The 3D reconstruction based on binocular stereo vision is a main part of the machine vision technology and has a variety of applications in robot navigation, aerial surveying and mapping, medical imaging, industrial inspection, etc.. With analysis of the relation between 3D coordinates of the mesh candidates and 2D coordinates in images, the mesh candidates similarity of gray value based stereo vision method was improved. Multi-resolution mesh grid was established corresponding to the wavelet decomposed images. The depth information of the object was reconstructed according to the similarity of the gray values between both images with the choosing strategy from the rough to the fine. The algorithm has obvious superiorities in computation and accuracy shown by experiments using Matlab.