I am working on action recognition from multi-views skeleton images captured using kinects. However, when I acquired skeleton images using three kinects with 3 notebooks the images frame across devices is not quite synchronized. Is there any ways to sync all frames from different devices? Or is there a public available database of synced skeleton coordinates that I can download to perfrom action recognition?
I think the easiest way to solve this is to coregister the three kinect images by calibrating the setting to the same (e.g. real world) coordinate system. This means you have to transform the skeletons of each kinect to the real world coordinate system first. If this is done with sufficient accuracy the registration of the joints coming from the different sensors should be possible by a simple nearest neighbor search.
You could let the three notebooks communicate with a time-server:
http://en.wikipedia.org/wiki/Time_server
The time-server does not need to return the correct time; as long as it returns good relative times, you're fine.
I think that will work much better than some kind of registration.
A time synchronisation is not sufficient, at least not if the viewing angles of the sensors overlap. You need to couple the data of the devices in time as well as in space with the two methods described above (by Lambert and me) if you want to get synced skeleton coordinates.
I disagree, Ralf.
I will not deny that the images from the three sensors need to be combined in order to best represent the 3D scene.
But the question was how to synchronize them. It is more accurate and faster to just 'ask the time' than to infer time from registration results.
If the question is only the time synchronisation, then of course you are right, Lambert. From Somying's problem description, however, I read that she also has problems with the spatial registration of the skeleton coordinates, even though the term "synced" might be misleading in this context.
Thank you very much for your answer. I will first try to come up with a client-server program to lunch the three kinect to make sure that timing is synched. And I think misaligned coordinates or spatial registration problem is always there due to characteristic of the cameras and also the acquisition environment. I probably have to research further how researchers deals with the spatial resolution. I may try to come up an algorithm to choose or to re-estimate the best coordinate among ones recieved from the three kinects. Thank you very much.
- Badges
- Science topic
- Similar topics
- Engineering
- Electrical Engineering
- Signal Processing
- Image Processing