Detection of Video Objects in Dynamic Scene Using Local Binary Pattern Subtraction Method

In this paper, the problem of video object detection in dynamic scene has been addressed. The dynamism is referred to the changes in the scene of interest, due to swaying of tree branches, leaves, fluctuation of surface in case of water bodies, variation of scene illumination, etc. The problem is formulated in a fixed camera scenario and with unavailability of reference frame (background model). The local binary pattern (LBP) is a very strong element used in object detection algorithms. In the literature, many methods exist, where the LBP histograms of current frame and previous frames are combined and used for background subtraction, to get the foreground detected. This histogram computation and construction of a final histogram for the background subtraction method is a very time-consuming and complex process. The complexity can be reduced to a large extent by using our proposed window-based LBP subtraction (WBLBPS) method. Moreover, the efficacy of the proposed method in terms of correct classification is quite satisfactory as compared to the other LBP-based methods.