Mini-Project during Summer 2020
Used a facial landmark detector and PnP algorithm to estimate orientation of head. Takes live video feed and estimates Euler angles representing head orientation.
Motivation: Many applications or games based on physical motion sensing, e.g. Kinect, require the user to be a sufficient distance away from the sensor and in a sufficiently large room. In tighter spaces, such as at a desk setting, head pose estimation offers a much more feasible method of human-computer interaction. The results of this project have be extended to allow users to control application or games in a convenient and intuitive manner.
Credits:
- inspired by PyImageSearch tutorial on facial landmark detection
- using dlib's pre-trained facial landmark detector
- using 3D facial model from https://github.com/yinguobing/head-pose-estimation/blob/master/assets/model.txt