Methodology used to evaluate computer vision algorithms in adverse weather conditions

Computer vision systems are increasingly present in road environments. These have not been evaluated in adverse weather conditions, particularly in rain. The objective of this article is to develop tools to validate these computer vision systems in such adverse weather conditions. This study begins by setting up a digital rain image simulator based on a set of physical laws. In addition to the simple visual effect, it makes it possible to obtain images of rain displaying physical reality. The simulator was then validated against data acquired in the Cerema R&D Fog and Rain platform. Finally, a protocol is proposed to evaluate the robustness of image features in rainy conditions. This protocol was then used to test the robustness of the Harris image feature in these conditions. Thanks to the protocol established, the study concludes that the rain has an impact on this conventional feature. (C) 2016 The Authors. Published by Elsevier B.V.