GoCV Documentation

GoCV includes support for various OpenCV contrib modules. Their implementation status varies:

Implemented Contrib Modules

• freetype (Drawing UTF-8 strings)
• img_hash (Image hashing algorithms)
• mcc (Macbeth Chart module)
• wechat_qrcode (WeChat QR code detector)
• xobjdetect (Extended object detection)
• xphoto (Additional photo processing algorithms)

Contrib Modules in Progress

• bgsegm (Background-Foreground Segmentation)
• face (Face Recognition)
• tracking (Tracking API)
• xfeatures2d (Extra 2D Features Framework)
• ximgproc (Extended Image Processing)

Other Contrib Modules

Many other modules (e.g., alphamat, barcode, dnn_superres, sfm, text) are listed in the roadmap but are not yet implemented in GoCV.

The GoCV project is progressively implementing OpenCV's CUDA modules. When choosing modules for your project, be aware of which ones are fully implemented, in progress, or not yet started.

Fully Implemented CUDA Modules

• core
• cudabgsegm (Background Segmentation)
• cudaobjdetect (Object Detection)
• cudawarping (Image Warping)

CUDA Modules in Progress

• cudaarithm (Operations on Matrices)
• cudacodec (Video Encoding/Decoding)
• cudafeatures2d (Feature Detection and Description)
• cudafilters (Image Filtering)
• cudaimgproc (Image Processing)
• cudaoptflow (Optical Flow)
• cudastereo (Stereo Correspondence)

Note: Check the specific function implementation status in the roadmap if you require a particular mathematical or processing operation.

Common computer vision algorithms are implemented in these utilities:

• Faceblur: Detects faces using CascadeClassifier and applies a Gaussian blur.
• Facedetect: Detects faces using CascadeClassifier and draws rectangles around them.
• Facedetect from URL: Detects faces in an image downloaded from a URL.
• Counter: Counts objects crossing a user-defined vertical or horizontal line in a video file.
• Hand Gestures: Counts fingers by analyzing convexity defects.
• Motion Detection: Processes a video feed to detect motion.
• Tracking: Uses the TrackerMOSSE algorithm from OpenCV Contrib to track a user-selected region of interest.

GoCV provides a Makefile target to simplify compiling OpenCV with CUDA support.

To compile with shared libraries:

make install_cuda

To compile with static OpenCV libraries:

make install_cuda BUILD_SHARED_LIBS=OFF

After compilation, verify the installation by running the CUDA test command:

cd $GOPATH/src/gocv.io/x/gocv
go run ./cmd/cuda/main.go

Successful verification will output the GoCV version and GPU device information (e.g., Device 0: "GeForce MX150" ...).

To use patented or non-free algorithms (like SIFT) that are part of the OpenCV Contrib modules, you must build OpenCV with the OPENCV_ENABLE_NONFREE=ON flag. If you are using the provided build scripts, you can run the following command:

make build_nonfree

Several utilities demonstrate how to interact with webcams and video files:

• ASCIIcam: Captures video from a webcam and displays it in the terminal using ASCII characters.
• Capture test: Verifies that video capture from a connected webcam is working correctly.
• Capture window: Captures video from a webcam and displays it in a standard GUI window.
• MJPEG Streamer: Opens a video capture device and streams the feed via MJPEG, viewable in any web browser.
• Save Image: Captures a single frame from a webcam and saves it to a file.
• Save Video: Captures 100 frames from a camera and saves them to a video file.
• Show Image: Opens an image file from disk and displays it in a window.

When using go run, go build, or go test with OpenVINO, you must include the -tags customenv flag. This ensures the CGo compiler uses the correct environment settings provided by the OpenVINO setup scripts instead of the default settings.

You can verify the installation by running the version command example. A successful setup will show opencv lib version containing -openvino.

To use GoCV with CUDA, you must install the NVIDIA CUDA toolkit and cuDNN libraries on your host system:

1. CUDA Toolkit: Download from NVIDIA CUDA Downloads.
2. cuDNN: Download the appropriate runtime and developer libraries from the cuDNN Archive (e.g., 'cuDNN Runtime Library' and 'cuDNN Developer Library' for your specific OS version).

GoCV provides support for the Intel OpenVINO Inference Engine via the gocv.io/x/gocv/openvino/ie package. This allows for executing convolutional neural networks using OpenVINO libraries.

Requirements

To use OpenVINO with GoCV, you must have Intel OpenVINO version 2019 R3 or higher installed.