Using JM-L To Develop Gesture Recognition for Kinect

edited December 2013 in Kinect

Hi,

I have been working on a project trying to develop a gesture recognition system based on kinect joint movements. I tried making my own tools but efficiency is very important with these applications and my algorithm is not fast enough. As you know kinect gives you x,y,z data for 15 points and what I want to do is monitor movements of the joints over a set number of frames (assuming the movement is meaningful and forms a gesture of some kind), then use this data as training input to a classifier (using Java Machine Learning 0.1.7). After training, I want to use the classifier to classify new movements. What is the best way to organize my data (Im thinking of storing training data in a csv), meaning how do I divide my datasets (per frame, per joint or some other pattern) for maximum efficiency?

Thanks a lot

Answers

  • I don't think there will be people helping you, as it is a difficult topic and algorithms that do so are still in development. I'm going to have the same problem as you in the next couple of months, so maybe we can share our conclusions when we found one. I was thinking in using OpenCV to help me in gesture recognition, but I'm still making up my mind.

  • I don't think there will be people helping you, as it is a difficult topic and algorithms that do so are still in development. I'm going to have the same problem as you in the next couple of months, so maybe we can share our conclusions when we found one. I was thinking in using OpenCV (not the ope

  • Hi,

    Thanks for the reply, this project was discontinued unfortunately. I would suggest to anyone that is going to use OpenCV and processing to reconsider and use matlab instead until they are happy with their prototype. Also, don't try to over complicate things; only move on to more complex algorithms such as machine learning techniques after you've completely exhausted your current resources.

    Best of luck!!

Sign In or Register to comment.