A High-Speed LiDAR Online Multiecho Ranging Method Based on Module-Level Pipelining Acceleration Gauss-Newton

In emergency rescue and autonomous driving, full-waveform light detection and ranging (FW-LiDAR) acquires multilevel 3-D imaging by decomposing the multiple echoes scattered from targets. To address the time-consuming issue of real-time multitarget detection, a Gauss-Newton-based multiecho online (GNMO) ranging method is proposed and implemented based on module-level pipelining acceleration. The relationship between the throughput rate of GNMO ranging method and hardware resource consumption of pipelining is established for providing an acceleration design foundation. The Gauss-Newton iteration containing four functional modules is optimized by inner module parallel pipeline and intermodule pipeline. In the inner module parallel pipeline, a matrix sparsification method with division-free Gauss-Jordan elimination solver is adopted to greatly reduce the input-to-output latency time of each module. The experimental results show that the maximum ranging rate (MRR) of three-subecho online ranging achieves up to 637.8 kHz, approximately 2.7 times as fast as the state-of-the-art performance in previous reports. The ranging standard deviation (RStD) achieves 3.1 mm@49.6 dB and 23.9 mm@18.4 dB; the absolute of mean ranging error (MRE) is lower than 5.0 mm. The imaging results revealed that the FW-LiDAR implemented by the proposed online ranging method can provide more uniformly distributed points and more dense structural characteristics of trees.