Moving Object Refining in Traffic Monitoring Applications

Moving object segmentation is an important task in vision-based traffic monitoring applications. In traffic scenes, various outliers such as sudden illumination changes, moving cast shadows, camera jitter, etc., often cause serious errors in image analysis due to misclassiflcation of moving objects. An efficient moving object refining approach is thus expected. In this paper, we address the problem of moving object refining by processing the background subtraction results. In an analytical multi-stage procedure, we remove sudden illumination changes and local reflected regions employing photometric color invariants, remove moving cast shadows based on a single Gaussian shadow model, uniform-region classification, and spatial analysis, and further remove other types of outliers in a postprocessing stage of area filtering and area-to-perimeter test. Experimental results on actual video sequences representative of different traffic scenes and illuminations are presented. The results illustrate that our approach is efficient when handling widely different conditions that can occur.