LiDAR–Monocular Camera Fusion for Non-Cooperative Target Perception: A Review

Relative pose measurement and 3D structural perception of space non-cooperative targets provide core technical support for on-orbit servicing, space debris removal, and defunct satellite capture. A single sensor can hardly achieve high-precision ranging and high-resolution texture perception simultaneously. Fusion of Light Detection and Ranging (LiDAR) and a monocular camera combines precise depth measurement with dense semantic representation, providing an accurate and information-rich solution for space non-cooperative target perception. This paper systematically reviews LiDAR–monocular camera fusion from three levels: fusion prerequisites, methodology, and perceptual outputs. First, sensor calibration as the prerequisite is introduced. Second, fusion strategies are summarized from three perspectives: classical geometric methods, deep learning methods, and extended methods based on Neural Radiance Fields (NeRF) and 3D Gaussian Splatting (3DGS). Third, the progress, limitations, public datasets, and evaluation metrics are analyzed for three outputs: depth completion, pose estimation, and 3D reconstruction. Finally, considering the contradiction between data requirements and the difficulty of space data acquisition, future directions are discussed from two perspectives: methodology transfer from ground to space and data transfer from simulation to reality.