Posture Recognition

Keywords: Python, Deep Learning

Introduction

This will train a model which used in my university graduate project "Human Posture Recognition Based on Neural Network in Robot Controlling".
In 2013, I trained the model that using C++ and OpenNN, and I got 84.04% classification accuracy.
Now I train this model again that using Python, Keras, and Theano as backend, then I get a better classification accuracy which is 99.04%.

Requirement

1. Python 3.5
2. Keras
3. Theano / TensorFlow (as Keras backend)
4. Numpy
5. Pandas

Data

Input :
We got the body coordinate from Kinect, then using Dot Product to find the angle of shoulder and elbow.

• angle of left shoulder
• angle of right shoulder
• angle of left elbow
• angle of right elbow

Output :
classified to 9 classes, example:

• 1 => 1 0 0 0 0 0 0 0 0
• 5 => 0 0 0 0 1 0 0 0 0
• 9 => 0 0 0 0 0 0 0 0 1