Space debris: pose estimation using stereo vision

[摘要] ENGLISH ABSTRACT: Tracking the relative attitude and position of uncooperative in-orbit objects is vitalfor autonomous operations in space. Vision-based solutions have low powerconsumption and can provide a system with valuable information to perform posedetermination. Estimation algorithms are required to extract the system states fromvisual measurements and many similar approaches have been investigated in mobilerobotics.In this thesis, a chaser satellite is fitted with stereo cameras which are used toextract unique features on the surfaces of an uncooperative, unknown target. Thescale invariant feature transform (SIFT) feature detector is used to identify andestablish correspondence of the target features. A full state kinematic estimatoris implemented using an extended Kalman filter (EKF) based on the simultaneouslocalisation and mapping (SLAM) approach. The filter makes use of the observationsfrom the feature extractor to estimate the position and orientation of the targetrelative to the chaser along with the angular and linear velocities of the target.Shape and size reconstruction of the target is made possible using the sparselytracked features.A simulation environment providing ground truth is used to verify the performanceof the estimation algorithm. The integration of the estimator with the feature extractoris assessed using real world data. Experimental data is obtained from imagesequences of a moving target in a laboratory set-up. Results show that the filter estimates the system states successfully and that the developed feature extractor iscapable of detecting robust features with reliable correspondence. It is concludedthat the use of a stereo-vision-based estimator is a viable option for autonomousoperations such as space debris removal and satellite service missions.

[发布机构] Stellenbosch University