Standalone Hand Keypoint Detector

Detects hands keypoints seen on any video or live camera.

Project updated! I wrote a Medium Story making an introduction on the problem of hand recognition, how it's done with OpenPose and why this Standalone Hand Keypoint Detector is needed. Have a read there! https://medium.com/@apofeniaco/into-the-problem-of-hand-recognition-da30797450fe

This hand detector is currently made by integrating two third party tools:

• Handtracking by Victor Dibia - This neural network is used to detect the bounding boxes of hands seen on a image
• OpenPose - The found bounding boxes are fed into OpenPose Hand Keypoint detector.

To run, do just

python run.py

It will detect and display hands found on your webcam stream.

Only left hands are find correctly. This is because the neural network the Handtracking project uses is unable to distinguish if the hand is a left or right hand, and OpenPose needs this information to fully detect the hand keypoints.

Set up

Prerequisites

• OpenPose - Make & install Openpose with the Python module enabled.
• OpenCv
• Tensorflow

Configuration

• config.py: Write there the path where you installed OpenPose
• detection_keypoints.py Make sure python is able to find OpenPose python lib path!

Future improvements

Stay tuned! Following versions coming soon, enabling left and right hand detection and a own Neural Network to make a fully contained project :)