See README for description of the Dove-eye project (poor man's Hawk-Eye).
The main goal is to make 2D object tracking from a single camera more stable, so that it's reliable for 3D localization.
Minor goal 1 is to extend calibration with definition of custom frame of reference.
Another minor (minor by importance, not difficulty) goal is to modify pipeline and GUI in such a way that it'll allow offline tracking (from video files) and frame-by-frame processing -- this might be prerequisite for making the 2D tracking work properly.
Update after Hackweek 13
- naïve approach to tracking proved to work in simple setups only -> identified need for more sophisticated tracking
- implemented video files tracking
- implemented frame-by-frame processing
Plan for Hackweek 14
Georg Nebehay published two very promising tracking algorithms OpenTLD and CMT. The plan is to compile these two in openSUSE (and learn some of packaging stuff) and see the announced result for myself. The next step is to integrate at least on of these tracking algorithms into Dove-eye. If time permits, it'd be nice to also to implement the minor goal 1 from previous HW.
Update after Hackweek 14
- (not as much time spent as projected)
- OpenTLD packaged for openSUSE distributions
- integrated OpenTLD based tracker into Dove-eye
- Dove-eye is now able to track a pendulum (on a desk, ~0.4 m long) quite reliably, the precision is not so great though (the bounding box changes size, is slightly moved -> noise)
- this seems to be sufficient to move attention to data processing (and then perhaps back to tracking, tick-tock)
This project is part of:
Hack Week 13 Hack Week 14
This project is one of its kind!