Spatial resolution metric for optimal viewpoints generation in visual inspection planning

The automation of visual quality inspection is becoming increasingly important in manufacturing industries. The objective is to ensure that manufactured products meet specific quality characteristics. Manual inspection by trained personnel is the preferred method in most industries due to the difficulty of identifying defects of various types and sizes. Sensor placement for 3D automatic visual inspection is a growing computer vision and robotics area. Although some methods have been proposed, they struggle to provide high-speed inspection and complete coverage. A fundamental requirement is to inspect the product with a certain specific resolution to detect all defects of a particular size, which is still an open problem. Therefore, we propose a novel model-based approach to automatically generate optimal viewpoints guaranteeing maximal coverage of the object’s surface at a specific spatial resolution that depends on the requirements of the problem. This is done by ray tracing information from the sensor to the object to be inspected once the sensor model and the 3D mesh of the object are known. In contrast to existing algorithms for optimal viewpoints generation, our approach includes the spatial resolution within the viewpoint planning process. We demonstrate that our approach yields optimal viewpoints that achieve complete coverage and a desired spatial resolution at the same time, while the number of optimal viewpoints is kept small, limiting the time required for inspection.