VIPE: Visible and Infrared Fused Pose Estimation Framework for Space Noncooperative Objects.

Area of Science:

• Spacecraft engineering
• Computer vision
• Robotics

Background:

• Accurate pose estimation of non-cooperative space objects is vital for space missions like debris removal and satellite servicing.
• Existing monocular pose estimation methods struggle in low-visibility space environments.

Purpose of the Study:

• To develop a novel deep learning framework for accurate pose estimation of non-cooperative space objects using fused visible and infrared imagery.
• To address the limitations of monocular methods in challenging space conditions.

Main Methods:

• Proposed a Visible and Infrared Fused Pose Estimation Framework (VIPE) integrating an image fusion subnetwork and a pose estimation subnetwork.
• The fusion subnetwork merges multi-scale features from visible and infrared images.
• Developed a new Bimodal-Vision Pose Estimation (BVPE) dataset with 3,630 visible-infrared image pairs.

Main Results:

• The VIPE framework significantly outperforms existing monocular pose estimation methods.
• Demonstrated superior performance in complex and low-visibility space environments.
• Achieved more reliable and accurate pose estimation results.

Conclusions:

• The proposed VIPE framework effectively fuses visible and infrared imagery for robust pose estimation of non-cooperative space objects.
• The BVPE dataset provides a valuable resource for advancing research in this domain.
• VIPE offers a promising solution for critical space operations requiring precise object localization.