Rust OpenCV bindings

Experimental Rust bindings for OpenCV 3 and 4.

The API is usable, but unstable and not very battle-tested; use at your own risk.

Quickstart

Make sure the supported OpenCV version (3.2, 3.4 or 4.x) and Clang (part of LLVM, needed for automatic binding generation) are installed in your system.

Update your Cargo.toml

opencv = "0.53"

Import prelude

use opencv::prelude::*;

Getting OpenCV

Linux

You have several options of getting the OpenCV library:

• install it from the repository, make sure to install -dev packages because they contain headers necessary for the crate build (also check that your package contains pkg_config or cmake files).

• build OpenCV manually and set up the following environment variables prior to building the project with opencv crate:

  • PKG_CONFIG_PATH for the location of *.pc files or OpenCV_DIR for the location of *.cmake files
  • LD_LIBRARY_PATH for where to look for the installed *.so files during runtime

Additionally, please make sure to install clang package or its derivative that contains libclang.so and clang binary.

• Gentoo, Fedora: clang
• Debian, Ubuntu: clang and libclang-dev

Windows package

Installing OpenCV is easy through the following sources:

• from chocolatey, also install llvm package, it's required for building:

  choco install llvm opencv
  

  also set OPENCV_LINK_LIBS, OPENCV_LINK_PATHS and OPENCV_INCLUDE_PATHS environment variables (see below for details).

  Also, check the user guides here and here.

• from vcpkg, also install llvm package, necessary for building:

  vcpkg install llvm opencv4[contrib,nonfree]
  

  You most probably want to set environment variable VCPKGRS_DYNAMIC to "1" unless you're specifically targeting a static build.

macOS package

Get OpenCV from homebrew:

• homebrew:

  brew install opencv
  

  You will also need a working C++ compiler and libclang, you can install Command Line Tools (xcode-select --install), XCode (from AppStore) or llvm (from Brew). You most probably need to also check the item 6 of the troubleshooting below.

Manual build

You can of course always compile OpenCV of the version you prefer manually. This is also supported, but it requires some additional configuration.

You need to set up the following environment variables to point to the installed files of your OpenCV build: OPENCV_LINK_LIBS, OPENCV_LINK_PATHS and OPENCV_INCLUDE_PATHS (see below for details).

Troubleshooting

1. One of the common problems is link errors in the end of the build.

   Be sure to set up the relevant environment variables that will allow the linker to find the libraries it needs (see below).

2. You're getting runtime errors like:

   thread 'main' panicked at 'called `Result::unwrap()` on an `Err` value: Error { code: -215, message: "OpenCV(4.2.0) /build/opencv/src/opencv-4.2.0/modules/highgui/src/window.cpp:384: error: (-215:Assertion failed) size.width>0 && size.height>0 in function \'imshow\'\n" }', src/libcore/result.rs:1188:5
   

   thread 'extraction::tests::test_contour_matching' panicked at 'called `Result::unwrap()` on an `Err` value: Error { code: -215, message: "OpenCV(4.1.1) /tmp/opencv-20190908-41416-17rm3ol/opencv-4.1.1/modules/core/src/matrix_wrap.cpp:79: error: (-215:Assertion failed) 0 <= i && i < (int)vv.size() in function \'getMat_\'\n" }', src/libcore/result.rs:1165:5
   

   These errors (note the .cpp source file and Error return value) are coming from OpenCV itself, not from the crate. It means that you're using the OpenCV API incorrectly, e.g. passing incompatible or unexpected arguments. Please refer to the OpenCV documentation for details.

3. You're getting errors that methods don't exist or not implemented for specific structs, but you can see them in the documentation and in the crate source.

   Be sure to import use opencv::prelude::*;. The crate contains a lot of traits that need to be imported first.

4. On Windows, you're getting the (exit code: 0xc0000135, STATUS_DLL_NOT_FOUND) error when running the compiled binary.

   That often means that Windows can't find the OpenCV library dll. Be sure to set up PATH environment variable correctly or copy the dll next to the binary you're trying to run. Check that guide too.

5. On Windows with VCPKG you're getting a lot of linking errors in multiple files like in this issue.

   Unless you're doing a very specific build, you want to have environment variable VCPKGRS_DYNAMIC set to "1".

6. On macOS you're getting the dyld: Library not loaded: @rpath/libclang.dylib error during the build process.

   OS can't find libclang.dylib dynamic library because it resides in a non-standard path, set up the DYLD_FALLBACK_LIBRARY_PATH environment variable to point to the path where libclang.dylib can be found, e.g. for Command Line Tools:

   export DYLD_FALLBACK_LIBRARY_PATH="$(xcode-select --print-path)/usr/lib/"
   

   or XCode:

   export DYLD_FALLBACK_LIBRARY_PATH="$(xcode-select --print-path)/Toolchains/XcodeDefault.xctoolchain/usr/lib/"
   

7. You're getting the panic: a `libclang` shared library is not loaded on this thread.

   Enable the clang-runtime feature. The reason for the issue is that some crates (like bindgen) depend on clang-sys with hard-enabled runtime feature and because of that cargo makes this feature also enabled for every other crate that depends on clang-sys (opencv in this case). During binding generation phase opencv crate tries to use multiple threads and clang-sys with runtime feature enabled doesn't like that (hence the panic). Enabling clang-runtime feature switches to using multiple processes instead of multiple threads. This makes the build a bit longer because of the need to build the helper binary, but the end result is the same.

