Radiation Knowledge Based Gaussian Shadow Detection

Intelligent traffic detection has been paid much attention in many years, and the shadow removal was the key problem in this technique. Usually, geometric characteristics were used for vehicle detection or person detection, but it is not robust to shadow. Shadow in this paper was formed by vehicle's occlusion in the direction of the incident light and we want to remove the part projected in background. Because neighboring regions on a smooth curved surface have similar surface normal and illumination conditions but have different reflection factor because they are of different materials. In this paper, we combine spectrum and radiation knowledge with Gaussian shadow detection algorithm, which makes use of the reflection and radiation difference among regions to provide referential information to Gaussian shadow detection model, thus separating the moving object from its shadow more accurately.