Robust and Efficient Person Re-Identification Model using K-Nearest Neighbor Graph

Abstract

the state-of-art person re-identification (prid) models for ranking generally depends on labeled pairwise feature sets information to learn a task-dependent distance metric. Further, in retrieval process, re-ranking is an important mechanism for enhancing the accuracy. However, very limited work is carried out for designing a re-ranking method, particularly for automatic and unsupervised strategies. The existing re-ranking based prid model is not efficient when multiple persons appears simultaneously in second camera. This is because the existing model identify person in second camera by matching the feature sets with feature sets in first camera, individually with respect to other person in the second camera. For overcoming research problem, this paper present robust and efficient prid (reprid) model. First, present a robust learning/ranking method using k-nearest neighbor (knn) graph. Then, this work present a re-ranking method to improve accuracy of prid by using information of co-occurrence persons for matching and reorganizing given rank lists. Experiment are conducted on standard dataset shows robustness and effectiveness of proposed prid method.