Pulse-based machine learning: Adaptive waveform centroid discrimination for LIDAR system

Abstract

Light detecting and ranging (LIDAR) full waveform echo detection is increasingly used benefiting from the gallop of high-speed sampling, nevertheless many waveform centroid algorithms are sensitive to noise and require a preset window width at present. We propose an adaptive waveform centroid algorithm for improving the ranging accuracy and robustness of LIDAR in this paper. The algorithm couples the idea of machine learning and the slope selection to get the peak of the fitting curve (PFC). The required waveform data can be adaptively selected by calculating the slope from pulse data, and then its waveform centroid is measured by pulse-based machine learning algorithm using least square method. The simulation and experiments are performed to show the performance of our algorithm compared to the existing algorithms. We demonstrate the proposed algorithm to achieve an average error of 0.0838 ns with a standard deviation of 0.1289 ns at the SNR of 10 dB, which is significantly more accurate and robust when compared with the double-scale waveform centroid algorithm in the simulation. In actual test experiments, our algorithm also complete superior performance. The proposed waveform centroid algorithm has potential for ranging task in LIDAR.