Abstract
Recently, support vector ranking has been adopted to address the challenging person re-identification problem. However, the ranking model based on ordinary global features cannot represent the significant variation of pose and viewpoint across camera views. Thus, a novel ranking method which fuses the dense invariant features is proposed in this paper to model the variation of images across camera views. By maximizing the margin and minimizing the error score for the fused features, an optimal space for ranking has been learned. Due to the invariance of the dense invariant features and the fusion of the bidirectional features, the proposed method significantly outperforms the original support vector ranking algorithm and is competitive with state-of-the-art techniques on two challenging datasets, showing its potential for real-world person re-identification.
Talk to us
Join us for a 30 min session where you can share your feedback and ask us any queries you have