Moving Object Detection using Compound Markov Random Field Model

Abstract

We propose a novel approach of moving object detection in a video sequence. The proposed scheme uses spatial segmentation and temporal segmentation to construct the video object plane (VOP) and hence the detection of a moving objects. The spatial segmentation problem is formulated in spatiotemporal framework. A compound Markov random field model is proposed to model the video-sequences. This compound model employs edge features in the temporal direction. The MRF model parameters are selected on a trial and error basis. The labels in the spatial segmentation are estimated using Maximum a posteriori (MAP) criterion. A hybrid algorithm is proposed to obtain MAP estimates. These estimated labels are used to obtain the temporal segmentation followed by construction of video object plane. The results obtained are compared joint segmentation scheme (JSEG). It is observed that the edge based scheme proved to be best as compared to edgeless and JSEG schemes.