9 unstable releases (3 breaking)

0.4.0 Sep 25, 2022
0.3.0 Sep 15, 2022
0.2.6 Oct 7, 2020
0.2.5 Sep 1, 2020
0.1.0 Aug 24, 2020

#723 in Machine learning

47 downloads per month
Used in 3 crates (2 directly)

MIT license

4.5MB
82K SLoC

C 37K SLoC // 0.2% comments Assembly 22K SLoC // 0.2% comments GNU Style Assembly 9K SLoC // 0.1% comments Bitbake 5K SLoC C++ 4.5K SLoC // 0.1% comments Java 3K SLoC // 0.4% comments JavaScript 1K SLoC // 0.0% comments Objective-C++ 1K SLoC // 0.2% comments Rust 51 SLoC // 0.1% comments Prolog 26 SLoC Batch 23 SLoC Shell 6 SLoC // 0.2% comments

OpenPNP Logo

openpnp-capture

OpenPnP Capture is a cross platform video capture library with a focus on machine vision. It's goals are:

  • Native camera access on Windows, Linux and Mac. Implemented with DirectShow, V4L2 and AVFoundation respectively.
  • Named device enumeration.
  • Strong, repeatable unique IDs.
  • Format control with support for at least YUV and MJPEG. MJPEG will allow the use of multiple USB cameras on a single hub.
  • Auto and manual exposure control.
  • Auto and manual focus control.
  • Simple, common C wrapper for the listed APIs.

Status

Windows

Feature Implemented
Device Enumeration Yes
Capturing Yes
MJPEG formats Yes
YUV formats Yes
Exposure control Yes
Focus control Yes / Untested
Zoom control Yes
Gain control Yes
White balance control Yes
Framerate control No

Linux

Feature Implemented
Device Enumeration Yes
Capturing Yes
MJPEG formats Yes
YUV formats Yes, YUYV/YUV2
Exposure control Yes
Focus control Yes / Untested
Zoom control Yes
Gain control Yes / Untested
White balance control Yes
Framerate control No

OSX

Feature Implemented
Device Enumeration Yes
Capturing Yes
MJPEG formats Yes (dmb1)
YUV formats Yes
Exposure control Yes / Experimental
Focus control Yes / Experimental
Zoom control Yes / Experimental
Gain control Yes / Experimental
White balance control Yes / Experimental
Framerate control No

TODO

  • support for re-enumeration.

Building OpenPnP Capture

Dependencies

  • CMAKE 3.1 or better
  • MAKE (osx, linux)
  • Visual Studio 2013 + NMake or Ninja Build (windows)
  • NASM for building libjpeg-turbo (linux)
  • libgtk-3-dev (linux, test program)

Build instructions (Windows)

Run the batch file 'bootstrap.bat' and choose the desired build system (VisualStudio/nmake or Ninja). Make sure the compiler (Visual Studio) is in the search path.

Go to the build directory and run nmake or ninja to build the library and the test application.

Build instructions (OSX)

Run 'bootstrap_osx.sh'. Run make.

Build instructions (Linux)

Run 'bootstrap_linux.sh'. Run make.

Supporting other platforms

  • Implement all PlatformXXX classes, like in the win or linux directories.
  • PlatformContext handles device and internal frame buffer format enumeration.
  • PlatformStream is responsible for capturing and decoding the camera stream to a 8-bit per channel RGB frame buffer.
  • Statically link external dependencies.

Releases

Releases are built automatically for all supported platforms. See https://github.com/openpnp/openpnp-capture/releases/latest to download the latest.

Platform Notes

MacOS

On MacOS as of 10.15 Camera permission is needed to open the camera. The library will automatically execute the permission request, but an Info.plist is required to exist in the application bundle. An example Info.plist is:

<?xml version="1.0" encoding="UTF-8"?>
<!DOCTYPE plist PUBLIC "-//Apple//DTD PLIST 1.0//EN" "http://www.apple.com/DTDs/PropertyList-1.0.dtd">
<plist version="1.0">
<dict>
	<key>NSCameraUsageDescription</key>
	<string>openpnp-capture needs permission to access the camera to capture images.</string>
</dict>
</plist>

You can reset the camera permissions in MacOS for testing purposes by running tccutil reset Camera.

No runtime deps