3D reconstruction of the dynamic scene with high-speed targets for GM-APD LiDAR. (June 2023)

• Record Type: Journal Article
• Authors: Liu, Di; Sun, Jianfeng; Lu, Wei; Li, Sining; Zhou, Xin
• Is Part Of: Optics & laser technology. Volume 161(2023)
• Publication Date: 2023-06
• Subjects: GM-APD -- LiDAR -- 3D reconstruction -- Dynamic scene -- Moving target -- Motion feature extraction; Optics -- Periodicals; Lasers -- Periodicals; Electronic journals; 621.366
• DOI: 10.1016/j.optlastec.2023.109114
• Languages: English
• ISSNs: 0030-3992

Abstract

Highlights: We propose a reconstruction algorithm for dynamic scenes, in which we split and correct the data based on the extracted motion features to eliminate the effects of background noise and motion. In this scheme, super-resolution reconfiguration of high-speed targets can be achieved readily with sufficient data acquisition. Detection, motion feature extraction, and position prediction of the target can be completed simultaneously as scene reconstruction with the proposed method. Abstract: With the advantages of high distance resolution, long detection distance, small size, and low power consumption, Geiger-mode avalanche photodiode (GM-APD) light detection and ranging (LiDAR) has excellent potential for applications such as three-dimensional earth mapping and autonomous driving. The reconstruction for GM-APD LiDAR is based on the statistics of multiple-laser-pulse data, leading to a long imaging time. There are problems such as blurring when the targets are high-speed, which limits its application scope. A reconstruction algorithm of the dynamic scene for GM-APD LiDAR is proposed in this paper to address this problem. Firstly, the motion features (such as velocity and position) of the targets in the scene are extracted by applying the Hough transform and used as a basis to isolate the targets' echo from the background noise, significantly reducing the noise interference. With these features, the data are corrected for the spatial location to attenuate or eliminate the effects caused by the targets' motion. Finally, the reconstruction is completed utilizing parameter estimation. Also, we discuss the super-resolution reconstruction capability of this algorithm with sufficient data. Reconstruction results of the scene with targets moving at 100–300 m/s are demonstrated at last. Compared to the conventional algorithms, the peak signal-to-noise ratio (PSNR) is improved by 3–4 dB, and the root-mean-square error (RMSE) of distance is improved by 6–10 times. In addition to a super-resolution reconstruction of the dynamic scene, this method also enables the detection, motion feature extraction, and position prediction of high-speed moving targets in the scene, significantly expanding the application scope of GM-APD LiDAR and having good practical application value.