Resolving Occlusion in Multi-Object Tracking through Integrated Fuzzy Similarity Measure

In multi-object tracking, occlusion is a situation where part of an object is covered by another object or any structure in the video scene. It is a very common problem in multi-object tracking for real world video scenes and is a cause for poor tracking performance. Considering its significance and inevitability, this problem has been a subject of numerous papers about multi-object tracking. In this paper, a method for occlusion handling based on fuzzy approach is proposed. Fuzzy techniques are used here as they can deal with uncertainty and imprecision which are inherent in image/video processing. The method consists of feature extraction, fuzzy feature representation, merge-split event detection, and track resolution. The main contribution of this paper is in the use of fuzzy similarity measure together with fuzzy integral for resolving object tracks after occlusion. The similarity measure is performed separately on color, texture, and shape after representing them as fuzzy features. Then, fuzzy integral combines them to calculate the overall similarity value. Experimental result shows that with moderately fast computational time, the proposed method can resolve occluded tracks accurately even in difficult situations. This result also shows the promising applicability of fuzzy approach for future automated video surveillance research. Index Term-Multi-object tracking, occlusion handling, fuzzy similarity measure, fuzzy integral.