Research on a Bayesian Estimation Image Reconstruction Method for Geiger Mode Avalanche Photodiode Light Detection and Ranging (GM‐APD LiDAR) Based on Spatiotemporal Features of Sequence Data

In scenes with strong background light, three-dimensional LiDAR imaging still faces problems such as incomplete imaging and residual background noise. This paper proposes a Bayesian-estimation image reconstruction method based on sequence data. Sparse range-intensity sequence data are constructed, and the signal probability and average intensity features of the sequence data are fused to build the likelihood term, thereby highlighting the differences between signal and noise. Through spatiotemporal average-intensity gradient analysis, the MRF prior parameters are adaptively estimated, and a spatial-continuity prior with edge-preserving characteristics is constructed. Under the Bayesian maximum a posteriori estimation framework, spatiotemporal feature fusion is transformed into an energy-minimization problem. The iterative conditional modes algorithm is then used to solve the optimal range image with spatial average intensity as a constraint, and morphological filtering is used to remove isolated noise points. With a data volume of 20,000 frames, the recovery ratio of the reconstructed range image in the 730 m scene reached 82.6%. In the 2,000 m long-range scene with an extremely low signal-to-background ratio (SBR), the recovery ratio reached 56.3%, representing a 46.2% improvement over the SPIRAL method, with an RMSE of 1.77 m. Partial reconstruction can also be achieved in low-SBR regions of long-range buildings, improving target completeness under low signal-to-background ratio conditions.