Damped Gauss-Newton based online ranging for point extraction from low SNR and high overlapping waveforms

Abstract

Extracting more valid points from the low SNR and high overlapping waveforms is a difficult task for full-waveform LiDAR using online waveform decomposition method. To solve this problem, a damped Gauss-Newton multi-echo online waveform decomposition algorithm (DGNM) implemented on the ZYNQ-7000 FPGA is proposed. In the implementation of DGNM, the sparse Jacobi matrix and the division-free Gauss-Jorden methods are advanced to accelerate the solving of the iteration step. The inexact line search is utilized to quickly obtain a reasonable damped factor to further modify the iteration step for a robust convergence. For the three-target experiments, the valid point extraction ratio of DGNM is six times higher than Gauss-Newton multi-echo decomposition (GNM) at an SNR of 18.44 dB, but maintaining the ranging accuracy at the same magnitude. It implies that DGNM can increase the valid points for enhancing the density of the 3D point cloud.