Novel Method of Monocular Real-Time Feature Point Tracking for Tethered Space Robots

Abstract

This paper proposes a visual perception system for a tethered space robot’s (TSR) automatic rendezvous from 100 to 0.15 m. The core problem, tracking the entire contour of noncooperative moving targets in real time, is emphasized in this work. Given numerous challenges in a dynamic scene, a novel feature tracking algorithm is developed, i.e., the monocular real-time robust feature tracking algorithm (MRRFT). To generate a robust target model, improved speeded-up robust features (SURF) are used to extract features from a marked target box. The tracker then uses the pyramid Kanade-Lucas-Tomasi (P-KLT) matching algorithm and eliminates mismatched points by a statistical method. The greedy snake algorithm is applied to obtain the exact location of the target box and to update it automatically. A discrete feature filter and an adaptive feature updating strategy are also designed to enhance robustness. A three-dimensional (3D) simulation and a semiphysical system are developed to evaluate the method. Numerous experiments demonstrate that the tracker can stably track satellite models with simple structures with improved accuracy and time savings than good features to track (GFTT)+P-KLT or scale invariant feature transform (SIFT)+P-KLT.