Given my recent job in the computer vision field, I wondered how easy is to create a cheap but powerful motion capture application for use in small animation projects that utilizes cutting edge computer vision algorithms. There are already available commercial (iPi soft) markerless mocap solutions based on MS Kinect, but not for free and producing questionable results. After giving it a thought for a while i excluded kinect as a primary sensor for mocap, because of its very slow capturing capability (30 fps). I was excited to discover that PlayStation3 Eye camera, is quite cheap (25 euros), can deliver up to 125 frames per second and its relative easy to hardware synch multiple cameras. The cameras can work under windows with drivers by Alex Popovich (codelaboratories.com), but with a small fee.
I decided to proceed with PS3 Eye approach and a classical mocap setup (the one with reflective markers), and see how thing goes before using more sophisticated algos. My setup includes two PS3 Eye hard-synched with IR LED modified ring lights and reflective spherical markers mimicking Vicon's kits (a classical mocap solution) on a cheaper version! The cost per camera is about 40 euros (including fee for the driver, the IR LED ring lights and light power unit). The software is still under development. It uses C++ utilizing Alex's drivers for frame-grabbing and CUDA's parallel threads for marker extraction and tracking (segmentation and tracking under development). The code for camera calibration and marker position estimation is ready.
Even such a mocap approach is old and relatively easy in theory, there were some issues. In the hardware domain the ring lights ought to be cheap but cover a lot of volume. The PS3 Eye Field of View (FoV) is about 75 degrees. Concerning software the tracking of multiple identical targets with few cameras is a challenge.
I'll post an update on the project as soon as I have some results!
No comments:
Post a Comment