Teaching scans are cumbersome and have poor versatility when performing scan reconstruction. Viewpoint planning has continued to focus on automatically obtaining the minimum set of viewpoints covering the model. To realize automated 3D scanning and reconstruction of parts with different complexity levels, we study issues such as viewpoint redundancy, viewpoint occlusion, and binocular reconstruction constraints that may occur during viewpoint planning. First, given the difficulty of completely scaning the model with existing viewpoint planning, Lloyd's algorithm is improved by analyzing the characteristics of surface structured light scanning and the energy function of Euclidean distance and normal vector deviation is applied to perform Voronoi partitioning of the model to generate an initial scanning viewpoint. Then, to address the viewpoint redundancy problem, an iterative algorithm for splitting the initial scanning viewpoints is proposed. Finally, given the problem that the generated viewpoints are prone to occlusion, a line-of-sight de-occlusion strategy is proposed. Moreover, to improve the model coverage, a method of using panning viewpoints is proposed. The experimental results show that under the optimal number of viewpoints, the coverage rate of automobile castings and shells reaches more than 94%, and that of the simple curved automobile sheet metal reaches more than 99.5%, and automatic planning and scanning of the automotive steering knuckle is realized. Planning scanning meets the coverage and efficiency requirements of automatic viewpoint planning and the adaptability requirements for parts with different complexity levels.