A Fast and Robust Re-Identification Method for Large Video Data

摘要

This paper proposes a fast and robust person re-identification method from videos. Naive re-identification methods may falsely recognize images of a same person as different people due to continuous change of environments around cameras even in a same place. This failure causes multiple result sets of images of different people, which should be one. This paper addresses the problem to propose a method that enhances the robustness and computation speed of the re-identification. The robustness of the re-identification is improved by employing video tracking technologies. The speed of that is increased with a Luigi index, which is a data structure for fast similarity search. Experimental results showed that the proposed method shortens execution time by 10% and reduces the number of redundant result sets by a factor of 12.6 compared to the naive one.