I would like to know how we can get Kinect motions into AnyBody. I’m asking myself if it is possible to connect such device in my computer and make it works in AnyBody really easily.
Also, you said that you have successfully recorded jumps, turns, sit-to-stand, and other full-body motions with the Kinect technology. Do you plan to add examples of these applications in the repository ?
We do of course plan to add these models to the repository. But right now we are experimenting and performing some validation studies to figure out just how well or poorly this new technology works. To make it work, we need several different elements:
- One or several Kinect cameras
- Image processing software
- A stick figure model
- A connection between the stick figure and the anatomical anybody model
- Ways of handling contacts between the model and the environment
For each of these elements we have several different options. Right now we are playing with a system comprising two cameras, Russian image processing software and a file interface based on forward kinematics joint angles. After eliminating a lot of bugs in all of the elements, it seems to work and we have processed several different movements. Recently we did an experiment with parallel recording with Kinect and with a Qualisys mocap system. We are currently processing those trials to be able to assess the accuracy.
So the bottom line is that this is not quite ready for release yet, but we are getting there.
Thank you for your complete answer. It sound very promising, considering the lower price of Kinect compared to other techniques. Did you know that they will release HD versions of the kinect that will be more precise ? You should consider using these HD versions.
I think we can sum up our preliminary conclusions by saying that it is not the accuracy of the camera that is the limiting factor but rather the quality of the subsequent image analysis. This is why we went with the Russian software solution, which distinguishes itself by relying on two cameras rather than one. This allows better identification of limbs and their centroids in the image analysis.
I see. Keep on the good improvements