Research On Hyper-graph And Application Of Relation Description In Tracking

This paper mainly focuses on hyper-graph matching and applying itscore idea, relation description, to real-time tracking. Finally, wepreliminarily make a summary and outlook for relation description. In thispaper, we present a non-rigid object matching algorithm based onhyper-graph with bundled features，a real-time tracking algorithm based onadaptive template matching with relation description and an occlusiondetection algorithm based on relation between screen and target.Experiments verify the good performance of our algorithms.In the study of hyper-graph matching, we use it to address non-rigidobject matching problem. According to deformation and interference, weextract bundled features based on regions, we adjust constraints of model,similarity formula and matching results. Finally, we present a non-rigidobject matching algorithm based on hyper-graph with bundled features.Experiments on simulation and real samples verify the good performance ofour algorithm.Based on the study of hyper-graph matching, we focus on its core idea,relation description, we use it to address real-time tracking problem. Basedon our affine invariant relation description which can adaptively adjust thewindow in template matching, we present a real-time tracking algorithmbased on adaptive template matching with relation description. Comparativeexperiments verify the good performance of our algorithm. In addition, byanalyzing occlusion in tracking from a new perspective, we present anocclusion detection algorithm based on relation between screen and target.Experiments verify the good performance of it.Finally, we preliminarily draw a conclusion and make an outlook for its future work and research approach.