Building a Face Detector with OpenCV in C++

Hi everyone! In this blog post, I explain how to build a face detection algorithm with the machine learning components in OpenCV. We will use OpenCV to read an image from a camera and detect faces in it.

You can find all code for this blog post on GitHub.

Installing OpenCV

We use parts of OpenCV and the OpenCV_contrib module. The most convenient way to make sure you have access to these modules is building OpenCV from source. I used OpenCV version 4.2.0 on Ubuntu 16.04. For convenience, I included a bash script that installs the correct OpenCV version and all necessary dependencies in the accompanying repository.

The cv::dnn::Net class we use was added to OpenCV in version 3.4.10, so earlier versions might also work, but I did not test this.

CMake setup

We build the code with CMake. Create a CMake project with a single executable and set the C++ standard to 14. Then find the OpenCV package and link the executable against it.

The full CMake setup is available in the repository:

• Initial CMakeLists.txt
• CMake version with FaceDetector files

Getting an image from the camera

The first thing we need is a camera image. Luckily, the cv::VideoCapture class makes this straightforward.

The flow is:

• create and open a cv::VideoCapture for the first camera
• grab frames into a cv::Mat
• display frames in a loop
• exit on Esc, then close window and release capture

See the full implementation here:

• gettingAnImageFromTheCamera.cpp

We can now display images captured from the camera.

Using cv::dnn::Net to load a pre-trained SSD face detection network

Now we start building the face detector. We use cv::dnn::Net and load weights from a pre-trained Caffe model.

To keep functionality in one place, we create a FaceDetector class with:

std::vector<cv::Rect> detect_face_rectangles(const cv::Mat& frame)

This method takes an input image and returns detected face rectangles.

Header and implementation:

• FaceDetector.h
• FaceDetector.cpp

Model files

Inside the constructor we use cv::dnn::readNetFromCaffe. It needs two files:

• deploy.prototxt
• res10_300x300_ssd_iter_140000_fp16.caffemodel

A practical way to pass these file paths from CMake to C++ is through compile definitions. I originally found this approach in this StackOverflow answer.

Detection pipeline

The detection pipeline in detect_face_rectangles is:

• convert frame to blob with cv::dnn::blobFromImage
• set blob as network input
• run forward pass
• iterate detections
• keep detections above confidence threshold
• convert box coordinates to cv::Rect

For a good explanation of blobFromImage, see this article.

Visualizing detected faces

Because the detector is implemented as a class, visualization is simple:

• create FaceDetector
• call detect_face_rectangles
• draw each rectangle using OpenCV's rectangle function

See the complete example:

• mainFaceDetector.cpp

If we run this, we get a rectangle around Beethoven's face.

Wrap-up

This concludes the post about face detection in OpenCV. We saw how to grab a camera image and detect faces with a pre-trained SSD network.