After got the scattered point cloud model from the 3D scanner, first search the K nearest neighbors for each point by constructing the K D tree; Second according to those neighborhood relationships, use the least square fitting planar method to estimate the normal vector of each point. Next determine the boundary of point cloud, select the extreme point as the initial point, and build seed triangle. Finally use the local optimal triangle mesh growing algorithm based on multiple constraints, start from the seed triangle, make the edge as extension condition, search for suitable point to construct new triangle layer by layer. During this process four constraints added, to better select the extension point, and extend the triangle edge outward, finally generate adjacent triangle meshes, realize the surface reconstruction more realistically.