Foreground Regions Extraction and Characterization Towards Real-Time Object Tracking

Abstract

Object localization and tracking are key issues in the analysis of scenes for video surveillance or scene understanding applications. This paper presents a contribution to the object tracking task in indoor environments surveyed by multiple fixed cameras. The method proposed uses a foreground separation process at each camera view. Then, a 3D-foreground scene is modeled and discretized into voxels making use of all the segmented views, preventing the difficulties of inter-object occlusions in 2D trackers, and increasing the robustness for not having to rely only in one view. The voxels are grouped into meaningful blobs, whose colors are modeled for tracking purposes, using a novel voxel-coloring technique that considers possible inter/intra-object occlusions. Finally, color information together with other characteristic features of 3D object appearances are temporally tracked using a template-based technique which takes into account all the features simultaneously in accordance with their respective variances. Extensive experiments dealing with several hours of video sequences in real-world scenarios have been conducted, showing a very promising performance.