OpenCV's Rust-native rival, piece by piece

What it does

Rust CV is a mono-repo of pure-Rust computer vision crates aiming to replicate the core capabilities of OpenCV, OpenMVG, and vSLAM frameworks without FFI dependencies. It bundles algorithms for feature extraction (AKAZE, FAST), camera modeling, descriptor matching, and geometric verification into a cohesive API surface with a “batteries-included” cv meta-crate.

The interesting bit

The project deliberately avoids the machine-learning hype cycle. Almost every algorithm here is classical—Sobel, Canny, Eight-Point, ARRSAC, Lambda Twist P3P—making it a rare modern toolkit for photogrammetry purists. The #[no_std] support where possible is a genuine constraint, not an afterthought.

Key highlights

• Sufficient photogrammetry primitives to run Structure-from-Motion and visual SLAM pipelines
• #[no_std] support in crates where feasible, enabling embedded or WASM targets
• Geometric verification through the sample-consensus abstraction: ARRSAC consensus, Lambda Twist P3P, Eight-Point algorithm
• Multiple matching backends: brute-force for tracking, HGG and HNSW for loop-closure search
• Tutorial book with runnable example code in the repo

Caveats

• Large surface area remains incomplete: no SIFT, no ORB, no RANSAC (only ARRSAC), no fisheye or equirectangular camera models, no PLY/NVM export
• The README’s exhaustive checklist makes clear this is a work-in-progress; image processing and pattern recognition are explicitly called weaker domains
• Some “basically done” algorithms (Nister-Stewenius) exist in archived repos but aren’t packaged

Verdict

Worth tracking if you’re building Rust-native robotics, AR, or photogrammetry tools and want to escape OpenCV’s C++ build complexity. Not yet a drop-in replacement—expect to contribute or work around missing pieces.