The current system uses the Microsoft Kinect 1 for its human skeletal tracking API and its depth camera capabilities. Unfortunately, the depth cameras do not provide complete depth data for every pixel. The depth data on the Kinect is calculated by looking at the distortions in an IR laser diffraction pattern. What this means is that the actual depth data returned by the camera is only accurate for a small number of points; the rest of the depth pixels are interpolated.
Due to the depth issues, the 3D location of the ball (or other fetch object) cannot be exactly calculated. As other objects in the room are used for determining the relative position of the robot in a room, the robot also suffers from problems calculating exactly where it is. These two things combined make repeatability of retrieving the fetch object lower than desired.
The Microsoft Kinect 2 (aka the XBOX ONE Kinect) uses a time-of-flight technology for depth sensing which returns the actual depth data for every pixel in a frame instead of an interpolated, low resolution mess. Once the Kinect 2 for Windows is released and the SDK is stable, development for LARRY will continue and the code, schematics, and mechanical drawings will be released to the public.
If you have any questions regarding my code or real-time, low overhead object tracking methods, feel free to shoot me an email at fox-ro-bo-tics (at) g-mail (dot) com [remove the dashes and replace "at" and "dot"].
Best regards,
Austin Steingrube
No comments:
Post a Comment