An Adaptive Cast Shadow Detection with Combined Texture and Color Models

Abstract

Cast shadows of moving foreground objects can cause miss tracking problem in object detection and tracking, thus shadow detection is an important step used after a moving foreground object is detected. Most of current methods have a significant trade-off between the shadow detection rate and the shadow discrimination rate. In this paper, an effective and adaptive method with combined texture and color models is proposed in order to achieve good shadow detection rate and shadow discrimination rate as well. Firstly, Scale Invariant Local Ternary Pattern (SILTP) is used to select a candidate shadow region. Then HSV color model is employed to detect a new candidate shadow region by using maximum likelihood estimation (MLE) to estimate the thresholds of HSV color model adaptively. Finally the two regions are combined by logical operation and a new shadow region can be obtained. Our experimental results show that the proposed method achieves a better performance in both shadow detection rate and discrimination rate compared to the other current methods. Moreover, the proposed method runs at 100 frames per second and is suitable for the real-time detection and tracking.