Super LiDAR Intensity for Robotic Perception

TL;DR

Experimental results validate the efficacy of the proposed approach, which successfully integrates computer vision techniques with LiDAR data processing, enhancing the applicability of low-cost LiDAR systems and establishing a novel paradigm for robotic vision via active optical sensing–LiDAR as a Camera.

Abstract

Conventionally, human intuition defines vision as a modality of passive optical sensing, relying on ambient light to perceive the environment. However, active optical sensing, which involves emitting and receiving signals, offers unique advantages by capturing both radiometric and geometric properties of the environment, independent of external illumination conditions. This work focuses on advancing active optical sensing using Light Detection and Ranging (LiDAR), which captures intensity data, enabling the estimation of surface reflectance that remains invariant under varying illumination. Such properties are crucial for robotic perception tasks, including detection, recognition, segmentation, and Simultaneous Localization and Mapping (SLAM). A key challenge with low-cost LiDARs lies in the sparsity of scan data, which limits their broader application. To address this limitation, this work introduces an innovative framework for generating dense LiDAR intensity images from sparse data, leveraging the unique attributes of non-repeating scanning LiDAR (NRS-LiDAR). We tackle critical challenges, including intensity calibration and the transition from static to dynamic scene domains, facilitating the reconstruction of dense intensity images in real-world settings. The key contributions of this work include a comprehensive dataset for LiDAR intensity image densification, a densification network tailored for NRS-LiDAR, and diverse applications such as loop closure and traffic lane detection using the generated dense intensity images. Experimental results validate the efficacy of the proposed approach, which successfully integrates computer vision techniques with LiDAR data processing, enhancing the applicability of low-cost LiDAR systems and establishing a novel paradigm for robotic vision via active optical sensing--LiDAR as a Camera.

Figures (19)

• Figure 1: Comparison between the intensity image densified by the proposed method and the true intensity images stitched from high-line LiDAR data. (a) Raw NRS-LiDAR input. (b) Densified 1380 $\times$ 240 intensity image using the proposed method. (c) Stitched 1024 $\times$ 128 intensity images using the high-cost Ouster 128-line LiDAR data from pfreundschuh2024coinshan2021robust. The proposed method effectively eliminates line artifacts and mitigates overexposure in challenging scenarios, generating high-definition, camera-grade intensity images.
• Figure 2: Illustration of the non-repeating scanning mechanism. When stationary, the LiDAR gradually produces denser point clouds over time within the same FoV, allowing the intensity image derived from the point cloud projection to achieve higher density as well. The color of point clouds or gray-scale images represents the intensity value.
• Figure 3: System diagram: During stationarity, the robot utilizes NRS-LiDAR to obtain sparse and dense LiDAR data. Sparse data is augmented to simulate motion characteristics, while dense data serves as supervision to train an intensity image densification network for motion perception tasks.
• Figure 4: Intensity images obtained by virtual-camera projection mode (top) and panoramic mode (bottom), respectively.
• Figure 5: Comparison of intensity images fused from 5 scans in stationary and in-motion cases. The image shows the corresponding intensity images. In the stationary case, the 5-scan inputs are aligned, whereas the in-motion case reveals irregularities and misalignment due to motions.