A background modeling and foreground segmentation approach based on the feedback of moving objects in traffic surveillance systems

摘要

Background modeling and foreground segmentation are the foundation of traffic surveillance systems. The preciseness of the background model and the accuracy of the foreground segmentation will directly affect the subsequent operations, such as object detection, target classification and behavior understanding. Additionally, the processing time is limited for real applications. The background modeling and foreground segmentation approaches, unfortunately, often have to make two tough trade-offs, including the one between the robustness to background changes and the sensitivity to foreground abnormalities and the other between suppressing noise and reducing the erroneous holes and splitting in foreground segmentation. To deal with these problems, an improved background modeling and foreground segmentation approach based on the feedback of the tracking results of moving objects is proposed. According to the achieved object tracking results, a frame image is divided into four kinds of regions, then a dual-layer background updating is done for these different regions with appropriate operations, which can significantly improve the quality of the background model. Based on the spatial relationship among the tracked objects, the predicted object blocks are merged into regions, among which adaptive segmentation thresholds are used for foreground segmentation. This adaptive threshold approach can efficiently avoid the erroneous holes and splitting in foreground segmentation. Our proposed approach is validated with several public data sets, which confirm its advantages over many existing approaches.