Robust stereo tracking for space applications

Abstract

Abstract-This paper proposes a real-time, robust and efficient 3D model-based tracking algorithm for visual servoing. A virtual visual servoing approach is used for 3D tracking. This method is similar to more classical non-linear pose computation techniques. Robustness is obtained by integrating an Mestimator into the virtual visual control law via an iteratively reweighted least squares implementation. The presented approach is also extended to the use of multiple cameras. Results show the method to be robust to occlusion, changes in illumination and miss-tracking.

Vision-based space autonomous rendezvous: A case study

“…Using cameras, [1], [2], [15] propose to rely on easy to detect and track markers installed on the target. Other techniques under study in a space context deal with natural features with stereo vision methods to recover the pose, given a 3D model of the object [8], [11]. With a monocular camera, a feature-matching computer vision approach had been selected in [14].…”

3D model based pose estimation for omnidirectional stereovision

“…The proposed approach is based on works of Marchand et al [7], [8] where the pose computation problem is defined as a virtual visual servoing one. This virtual servoing process is similar to a full scale non-linear estimation method of the pose.…”

“…Dionnet et al in [7], modelled the virtual visual servoing problem for a two perspective cameras stereo rigid system saying that knowing the system calibration (two sets of intrinsic parameters and the second camera pose w.r.t. the first), we can rewrite the ∆ criterion to take into account two cameras.…”

“…So, as in [7], we use a M-estimator [9] to deal with outliers. A lot of functions have been proposed in the literature.…”

“…This is the Virtual Visual Servoing (VVS) framework introduced by Sundareswaran et al in [4] and then by Marchand et al in [5], a full scale non-linear optimisation technique which can be used for pose computation. Some works have been done, for perspective camera [6], stereo perspective rig [7] and even for monocular omnidirectional vision [8]. These works also deal with corrupted data, which is frequent in image feature extraction, using the widely accepted statistical techniques of robust M-estimation [9].…”