proposed, which transforms the disparity estimation problem into a point cloud matching task in 3D space. By leveraging nearest neighbor search to establish precise correspondences within the 3D point cloud, this approach significantly reduces computational complexity and improves registration accuracy compared to traditional methods. Experimental results on the KITTI 2012, KITTI 2015, and Virtual KITTI 2 datasets demonstrate that the proposed method outperforms classical unsupervised methods such as DispNet and achieves performance close to that of supervised methods like AANet and GwcNet, validating its advantages in reconstruction accuracy for texture-less regions and generalization capability.