Head Pose estimation in real-time video

Question

I need to do head pose estimation in live camera video. I have tried with few options like POSIT and shervin Emami's code.

The problem is that I couldnt find any code that uses live video as input. Can anyone suggest good opensource code or methods to find head-pose.

Input: Frames from the camera.

Expected output: Finding the pose of the face.

It would be great if I get an links or codes if any. Also I am planning to do the same application in Android and IOS so like to have few useful links which may help me.

Answer 1

Take a look at OpenPose, it's a real-time multi-person system to jointly detect human body, hand, and facial keypoints (in total 130 keypoints) on single images. Using a CNN, it's a great example of Deep Learning in the field of computer vision.

OpenPose is originally build with Caffe, but ports to TensorFlow are starting to emerge. For example this github repo, is a implementation of OpenPose using TensorFlow. In addition, the new model is bases on Mobilenet, which makes it more fit to run on CPU or low-power embedded devices (like a smartphone for example). BTW, TensorFlow supports both Android and iOS!