Skip to contents

parallax

Multi-platform point cloud analysis for R — registration, segmentation, surface reconstruction, and feature extraction for aerial, terrestrial, and UAV LiDAR.

R has lidR for aerial LiDAR processing and it’s excellent — but it was designed for overhead perspective at moderate point densities. Throw a terrestrial laser scan at 50,000 pts/m² at it and things get awkward. Give it a UAV photogrammetric point cloud with oblique angles and variable density and you’re fighting the assumptions. There’s no R package that treats acquisition platform as a first-class concept, where ground classification, segmentation, and reconstruction automatically adapt their parameters based on whether you’re working with aerial, terrestrial, or UAV data.

parallax fills that gap. One API, platform-aware defaults, Rust-powered performance where it counts, pipe-friendly outputs that work naturally with the tidyverse.

Documentation is published at https://chrislyonsky.github.io/parallax/.

Install 

# install.packages("pak")
pak::pak("chrislyonsKY/parallax")

Requires a Rust toolchain for building from source. Binary packages (when available from CRAN or R-universe) don’t require Rust.

Quick start 

library(parallax)

# Use the installed demo files for a quick tour
aerial_path <- system.file("extdata", "aerial-demo.csv", package = "parallax")
source_path <- system.file("extdata", "registration-source.csv", package = "parallax")
target_path <- system.file("extdata", "registration-target.csv", package = "parallax")

aerial <- px_read(aerial_path, platform = "aerial")
source <- px_read(source_path, platform = "terrestrial")
target <- px_read(target_path, platform = "terrestrial")

# Filtering and segmentation
filtered <- px_filter_statistical(aerial, k = 6, std_ratio = 1.5)
ground <- px_classify_ground(filtered)
trees <- px_segment_trees(filtered, min_height = 2.0, crown_threshold = 0.75)
metrics <- px_metrics(filtered)

# Registration
icp <- px_icp(source = source, target = target, max_correspondence_distance = 2.0)

# Surface reconstruction
target <- px_estimate_normals(target, k_neighbors = 4L)
mesh <- px_poisson_reconstruct(target, depth = 4L)

The same files back the package vignettes in vignettes/getting-started.Rmd and vignettes/platform-aware-workflows.Rmd.

Platform-aware types

Platform Class Typical density Default ground params
Aerial px_aerial 2–25 pts/m² CSF 2.0m cloth, nadir perspective
Terrestrial px_terrestrial 1,000–100,000 pts/m² CSF 0.5m cloth, oblique-aware
UAV px_uav 20–500 pts/m² CSF 1.5m cloth, variable density

Features

Registration — ICP (point-to-point, point-to-plane), coarse alignment (FPFH + RANSAC), multi-scan px_align_scans()

Segmentation — ground classification (CSF, PMF, platform-aware), object segmentation (DBSCAN, region growing), tree delineation (watershed, Li2012)

Surface reconstruction — Poisson (watertight), Ball Pivoting, alpha shapes

Feature extraction — RANSAC plane/cylinder detection, edge detection

Filtering — voxel downsample, statistical outlier removal, radius outlier, crop

Metrics — summary statistics, canopy height model, point density maps

Examples and articles

Architecture

Rust core (via extendr) for performance-critical operations. R layer for the type system, pipe-friendly API, and platform-aware dispatch. All px_* functions accept any px_cloud subtype.

lidR interop 

las <- to_lidr(cloud)           # parallax → lidR
cloud <- from_lidr(las)         # lidR → parallax

Companion: occulus (Python)

occulus provides the same capabilities with Python-native idioms. C++ core via pybind11. Same concepts, same algorithm names, independent implementation.

parallax (R) occulus (Python)
Core Rust via extendr C++ via pybind11
Returns S3/S7 classes, tibbles NumPy arrays
Interop lidR, sf, terra Open3D, laspy
Registry CRAN PyPI

Development 

# Clone and install from source
git clone https://github.com/chrislyonsKY/parallax.git
cd parallax
# In R:
devtools::install()
devtools::test()
devtools::check()
pkgdown::build_site()

License

GPL-3.0 — see LICENSE.

Author

Chris Lyons — GIS Developer, Kentucky Energy & Environment Cabinet - GitHub: @chrislyonsKY - Email: