3D Reconstruction of Space Objects from Multi-Views by a Visible Sensor.

Area of Science:

• Computer Vision
• Robotics
• Aerospace Engineering

Background:

• Accurate 3D structural modeling of space objects is crucial for on-orbit servicing, debris removal, and asset management.
• Existing 3D reconstruction methods struggle with symmetric structures and repetitive textures common in space objects, leading to inaccurate models.

Purpose of the Study:

• To develop a novel 3D reconstruction framework for generating accurate and complete 3D structural models of space objects from visible sensor imagery.
• To address challenges in 3D reconstruction caused by object symmetry and texture repetition.

Main Methods:

• Utilizes Structure from Motion (SFM) to estimate camera poses and sparse point depths, followed by Patch-based Multi-View Stereo (PMVS) for dense point cloud generation.
• Introduces an image sequencing strategy within SFM to mitigate matching errors from symmetric structures and textures.
• Incorporates a refinement process leveraging prior knowledge of geometric primitives for artificial space objects.

Main Results:

• The framework successfully generated accurate and fully-covered 3D point cloud models of space objects from both simulated and real image datasets.
• The proposed refinement strategy effectively removed outliers and noise, significantly improving the structural integrity and visual quality of the reconstructed models.

Conclusions:

• The novel 3D reconstruction framework provides a robust solution for accurately modeling space objects, overcoming limitations of traditional methods.
• The refinement process enhances the practical utility of the reconstructed models by improving their geometric accuracy and visual clarity.