Reporting issues

If you still have trouble using the crate after going through the Troubleshooting steps please fill free to report it to the bugtracker.

When reporting an issue please state:

1. Operating system
2. The way you installed OpenCV: package, official binary distribution, manual compilation, etc.
3. OpenCV version
4. Attach the full output of the following command from your project directory:

   RUST_BACKTRACE=full cargo build -vv 
   

Environment variables

The following variables must be set when building without pkg_config, cmake or vcpkg. You can set them on any platform, the specified values will override those automatically discovered.

• OPENCV_LINK_LIBS Comma separated list of library names to link to. .lib, .so or .dylib extension is optional. If you specify the ".framework" extension then build script will link a macOS framework instead of plain shared library. E.g. "opencv_world411".

  If this list starts with '+' (plus sign) then the specified items will be appended to whatever the system probe returned. E.g. a value of "+dc1394" will do a system discovery of the OpenCV library and its linked libraries and then will additionally link dc1394 library at the end. Can be useful if the system probe produces a mostly working setup, but has incomplete link list, or the order is wrong (especially important during static linking).

• OPENCV_LINK_PATHS Comma separated list of paths to search for libraries to link. E.g. "C:\tools\opencv\build\x64\vc15\lib". The path list can start with '+', see OPENCV_LINK_LIBS for a detailed explanation (e.g. "+/usr/local/lib").

• OPENCV_INCLUDE_PATHS Comma separated list of paths to search for system include files during compilation. E.g. "C:\tools\opencv\build\include". One of the directories specified therein must contain "opencv2/core/version.hpp" or "core/version.hpp" file, it's used to detect the version of the headers. The path list can start with '+', see OPENCV_LINK_LIBS for a detailed explanation (e.g. "+/opt/cuda/targets/x86_64-linux/include/").

The following variables are rarely used, but you might need them under some circumstances:

• OPENCV_PACKAGE_NAME In some cases you might want to override the pkg-config, cmake or vcpkg package name, you can use this environment variable for that. If you set it pkg-config will expect to find the file with that name and .pc extension in the package directory. Cmake will look for that file with .cmake extension. And vcpkg will use that name to try to find package in packages directory under VCPKG_ROOT. You can also use separate environment variables to set different package names for different package systems:

  • OPENCV_PKGCONFIG_NAME
  • OPENCV_CMAKE_NAME
  • OPENCV_VCPKG_NAME

• OPENCV_CMAKE_BIN Path to cmake binary (used in OpenCV discovery process using cmake). If not set then just "cmake" will be used. For example, you can set something like "/usr/local/bin/cmake" here.

• OPENCV_DISABLE_PROBES Comma separated list of OpenCV package auto-discovery systems to exclude from running. Might be useful if one of the higher priority systems is producing incorrect results. Can contain the following values:

  • environment - reads data only from the OPENCV_LINK_LIBS, OPENCV_LINK_PATHS and OPENCV_INCLUDE_PATHS environment variables
  • pkg_config
  • cmake
  • vcpkg_cmake - like vcpkg, but only uses vcpkg for path discovery, the actual OpenCV probe is done using cmake (cmake related environment variables are applicable with this probe)
  • vcpkg

• OPENCV_MODULE_WHITELIST and OPENCV_MODULE_BLACKLIST Comma separated lists that affect modules that get their bindings generated. Setting whitelist will only generate the specified modules, setting blacklist will exclude the specified modules from generation. If the same module is specified in both list it will be excluded (i.e. blacklist has precedence). E.g. "core,dnn,features2d" .

The following variables affect the building the of the opencv crate, but belong to external components:

• PKG_CONFIG_PATH Where to look for *.pc files see the man pkg-config Path specified here must contain opencv.pc (pre OpenCV 4) or opencv4.pc (OpenCV 4 and later).

• VCPKG_ROOT and VCPKGRS_DYNAMIC The root of vcpkg installation and flag allowing use of *.dll libraries, see the documentation for vcpkg crate

• OpenCV_DIR The directory that contains OpenCV package cmake files. Usually there are OpenCVConfig.cmake, OpenCVConfig-version.cmake and OpenCVModules.cmake in it.

• LD_LIBRARY_PATH On Linux it sets the list of directories to look for the installed *.so files during runtime. Linux documentation has more info. Path specified here must contain libopencv_*.so files.

