Shadow removal for robust vehicle detection

Shadow removal is a critical process for any computer vision based vehicle detection system. Traditional shadow removal methods are not sufficiently robust since they frequently remove parts of the vehicles together with the shadows. For robust shadow identification and removal, shadow information must be compiled at every step of the computer vision chain from the time the shadow enters the scene until it finally disappears. In this paper we present a shadow removal algorithm that preserves the compactness of the vehicles' object masks while allowing for dealing with changing illumination conditions, long and broken shadows and multiple shadows at night. The method is based on a complex shadow model which includes, among others, information regarding: luminance and chromaticity, morphology, dynamics and spatial relations analysis.