Optimized Spatial Matching for Visual Object Tracking

Abstract

The biggest challenge for visual object tracking is the simultaneous requirements for robustness and discrimination. Although there are many algorithms to study current problems, this problem still cannot be overcome. In this paper, inspired by Siamese network SPM-Tracker, a new target tracking algorithm-OSM-Tracker is proposed. The algorithm is a two-stage Siamese network tracker composed of an optimized space network and a correction network. Through the cooperation of these two aspects with SPM-Tracker, compared OSM-Tracker has produced good results. Experiments have proved that our tracker has achieved a considerable performance improvement and achieved real-time effects on OTB-100and LaSOT.