A Lightweight RGB-guided Depth Completion Neural Network and Microlens Array Are Features of This 200 mW SPAD-based SoC-based 256 × 256 LiDAR Imaging System

Lila Crosswell^*
*Correspondence: Lila Crosswell, Department of Electrical Electronic and Computer Engineering, University of Catania, 95125 Catania, Italy, Email:

Introduction

This article introduces a cutting-edge LiDAR (Light Detection and Ranging) imaging system, integrating a lightweight RGB-guided depth completion neural network with a low-power Single-Photon Avalanche Diode (SPAD)-based System-on-Chip (SoC). The system operates at a minimal power consumption of 200 mW and achieves a resolution of 256 × 256. The synergy of this setup includes the utilization of a microlens array, enabling enhanced spatial information gathering [1]. This review covers the architecture, functioning principles, and key features of this innovative LiDAR system, emphasizing its advancements in depth completion, power efficiency, and image resolution. Advancements in LiDAR technology have revolutionized the field of 3D imaging and spatial mapping, playing a pivotal role in various industries such as automotive, robotics, and environmental monitoring. This article presents a state-of-the-art LiDAR imaging system characterized by its innovative features: a lightweight RGB-guided depth completion neural network and a Single- Photon Avalanche Diode (SPAD)-based System-on-Chip (SoC) operating at an ultra-low power of 200 mW while achieving a high-resolution output of 256 × 256 [2].

Description

The cornerstone of this LiDAR system is the SPAD-based SoC, leveraging the efficiency of Single-Photon Avalanche Diodes for high-performance depth sensing. The SPAD technology enables precise detection of individual photons, facilitating accurate distance measurement and enabling high-speed operation at minimal power consumption. A significant innovation within this LiDAR system is the integration of a neural network that utilizes RGB data to guide and enhance the depth completion process. This lightweight neural network efficiently fuses RGB information with LiDAR depth data, improving the accuracy and completeness of the 3D point cloud reconstruction [3].

The system incorporates a microlens array that optimizes light collection, enabling the acquisition of enhanced spatial information. This array complements the LiDAR's capabilities by improving the resolution and fidelity of the captured 3D images. Achieving a resolution of 256 × 256 while operating at a power consumption of merely 200 mW signifies a remarkable advancement in LiDAR technology. This exceptional power efficiency coupled with high resolution expands the applicability of LiDAR systems in resourceconstrained environments and battery-powered devices. The advancements in resolution, efficiency, and accuracy offered by this LiDAR system hold immense potential for autonomous vehicles, robotics, and navigation systems. The ability to provide high-resolution, detailed 3D maps in real-time contributes significantly to safety and precision in such applications [4].

The incorporation of an RGB-guided depth completion neural network enhances the accuracy and completeness of depth information. This neural network efficiently utilizes RGB data to refine and supplement the LiDAR depth map, enabling more detailed and accurate 3D reconstructions. The integration of a microlens array optimizes light collection, improving the spatial resolution and fidelity of the LiDAR imaging system. This feature enhances the system's ability to capture fine details and subtle nuances in the environment, facilitating more comprehensive 3D mapping [5].

Conclusion

The increased resolution and accuracy of LiDAR imaging enable better environmental monitoring and mapping. Applications in forestry, urban planning, and disaster management benefit from the system's ability to capture detailed spatial information efficiently. The integration of a lightweight RGBguided depth completion neural network with a low-power SPAD-based SoC in a high-resolution LiDAR system marks a significant leap forward in 3D imaging technology. The synergy of these components enables exceptional power efficiency, enhanced depth completion, and improved spatial information gathering. The applications span various industries, from autonomous vehicles to environmental monitoring, promising safer, more efficient, and detailed spatial mapping and navigation capabilities in the near future. Continued research and advancements in LiDAR technology are anticipated to further expand the horizons of applications and usability in diverse domains.

Acknowledgement

None.

Conflict of Interest

None.

References

1. Yuan, Wei, Cheng Xu, Li Xue and Hui Pang, et al. "Integrated double-sided random microlens array used for laser beam homogenization." Micromachines 12 (2021): 673.

   Google Scholar, Crossref, Indexed at

2. Xue, Li, Yingfei Pang, Wenjing Liu and Liwei Liu, et al. "Fabrication of random microlens array for laser beam homogenization with high efficiency." Micromachines 11 (2020): 338.

   Google Scholar, Crossref, Indexed at

3. Silberman, Nathan, Derek Hoiem, Pushmeet Kohli and Rob Fergus. "Indoor segmentation and support inference from rgbd images." Proceedings, Part V 12, Springer Berlin Heidelberg (2012): 746-760.

   Google Scholar, Crossref, Indexed at

4. Jin, Yuhua, Ali Hassan and Yijian Jiang. "Freeform microlens array homogenizer for excimer laser beam shaping." Opt Express 24 (2016): 24846-24858.

   Google Scholar, Crossref, Indexed at

5. Liu, Lina, Yiyi Liao, Yue Wang and Andreas Geiger, et al. "Learning steering kernels for guided depth completion." IEEE Trans Image Process 30 (2021): 2850-2861.

   Google Scholar, Crossref, Indexed at