Search papers, labs, and topics across Lattice.
This paper introduces Cross-Coordinate Correspondence Pruning (CCP) to enhance image-to-point cloud registration by addressing the challenges posed by varying point cloud densities. By projecting coarse correspondences into a unified 2D image coordinate system and employing a lightweight pruning network to filter outliers based on inlier confidences, the method effectively balances the trade-off between inlier quantity and outlier reduction. The proposed approach demonstrates a substantial improvement in Registration Recall, outperforming existing state-of-the-art methods by at least 8.6% across multiple benchmarks.
Achieving an 8.6% boost in registration accuracy by intelligently pruning correspondences based on point cloud density could redefine standards in image-to-point cloud alignment.
Recent detection-free approaches have shown significant efficacy in image-to-point cloud (I2P) registration by employing a coarse-to-fine matching pipeline. In the coarse stage, down-sampled image features and voxelized point cloud features are typically fused to establish initial coarse correspondences for subsequent refinement. However, existing methods largely overlook the critical role of point cloud density, which fundamentally dictates the quality of coarse correspondences and the final registration results. Specifically, excessively sparse point clouds lead to an insufficient number of inliers, while overly dense ones often introduce a high outlier ratio. Consequently, this creates an inherent density trade-off, thereby significantly limiting the registration accuracy of current approaches. For mitigating this trade-off, we propose a novel Cross-Coordinate Correspondences Pruning (CCP) strategy to acquire sufficient inliers while ensuring a low outlier ratio. To minimize interference from inter-modal coordinate discrepancies, we first project cross-coordinate coarse correspondences to the 2D image coordinate system for spatial unification. Subsequently, a lightweight pruning network is responsible for predicting the inlier confidences, which are used to filter coarse outliers, from coordinate geometric and modal feature dimensions. To maximize inlier recall, we further design a Multi-Density Point Ensemble (MDPE) strategy that consolidates and deduplicates pruned coarse correspondences across varying point cloud densities. Our method achieves a significant performance improvement, surpassing existing state-of-the-art methods by at least 8.6% in Registration Recall across various benchmarks.
