Abstract
We introduce cilantro, an open-source C++ library for geometric and general-purpose point cloud data processing. The library provides functionality that covers low-level point cloud operations, spatial reasoning, various methods for point cloud segmentation and generic data clustering, flexible algorithms for robust or local geometric alignment, model fitting, as well as powerful visualization tools. To accommodate all kinds of workflows, cilantro is almost fully templated, and most of its generic algorithms operate in arbitrary data dimension. At the same time, the library is easy to use and highly expressive, promoting a clean and concise coding style. cilantro is highly optimized, has a minimal set of external dependencies, and supports rapid development of performant point cloud processing software in a wide variety of contexts.
Introduction
As robotics, autonomous driving, and AR/VR applications expand, the demand for efficient 3D point cloud processing has become critical. Developers have traditionally faced a choice between the comprehensive but complex Point Cloud Library (PCL) and newer, more user-friendly but less feature-rich libraries like Open3D. cilantro bridges this gap, providing a high-performance, open-source C++ framework that is significantly faster than PCL, more versatile than Open3D, and supports data processing in arbitrary dimensions.
Methodology
cilantro is built on modern software engineering principles to ensure both speed and flexibility:
- Template-Based Design: Most algorithms are templated, allowing them to work with various numerical types and, uniquely, arbitrary dimensions beyond standard 3D space.
- Zero-Copy Data Handling: By utilizing a lightweight wrapper around the Eigen linear algebra library (
ConstDataMatrixMap), cilantro operates directly on existing memory buffers, avoiding the performance bottlenecks of data copying. - Modular ICP Framework: The library implements Iterative Closest Point (ICP) using the Curiously Recurring Template Pattern (CRTP), allowing users to swap components like distance metrics without sacrificing execution speed.
- Parallelization: It leverages OpenMP to distribute computationally intensive tasks across multiple CPU cores efficiently.
Experimental Evaluation
The authors evaluated cilantro against PCL and Open3D using two primary datasets:
- BundleFusion (apt0): A large-scale 3D apartment reconstruction consisting of approximately 7.8 million points.
- TUM RGBD (fr1/desk): A standard benchmark for indoor robotic mapping, used to test surface normal estimation and frame-to-frame alignment (ICP).
Key Results
Benchmarks demonstrate that cilantro consistently outperforms its predecessors in both speed and ease of use. In segmentation tasks on the 7.8-million-point apt0 model, cilantro achieved a nearly 5x speedup over PCL, completing the task in 3.65 seconds compared to PCL's 17.34 seconds.
In geometric registration tests, cilantro's point-to-plane alignment was 3.82x faster than Open3D and 14.99x faster than PCL. Surface normal estimation benchmarks also showed cilantro maintaining a consistent lead in execution time across various search methods.
Beyond performance, cilantro's "expressive" API allows for concise code through method chaining (e.g., cloud.downsample().estimateNormals()), drastically reducing the boilerplate code required by older libraries.
Conclusion
cilantro offers a lean, high-performance alternative for point cloud processing. While currently optimized for CPUs, future development may include GPU acceleration and expanded tools for surface reconstruction and non-rigid registration. The library is available under the MIT license on GitHub.
Summary updated