Embedded local feature based background modeling for video object detection

Background modeling has been one of the approaches for detecting foreground in a video. The challenge is when some of the entities of background are dynamic instead of being static. In this paper, we propose a feature embedding scheme to model background having some dynamic objects and varying illumination conditions. The two local feature extracting operators such as Local Binary Pattern (LBP) operator and Gabor filter have been appropriately embedded to model texture backgrounds with dynamic entities. The embedding has been in non linear frame work and the notion of information theoretic measure has been used to take care of the above two condition in the background. This background model has learned to efficiently model the background of the video. The performance of the proposed feature embedded algorithm has been found to be better than those of Huerta et al.'s [11] algorithm and Heikkila et al.'s [12] algorithm. Simulation results have been presented for video frames of PETS sequences.