Skip to content

FesianXu/KinectGestureControl

Repository files navigation

kinectLOGO

Introduce


Kinect Gesture Control System(KGCS) Demo is used for Kinect gesture control in embedded environment, including a WPF main program and a model based on TensorFlow(have not opened source now). If you have advise, please contact me with:

e-mail: FesianXu@163.com QQ: 973926198


Prerequisites you need

You need several prerequisites knowledge to access this project including:

  1. The basic knowledge about Computer Vision
  2. The basic knowledge about Deep Learning and Machine Learning
  3. The basic knowledge about C# WPF Programming
  4. The basic usage about python3.0 and TensorFlow 1.3
  5. The basic usage about Git of course.

Also the project need some development environment including:

  1. Kinect 1.0 SDK and Kinect Studio(additional)
  2. TensorFlowSharp(I suggest you to use NuGet to install it and you should use Visual Studio 2015 and later version in order to avoid the confusing mistakes in the installing)
  3. TensorFlow in python interface
  4. It is better if you install some serial port driver and monitor like PL2303 and CH340, you will meet the need that communicating with the MCU

Version Review


Version 1.0

Date: 2017/8/17

Functions:

  • A Basic Kinect Skeletons drawing.
  • A Basic color hands tracking, drawing the color both hands images.
  • A Basic CNN model for palm and fist recognition.
  • A Basic Voice Assistant Chris, who can manage the those task like:
    1. log in and log out, identity authorization.
    2. run and stop the kinect.
    3. refuse the limited authorization operation.
    4. Voice Services
  • Serial Communication with Pixhawk to control the model car.
  • single hand driving.

Bugs:

  • Left hand in hold single hand driving mode is invalid and in bug.
  • have a problem in serial communication receive.
  • have to manully initiate the RC by sending each channels values.

TODO list:

  • add a more efficient hand gesture recognition model than the single spatial CNN model, which is both time-consuming and in high-error rate. May be some model like LSTM-CNN attention based model is ideal.
  • rewrite the main thread code structure, it is mess now.
  • add more functions to Chris.
  • add 'forward-speed up' gesture command.
  • add the function of distinguishing the operating person and the passer-by.

version1.0main


About

Kinect Gesture Control System Demo

Resources

Stars

Watchers

Forks

Packages

No packages published

Languages