Detecting Moving Shadows in Video Sequences Using Region Level Evaluation for Vision-Based Vehicle Detection

In this paper, a multistage algorithm, which is applicable to vision-based vehicle detection system, for detecting and removing cast shadows is proposed. In the first stage, a novel noise level adapted method is presented to segment foreground which consists the moving objects and moving shadows. In the second stage, the foreground is partitioned into sub-regions using multilevel thresholding. Then, a three-step region-based evaluation is employed to check whether a region is a shadow or not. Results on various real surveillance sequences, which contains shadows range from dark and small to light and large, show the effectiveness of the proposed approach.