Surface sampling for optimal viewpoint generation

The automation of visual quality inspection is becoming increasingly important in manufacturing industries. Visual quality inspection aims to ensure that intermediate and/or final products meet certain quality features. However, automating quality inspection is challenging because the inspection process depends on the object geometry and the types of defects often encountered during manufacturing steps. As a result, manual inspection by pre-trained personnel is currently still used as the method of choice in most industries. Although some methods for automatic visual inspection planning have been proposed, each of them struggles to provide high-speed inspection with high resolution and complete (100%) coverage. In this paper, we propose a novel approach to automatically generate viewpoints, which is the first step in an automatic and flexible visual inspection pipeline. Our viewpoint generation algorithm consists of three main steps: (i) Poisson disk sampling on the target mesh and translation of the resulting samples to a predefined focusing distance along their corresponding normal vectors, (ii) sight ray tracing, and (iii) selection of the viewpoints by solving an integer linear programming problem. Additionally, we perform hyperparameter tuning using Bayesian optimization. Compared to the algorithms presented in the literature, our approach does not require a preprocessing step, and simulation results show that complete object surface coverage can be achieved even for objects with complex geometries. We demonstrate the performance of our method in terms of coverage by comparing simulation results directly with results from the literature.