2019/07/01 by Yi Zhang, Chao Zhang, Zhang, Yi +3
Computer Science · #Advanced Image and Video Retrieval Techniques #Video Surveillance and Tracking Methods #Human Pose and Action Recognition
paper · pdf · doi:10.48550/arxiv.1907.01150
We propose a novel multi-scale template matching method which is robust\nagainst both scaling and rotation in unconstrained environments. The key\ncomponent behind is a similarity measure referred to as scalable diversity\nsimilarity (SDS). Specifically, SDS exploits bidirectional diversity of the\nnearest neighbor (NN) matches between two sets of points. To address the\nscale-robustness of the similarity measure, local appearance and rank\ninformation are jointly used for the NN search. Furthermore, by introducing\npenalty term on the scale change, and polar radius term into the similarity\nmeasure, SDS is shown to be a well-performing similarity measure against\noverall size and rotation changes, as well as non-rigid geometric deformations,\nbackground clutter, and occlusions. The properties of SDS are statistically\njustified, and experiments on both synthetic and real-world data show that SDS\ncan significantly outperform state-of-the-art methods.\n