Multi-View Camera Pose Estimation for Robotic Arm Manipulation

Abstract

This paper proposes a novel approach aimed at estimating the pose of a camera, affixed to a robotic manipulator, against a target object. Our approach provides a way to exploit the redundancy of the robotic arm kinematics by directly considering manipulator poses in the model formulation for camera pose estimation. We adopt a single camera multi-shot technique that minimizes the reprojection error over all the rigid poses. The results of the proposed method are compared to four other studies employing either monocular or stereo setup. The experimental results on synthetic and real data show that the proposed monocular approach achieves better and in some cases comparable results to the stereo approach. Moreover, the proposed approach is significantly more robust and precise compared to other methods.

A Comparative Review of Hand-Eye Calibration Techniques for Vision Guided Robots

“…While the hand pose errors can be minimised by calibrating the robot [67], through reprojecting the image of the calibration pattern at each hand position and minimising the error between the real image and the reprojected image, the required hand-eye transform can then be estimated as shown in the process flow in Figure 6. Reprojection error minimisation is a well-known technique used in computer vision for pose estimation [68], [69], 3D measurements [70] and shape reconstruction [71] [72], with high level of accuracy and robustness. It shows how precise an estimated 3D world pointX recreates the true projection x on the image (see Figure 7).…”

Robust vision using retro reflective markers for remote handling in ITER

“…The possibility of optimizing over several images taken by the same camera at different poses is studied in another work [18]. In this work two multiview estimation methods have been considered: (1) using the poses reported by the robot's CS and (2) not using those.…”

“…To be able to establish a comparison with the marker-based multiview estimation methods of [18] (Multiview 1 and Multiview 2), for each image in the dataset described in section 4.1, we record an extra set of 5 random poses.…”

Reflection-Aware Generation and Identification of Square Marker Dictionaries

“…Camera localization is a fundamental process in many computer vision applications, such as augmented reality [ 1 , 2 ], autonomous driving [ 3 , 4 ], or robotics [ 5 , 6 ]. This problem is usually tackled by finding correspondences between the real environment and their projections on the camera image, followed by a Perspective-n-Point (PnP) optimization [ 7 ] to estimate the 3D rotation and translation of the camera.…”