This repo shows how to estimate human head pose from videos using TensorFlow and OpenCV.

• TensorFlow 1.4
• OpenCV 3.3
• Python 3

The code is tested under Ubuntu 16.04.

There are three major steps:

1. Face detection. A face detector is adopted to provide a face box containing a human face. Then the face box is expanded and transformed to a square to suit the needs of later steps.

2. Facial landmark detection. A custom trained facial landmark detector based on TensorFlow is responsible for output 68 facial landmarks.

3. Pose estimation. Once we got the 68 facial landmarks, a mutual PnP algorithms is adopted to calculate the pose.

• The marks is detected frame by frame, which result in small variance between adjacent frames. This makes the pose unstaible. A Kalman filter is used to solve this problem, you can draw the original pose to observe the difference.

• The 3D model of face comes from OpenFace, you can find the original file here.

• The build in face detector comes from OpenCV. https://github.com/opencv/opencv/tree/master/samples/dnn/face_detector

The code is licensed under the MIT license. However, the pre-trained TensorFlow model file is trained with various public data sets which have their own licenses. Please refer to them before using this code.

• 300-W: https://ibug.doc.ic.ac.uk/resources/300-W/
• 300-VW: https://ibug.doc.ic.ac.uk/resources/300-VW/
• LFPW: https://neerajkumar.org/databases/lfpw/
• HELEN: http://www.ifp.illinois.edu/~vuongle2/helen/
• AFW: https://www.ics.uci.edu/~xzhu/face/
• IBUG: https://ibug.doc.ic.ac.uk/resources/facial-point-annotations/

(Currently we are trying to empower embeded devices to be more powerful with deep learning, please get in touch if you are interested and would like to join us.)