Search papers, labs, and topics across Lattice.
This paper introduces a novel method for Image-to-Point Cloud Registration (I2P) that utilizes Conditional Rectified Flow to convert sparse LiDAR scans into dense intensity images, facilitating effective matching with camera images. By employing a self-supervised image completion task for pre-training and fine-tuning on minimal LiDAR data, the approach achieves high accuracy and generalization across various sensor configurations without requiring extensive labeled datasets. Experimental results on the R3LIVE dataset demonstrate a mean error of 4.89掳 / 1.63 m and a registration speed of approximately 0.68 seconds, significantly outperforming existing techniques.
Achieving a mean error of just 4.89掳 / 1.63 m, this method redefines the efficiency and accuracy of integrating LiDAR and camera data for autonomous systems.
Image-to-Point Cloud Registration (I2P) is essential for integrating camera and LiDAR in perception and autonomous systems, yet the modality gap between images and point clouds makes it difficult to achieve both high accuracy and strong generalization. In this paper, we propose a simple yet effective I2P method that treats LiDAR as an imaging sensor: from a single sparse LiDAR scan, we generate a dense LiDAR intensity image using Conditional Rectified Flow, match it with a camera image using a pre-trained feature matcher, and estimate the 6-DoF relative pose via PnP-RANSAC. The proposed model is pre-trained through a self-supervised image completion task and fine-tuned on a small amount of LiDAR data (neither image-point cloud pairs nor ground-truth sensor poses are required), enabling it to scale to diverse LiDAR and camera configurations. Experiments on the R3LIVE dataset show that the proposed method achieves a mean error of 4.89{\deg} / 1.63 m, outperforming existing methods, while completing a single registration in approximately 0.68 s.
