Risley-prism-based multi-beam scanning LiDAR for high-resolution three-dimensional imaging

Highlights

• Coherent light detection and ranging architecture for three-dimensional imaging.

• Risley-prism-based multi-beam scanning enables high-resolution spatial sampling.

• Multi-mode integrated point cloud filtering method for efficient outlier removal.

• Coarse-fine coupled alignment strategy for multi-channel point cloud registration.

Abstract

Light detection and ranging (LiDAR) as a significant approach to three-dimensional perception has increasingly suffered from challenges for compact structure, high resolution and strong adaptability. This paper presents a versatile LiDAR system that incorporates multi-beam scanning using Risley prisms and coherent detection based on triangular frequency modulation. By combining Risley-prism-based multi-beam steering model with simultaneous distance and velocity measurements, the LiDAR architecture for three-dimensional imaging is theoretically demonstrated. A unified LiDAR point cloud processing framework is developed, where the multi-mode integrated filtering method is proposed for outlier removal using point cloud characteristics. The coarse-fine coupled strategy is also formulated to combine statistical modeling with iterative optimization for multi-channel point cloud registration. It is experimentally validated that our LiDAR can achieve high-resolution three-dimensional information acquisition against long range, and the proposed point cloud processing technique can fully reconstruct the spatial form of object while preserving sufficient details.

Introduction

Three-dimensional (3D) imaging has garnered emerging interest as a technology to recover the surrounding environment from the signals acquired with various sensors or instruments, which can significantly contribute to object identification and situational perception. In contrast with passive methods represented by stereo vision, light detection and ranging (LiDAR) offers an essential approach to active 3D imaging. Due to its potential to achieve wide coverage, high resolution and ambient immunity [1], LiDAR has found extensive applications in many fields, such as remote sensing, topographic mapping and autonomous driving [2], [3], [4].

The existing LiDAR systems normally fall under two categories, in which one relies on flash illumination and the other requires beam scanning. Based on time of flight (TOF) ranging with large-area light pulses [5], the flash LiDAR usually necessitates a high peak laser power and high-performance detectors such as avalanche photodiodes and charge-coupled devices [6], [7], [8]. Comparatively, the scanning LiDAR working in a serial information acquisition mode allows performance enhancement at relatively low cost [9] Most beam scanning methods for 3D imaging LiDAR are based on mature technologies such as opto-mechanical devices or micro-electro-mechanical systems (MEMS). The conventional opto-mechanical LiDARs may incorporate different types of scanning systems to generate specific beam scan patterns, such as rotors, polygonal mirrors, galvanometric scanners or oscillatory scanners [10]. However, such mechanisms have some shortcomings in complex internal structure, high power consumption and limited scan rate [11,12]. The MEMS scanning LiDARs employs deformable mirrors driven by micro actuators for high dynamic response, which are still limited due to the inherent sensitivity to vibration and the angular resolution restricted by aperture size [13,14]. In recent years, solid-state LiDARs with inertia-free designs have attracted increasing attention in industries where reliability and stability are highly desired. One potential solid-state beam steering approach is to use optical phased arrays (OPA), for which the optical phase control and the complicated fabrication process remain challenging [15], [16], [17]. Alternatively, nonmechanical beam steering is made possible through lens-assisted integrated optical switch array [18], time-stretch spectrally scanning with tunable lasers [19], or wideband frequency swept source [20]. But those methods can only achieve beam scanning with a limited range or in one dimension. Differing from the above solutions, Risley-prism beam scanner offers significant advantages in compact structure, conformal aperture, low moment of inertia and great flexibility [21]. There have already been some basic studies about Risley prisms, mostly related to system setup and performance evaluation [22,23], analytical solutions to inverse problem [24,25], and graphical analysis for scan patterns [26]. By producing a wide variety of scan patterns through the coaxial rotation of two wedge prisms, this scanner can provide a multi-mode beam scanning mechanism for 3D imaging LiDAR [21,27].

As for light detection methods used in 3D LiDAR systems, TOF ranging based on direct pulsed laser detection has been dominant in many applications due to matured hardware setting and signal processing methods [[5], [6], [7], [8],11,14,[16], [17], [18], [19],27]. However, the TOF ranging performance is constrained by the limited laser pulse power for safety requirements and the returned signals that may be affected by strong sunlight or laser beams from other systems. There is a growing interest to develop frequency modulated continuous wave (FMCW) ranging based on coherent detection [15,20]. By reducing system complexity and resisting ambient influence in principle, the FMCW ranging is advantageous in high sensitivity, superior resolution and large dynamic range while performing simultaneous distance and velocity measurements.

In our previous work [28], we have proposed a 3D coherent imaging LiDAR that combines Risley prisms for wide-FOR beam scanning with FMCW for high-accuracy ranging, which has been validated in an airborne flight experiment. However, the LiDAR adopted a single-beam scanning scheme which leads to inherent limitations on spatial sampling and 3D sensing. Here we present, to the best of our knowledge, the first demonstration of Risley-prism-based multi-beam scanning (RMS) LiDAR using FMCW coherent detection. By steering multiple beams simultaneously through Risley prisms, the RMS LiDAR can significantly enhance the spatial sampling agility and information acquisition efficiency, thus enabling multi-channel point cloud generation with variable density and flexible resolution. The LiDAR data processing framework is developed to handle these large-scale point clouds, which incorporates a multi-mode integrated filtering method and a coarse-fine coupled registration method. It is intended that the proposed technique can achieve high-quality 3D imaging and reconstruction while balancing between computational efficiency and processing performance.

The remainder of this paper is outlined as follows. In Section 2, the 3D imaging architecture using RMS LiDAR is presented through theoretical demonstration and performance evaluation. In Section 3, the multi-channel data processing framework is developed with an emphasis on multi-mode integrated filtering and coarse-fine coupled registration. In Section 4, the proposed technique is validated by a 3D imaging experiment. The conclusions are drawn in Section 5.