Fake Shadow Detection Based on SIFT Features Matching

Abstract

Shadow detection is crucial for robust and reliable visual surveillance system. The shadow detection method based on gray level and color information will fail when object parts have very similar properties with real shadow. In this paper, we present a novel shadow detection scheme which can conveniently verify detected shadows utilizing the extracted SIFT (Scale Invariant Feature Transform) features. After candidate shadow regions are detected by utilizing rgb color model, SIFT algorithm is exploited for local feature detection in two consecutive frames. Then location information of SIFT features in moving object and shadow is analyzed. Finally fake shadow regions are successfully identified. Experiments indicate that the proposed algorithm is fast and exact, and improves the accuracy of the moving object detection. The proposed method can be used for shadow detection in SIFT features tracking and will be highly effective.