Multipath Interference Suppression of Amplitude-Modulated Continuous Wave Coaxial-Scanning LiDAR Using Model-Based Synthetic Data Learning

This article proposes a novel model-based synthetic data learning approach to suppress the multipath interference (MPI) in the amplitude-modulated continuous wave (AMCW) coaxial-scanning LiDAR. Previous works have focused on the MPI suppression in conventional AMCW time-of-flight (ToF) sensors with flash-type illumination sources based on the various MPI assumptions, whose MPI errors still remain in cm-scale. To achieve mm-scale MPI error suppression, this article proposes a novel learning-based MPI error suppression method implemented in a coaxial type AMCW scanning LiDAR in which the MPI phenomenon can be precisely modeled. Specifically, to model the MPI-generating mechanism in the AMCW coaxial-scanning LiDAR, a novel dedicated light transport model is designed at the simulation level. Using this precise MPI simulation model, the synthetic dataset reflecting the MPI mechanism in coaxial scanning LiDAR is generated and used for the training of the Bayesian-optimized eXtreme gradient boosting (XGBoost) ensemble. This trained XGBoost is then used for the MPI error suppression of the AMCW coaxial-scanning LiDAR. Experimental validation showed that the mean absolute error (MAE) of MPI error can be reduced to less than 2 mm by the proposed method. Such precise MPI suppression results are also maintained in real object scenes. Specifically, the MAE of MPI error in an object scene with a sharp corner is reduced to 2.8 mm, which is extremely low compared to previous works.