Statistical approaches to tracking-based moving object extraction

This paper reports a tracking-based moving object extraction algorithm, where the object location and tracking is achieved by using a template matching scheme. The structural regions of the moving objects are first detected by using a thresholding-based segmentation technique and represented coarsely on block resolution, then the interest regions of the moving objects are further formed by iterative region merging according to the spatiotemporal similarity measure and the meaningful moving objects are finally located by the temporal tracking procedure. The experimental results have confirmed that this proposed algorithm can provide more meaningful moving objects because both the spatial homogeneity of the grey levels and the temporal coherence of the motion fields are jointly exploited. This algorithm can also detect the appearance of new objects as well as the disappearance of existing objects efficiently because the relationships of the moving objects among frames are also established by the temporal tracking procedure.