Abstract:Person Re-identification (Re-ID) has become an important application in video surveillance. This paper investigates the problem of fusing multiple matching and ranking results from different Re-ID methods. Ranking consensus information of various results are utilized to achieve proper fusion parameters for further improving Re-ID performances. We further introduce a supervised process to learn the fusion parameters. Experimental results on three public datasets demonstrate the superior performance of our approach compared to existing methods.

Person re-identification fusing via ranking consensus