Robust Reflection Detection and Removal in Rainy Conditions using LAB and HSV Color Spaces

Abstract

In the field of traffic monitoring systems, shadows are the main causes of errors in computer vision-based vehicle detection and classification. A great number of  research have been carried out to detect and remove shadows. However, these research works only focused on solving shadow problems in daytime traffic scenes. Up to now, far too little attention has been paid to the problem caused by vehicles’ reflections in rainy conditions. Unlike shadows in the daytime, which are homogeneous gray shades, reflection shadows are inhomogeneous regions of different colors. This characteristic makes reflections harder to detect and remove. Therefore, in this paper, we aim to develop a reflection detection and removal method from single images or video. Reflections are detected by determining a combination of L and B channels from LAB color space and H channel from HSV color space. The reflection removal method is performed by determining the optimal intensity of reflected areas so that they match with neighbor regions. The advantage of our method is that all reflected areas are removed without affecting vehicles’ textures or details.