KISS-Matcher: Fast and Robust Point Cloud Registration Revisited (2409.15615v2)

Abstract: While global point cloud registration systems have advanced significantly in all aspects, many studies have focused on specific components, such as feature extraction, graph-theoretic pruning, or pose solvers. In this paper, we take a holistic view on the registration problem and develop an open-source and versatile C++ library for point cloud registration, called \textit{KISS-Matcher}. KISS-Matcher combines a novel feature detector, \textit{Faster-PFH}, that improves over the classical fast point feature histogram (FPFH). Moreover, it adopts a $k$-core-based graph-theoretic pruning to reduce the time complexity of rejecting outlier correspondences. Finally, it combines these modules in a complete, user-friendly, and ready-to-use pipeline. As verified by extensive experiments, KISS-Matcher has superior scalability and broad applicability, achieving a substantial speed-up compared to state-of-the-art outlier-robust registration pipelines while preserving accuracy. Our code will be available at \href{https://github.com/MIT-SPARK/KISS-Matcher}{\texttt{https://github.com/MIT-SPARK/KISS-Matcher}}.

• The paper introduces KISS-Matcher, a novel pipeline combining enhanced feature extraction (Faster-PFH) and graph-theoretic pruning for fast, robust, and scalable point cloud registration.
• KISS-Matcher employs Faster-PFH for feature extraction, achieving approximately 4.5 times faster performance than FPFH without sacrificing accuracy.
• A k-core based graph-theoretic pruning method is used for efficient outlier rejection, demonstrating scalability and performance comparable to state-of-the-art methods like TEASER++ with significant speed improvements, especially on large datasets.

Overview of "KISS-Matcher: Fast and Robust Point Cloud Registration Revisited"

The paper introduces KISS-Matcher, a novel approach for 3D point cloud registration, leveraging enhanced feature extraction, graph-theoretic pruning, and a comprehensive pipeline for robust, scalable, and fast registration. KISS-Matcher addresses challenges in point cloud registration by integrating efficient feature detection, robust graph-theoretic strategies, and scalability considerations, particularly for robotics and computer vision applications such as mapping, localization, and object pose estimation.

Key Contributions

The KISS-Matcher approach stands out for several reasons:

1. Feature Extraction with Faster-PFH: KISS-Matcher introduces Faster-PFH as an improvement over the classical FPFH descriptor by reducing computational redundancies and incorporating geometric filtering. This results in significant speed gains (approximately 4.5 times faster in single-threaded execution) without sacrificing performance.
2. Graph-Theoretic Pruning: To enhance scalability and efficiency, KISS-Matcher employs $k$-core-based graph-theoretic pruning for outlier rejection. This method offers a linear time complexity improvement over previous maximum clique inlier selection approaches like those in TEASER++. This pruning method achieves substantial runtime improvements, particularly as the scale of data increases.
3. A Comprehensive Pipeline: KISS-Matcher combines the above components into a fully-fledged C++ library, delivering a user-friendly and practical solution that performs well in both scan-level and large-scale map registration scenarios. Experiments demonstrate that KISS-Matcher is on par with state-of-the-art methods and offers scalability and speed benefits.

Numerical and Experimental Results

The paper presents extensive experiments validating the performance of KISS-Matcher. Notably, the system achieves a considerable speed-up compared to TEASER++, with scalability verified on both scan-level and map-level registration tasks using datasets like KITTI and MulRan.

• Quantitative Metrics: In scan-level registration, KISS-Matcher achieves a 100% success rate on the KITTI dataset, with competitive relative translation and rotation errors. Compared to learning-based methods, KISS-Matcher maintains performance across datasets without requiring training, highlighting its robustness and generalizability.
• Scalability: The robustness of the pipeline is demonstrated in various registration scenarios, including large-scale environments where the approach efficiently processes thousands of correspondences.

Implications and Future Directions

The proposed KISS-Matcher pipeline holds significant implications for real-time robotic applications, particularly in areas where speed and robustness in point cloud registration are crucial. The system's design to accommodate both local and global registration tasks makes it versatile for real-world scenarios.

Future work may explore further optimizations in feature extraction and pruning methodologies to extend KISS-Matcher's applicability to even larger cloud datasets and potentially integrate learning-based elements to enhance environment-specific adaptability.

KISS-Matcher represents a significant stride in robust point cloud registration, offering a scalable, efficient, and ready-to-use solution for diverse computer vision tasks. This robustified registration pipeline is poised to impact the development of SLAM systems and other domains where accurate point cloud processing is critical.