Adaptive Moving Cast Shadow Detection by Integrating Multiple Cues

Moving cast shadow detection and removal is a key step for accurate object detection in intelligent transportation system. This paper proposes a robust cast shadow detection algorithm by integrating multiple cues. Firstly, a weak shadow detector is adopted to detect these potential shadow pixels; Then three adaptive shadow estimators are designed and cascaded to integrate texture, chromaticity, brightness as well as spatial-temporal context for eliminating the object pixels so that this algorithm can robustly detect the moving cast shadow in the various environments; Lastly, spatial adjustment is employed to verify the shadow detection results of these three shadow estimators. Experimental results on indoor and outdoor video sequences show that this proposed algorithm can robustly detect moving cast shadow and rapidly adapt to variations in traffic surveillance scenarios.