f-trak

f-trak is a neural network based face detection program that tracks face movement in screen space. I originally built this as a cool means of controlling a player character in a POC game I made a while back called bongosero. So it's only intended purpose is to report a single bbox back representing the portion of a camera frame containing a face.

Design

f-trak makes use of a pretrained face detection neural network and opencv's Deep Neural Network module to find a face in an image frame captured from a video device.

A prototype was written in python, based on an example by Dr. Adrian Rosebrock, which serves as the design for the current iteration. A sample of this code is provided in this repository.

Set up

f-trak is entirely dependent on the opencv-rust crate. Please follow the set up procedure in their documentation. I found setting up for Linux a painless experience, but Windows is a tiny bit fiddly.

Windows Setup

It's worth noting that when compiling for windows the following environment variables must be set.

OPENCV_DIR "$\opencvLocation\build\x64\vc15\lib"

OPENCV_INCLUDE_PATHS "$\opencvLocation\build\include"

OPENCV_LINK_PATHS "$\opencvLocation\build\x64\vc15\lib"

OPENCV_LINK_LIBS "opencv_world412"

Path "$\opencvLocation\build\x64\vc15\bin"

Other environment variables may be needed as the documentation describes. You'll also need to install llvm, the opencv-rust crate readme documentation explains further.

How to use

f-trak is designed to be run on a separate thread and polled for the current location of a detected face. See the f-trak-test directory for an example application.