Viewpoint reduction in vision sensor planning for dimensional inspection

This paper addresses the vision sensor planning for part dimensional inspection. To efficiently inspect large sheet metal parts from automotive manufacturing, it is highly desirable to obtain the minimum number of camera viewpoints with each satisfying all the given task constraints. However, the general minimum-viewpoint problem is NP-hard. Based on our previous work, a novel method is developed to solve the problem into its sub-optimality. This method first generates candidate viewpoints using a decomposition-based approach. Then the minimum-viewpoint problem, is rendered as an integer optimization problem -set-partition problem, which can be solved to its sub-optimality using existing algorithms and software. Experimental results on real-world parts demonstrate the effectiveness of the new method.