Robustness of visual perception system in progressive challenging weather scenarios

Traditional field test and laboratory test can only evaluate hardware performance, and cannot test the robustness of artificial intelligence (AI) device for object detection, instance segmentation, etc. under task scenarios. For AI-based visual perception system, this need for evaluation and testing is very urgent. In this paper, we consider the adverse effects of challenging weather on the visual perception system, investigate the robustness of visual perception system affected by adverse weather through imaging, and propose a method to evaluate the robustness of object detection model under different weather variations. To begin, we analyze the components that influence the visual perception system’s robustness and develop a weather generation model under rainy and foggy scenes. After that, quantifiable weather meta-datasets with visibility in fog and rain intensity as variable parameters are built. Finally, regression analysis and statistical test are used to verify the nonlinear relationship between model performance and weather variations, the robustness of state-of-the-art object detection algorithms under progressive challenging weather images is evaluated.