• DYLD_LIBRARY_PATH and DYLD_FALLBACK_LIBRARY_PATH Similar to LD_LIBRARY_PATH, but for loading *.dylib files on macOS, see man dyld and this SO answer for more info. Path specified here must contain *.dylib files.

• PATH Windows searches for *.dlls in PATH among other places, be sure to set it up, or copy required OpenCV *.dlls next to your binary. Be sure to specify paths in UNIX style (/C/Program Files/Dir) because colon in PATH might be interpreted as the entry separator. Summary here.

• clang crate environment variables See crate's README

Cargo features

• clang-runtime - enables the runtime detection of libclang (runtime feature of clang-sys). Useful as a workaround for when your dependencies (like bindgen) pull in clang-sys with hard runtime feature.
• docs-only - internal usage, for building docs on docs.rs

API details

API Documentation is automatically translated from OpenCV's doxygen docs. Most likely you'll still want to refer to the official OpenCV C++ documentation as well.

OpenCV version support

The following OpenCV versions are supported at the moment:

• 3.2
• 3.4
• 4.x

Minimum rustc version

Generally you should use the latest stable rustc to compile this crate.

Platform support

Currently, the main development and testing of the crate is performed on Linux, but other major platforms are also supported: macOS and Windows.

For some more details please refer to the CI build scripts: Linux OpenCV install, macOS OpenCV install as framework, macOS OpenCV install via brew, Windows OpenCV install via Chocolatey, Windows OpenCV install via vcpkg, Test runner script.

Functionality

Generally the crate tries to only wrap OpenCV API and provide some convenience functions to be able to use it in Rust easier. We try to avoid adding any functionality besides that.

Errors

Most functions return a Result to expose a potential C++ exception. Although some methods like property reads or functions that are marked CV_NOEXCEPT in the OpenCV headers are infallible and return a naked value.

Properties

Properties of OpenCV classes are accessible through setters and getters. Those functions are infallible, they return the value directly instead of Result.

Infallible functions

For infallible functions (like setters) that accept &str values the following logic applies: if a Rust string passed as argument contains null byte then this string will be truncated up to that null byte. So if for example you pass "123\0456" to the setter, the property will be set to "123".

Callbacks

Some API functions accept callbacks, e.g. set_mouse_callback. While currently it's possible to successfully use those functions there are some limitations to keep in mind. Current implementation of callback handling leaks the passed callback argument. That means that the closure used as a callback will never be freed during the lifetime of a program and moreover Drop will not be called for it. There is a plan to implement possibility to be able to free at least some of the closures.

Unsafety

Although the crate tries to provide an ergonomic Rust interface for OpenCV, don't expect Rust safety guarantees at this stage. It's especially true for the borrow-checking and the shared mutable ownership. Notable example would be Mat which is a reference counted object in its essence. You can own a seemingly separate Mat in Rust terms, but it's going to be a mutable reference to the other Mat under the hood. Treat safety of the crate's API as you would treat one of C++, use clone() when needed.

Contrib modules

To be able to use some modules you need to have opencv_contrib installed. You can find the full list of contrib modules here.

Missing modules and functions

While most of the API is covered, for various reasons (that might no longer hold) there are modules and functions that are not yet implemented. If a missing module/function is near and dear to you, please file an issue (or better, open a pull request!).

The binding strategy

This crate works similar to the model of python and java's OpenCV wrappers - it uses libclang to parse the OpenCV C++ headers, generates a C interface to the C++ API, and wraps the C interface in Rust.

All the major modules in the C++ API are merged together in a huge cv:: namespace. We instead made one rust module for each major OpenCV module. So, for example, C++ cv::Mat is opencv::core::Mat in this crate.

The methods and field names have been snake_cased. Methods arguments with default value lose these default values, but they are reported in the API documentation.

Overloaded methods have been mostly manually given different names or automatically renamed to *_1, *_2, etc.

OpenCV 2 support

If you can't use OpenCV 3.x or higher, the (no longer maintained) 0.2.4 version of this crate is known to work with OpenCV 2.4.7.13 (and probably other 2.4 versions). Please refer to the README.md file for that version because the crate has gone through the considerable rewrite since.

Contributor's Guide

The binding generator code lives in a separate crate under binding-generator. During the build phase it creates bindings from the header files and puts them into bindings directory. Those are then transferred to src for the consumption by the crate users.

The crate itself, as imported by users, consists of generated rust code in src committed to the repo. This way, users don't have to handle the code generation overhead in their builds. When developing this crate, you can test changes to the binding generation using cargo build -vv. When changing the binding-generator, be sure to push changes to the generated code!

If you're looking for things to improve be sure to search for todo and fixme labels in the project source, those usually carry the comment of what exactly needs to be fixed.

The license for the original work is MIT.

Special thanks to ttacon for yielding the crate name.