一种面向在轨服务目标的物理退化感知两阶段三维重建方法及系统 A Physics-Degradation-Aware Two-Stage 3D Reconstruction Method and System for On-Orbit Servicing Targets

This invention discloses a physics-degradation-aware two-stage 3D reconstruction method and system for on-orbit servicing targets, addressing proximity perception and geometric reconstruction for satellite life extension, space-debris removal, and on-orbit assembly and maintenance. To counter coupled in-orbit physical degradations—solar irradiation saturation, specular multipath from multi-layer insulation (MLI), thermally induced boresight and extrinsic drift, platform micro-vibration, and data latency—that cause false alarms, trailing, layering, and geometric stretching in point clouds, the method comprises: S1 a high-fidelity simulation platform integrating an orbital-dynamics engine, a target 3D model library, illumination/thermal environment models, and ray-traced rendering; S2 coupled physical degradation models (LiDAR beam-divergence trailing, multipath ghost points, walk error; depth-camera saturation and stereo-matching failure; IMU attitude error; extrinsic thermal-drift harmonics; stray-light false alarms and data latency) injected into simulated point clouds via an error-injection interface; S3 "manifold-to-point" two-stage observation—stage one performs latitude-band uniform spherical fly-around at the keep-out-sphere radius to build a topologically complete global manifold point cloud, while stage two actively approaches along the safe stand-off vector with a trapezoidal velocity profile and fuses via ICP/Scan-to-Map registration to raise local density and geometric accuracy; S4 adaptive switching between LiDAR-only and LiDAR-Depth fusion modes based on the solar geometric-alignment factor, saturation risk, and observing range; S5 a degradation-aware reconstruction pipeline (IMU-interpolated motion compensation, multi-frame fusion, voxel downsampling, and cascaded SOR/ROR denoising with thresholds adaptively set from degradation priors); S6 output of the 3D model together with Chamfer distance, RMSE, Z-axis systematic bias, and topological completeness, forming a closed modeling–observation–reconstruction–evaluation loop. Tests show completeness above 99% with centimeter-level RMSE and over 60% lower Chamfer distance than a baseline without degradation priors.