Hi,
This is the gotcha chart and timeline Jeremy and I came up with for the
stereo vision team.
-H
1/17
at the very least:
* stereo cameras setup on vehicle and logging pictures using steve's code (needs cameras and iridium in truck )
shoot for:
* logging time-stamped SVS range images at ~15 Hz
1/29
at the very least:
* logging SVS range images at ~15 Hz
shoot for:
* processing SVS range images into goodness maps of some sort at ~15 Hz (needs state, may need MTA)
Midterm review:
Definitely be processing SVS range images into something
2/14
at the least:
some simple vote generation for arbiter based on goodness maps
afterwards:
Debugging, reliability, and improving performance
Goals:
Deliver a stereo vision system that will guide the DGC Vehicle through 210 miles of DGC course in <10 hrs
Objectives:
- Detect obstacles of height >.5m and radius .25m at 10mph
- Generate votes to avoid those obstacles
Technical Challenges:
- Interfacing SVS Software
- creating software that will translate dense stereo range image output from SVS into goodness maps
- be able to process images and produce votes at rates >15 fps
- don't crash or segfault or have memory leaks
Approach:
- Use Point Grey digital firewire cameras to generate stereo images
- use SVS Software to generate range maps
- write our own frickin' software to coordinate range maps with state data to generate bloody goodness maps and votes
- work loooooong hours until it's done
- Ask JPL people for help