The 3D reconstruction based on binocular stereo vision is a main part of the computer vision technology and it has a variety of applications such as robot navigation, aerial surveying and mapping, medical imaging, industrial inspection, etc.. A new binocular stereo vision method based on comparing gray value of the projection points is proposed to reconstruct 3D world. This method divides the world coordinate system into grids and the nodes are considered as candidates of the object coordinates. A group of points with varying depth information are projected on the two pictures. The difference of the gray values of the corresponding pairs implies the depth information of the object in 3D world. It is proved superior over traditional methods in the effectiveness and efficiency by simulatron experiments using Matlab. Compared with traditional method based on image matching, this algorithm has obvious superiority in simplicity, computation and accuracy. Also, it will not be affected by the nonlinear distortion from the camera.