Testing Photometric Stereo Applications

Testing vision software and systems is an important activity due to its increasing use in applications like autonomous driving or quality assurance in production. In this paper, we introduce an approach for testing vision applications relying on photometric stereo that is used for obtaining 3D models from a single camera using multiple light sources. In particular, we focus on testing an application prototype implementation used for estimating the quality of riblet surfaces used to reduce different kinds of friction. The approach is based on modifying images used as inputs of photometric stereo. The underlying modification operations like darkening or brightening correspond to specific fault situations that may arise during operation. The generated modified images are given as input to the vision software leading to deviations on side of the classification o utput. I deally, such systems should detect various faults and deliver corresponding error messages. Besides outlining the foundations, we present results obtained from the riblet surface inspection software currently under development.