An improved shadow removal algorithm based on gradient amendment

This paper presents an improved method for removing cast shadows from multiple objects in a static background using gradient amendment. Shadow can decrease object detection rate and increase the likelihood of tracking failure, which are the two most important measure of algorithm in intelligent video system. The drawback of existed shadow detection methods is the fracture problem of objects, especially for pixel-based method such as chromaticity-based method and texture-based method. A novel method based on gradient amendment is proposed to reduce the possibility of this issue, which has obvious advantages on the protection of object´s integrity and the resistance of objects´ shadow. We have verified the improved method on a series of indoor and outdoor video sequences.