A method for vehicle classification and resolving vehicle occlusion in traffic images

This paper presents a new method to classify vehicles with resolving vehicle occlusions in traffic images. Moving objects are detected in an image sequence using a probability-based background extraction and object segmentation algorithm. The partially occluded vehicles are detected by evaluating the convexity of the moving objects and split by the so-called “dividing line” of the occlusion region. Then the divided objects are classified by evaluating their normalized size. If the object is not partially occluded, its width and the ratio between length and width is extracted to detect if it is a full occlusion and classify it by developing a hierarchical classifier. The proposed method has been evaluated to see if the results are satisfying. Experimental results exhibit that the method is efficiently able to classify vehicles and process occlusions.