search

ProgrammingRobotsStudyGroup / robo_magellan

Apache License 2.0
4 stars 6 forks source link

Examine video from 04-20 to determine minimum cone area for good cone detection #39

Closed merose closed 6 years ago

merose commented 6 years ago

Several runs which were otherwise successful required executing the recover turn at one or more cones. It's possible that decreasing the minimum cone area will help the robot spot the cone earlier, to better handle GPS drift. The video from 04-20 should be examined to determine what the minimum cone area should be.

Testing on 04-20 used either 400 pixels or 350 pixels as the minimum cone area.

jeremyolsen commented 6 years ago

Here are the 4.19 videos - https://drive.google.com/open?id=15tMdN9OMkXJGFfAKV19_WGj8m0biE3R6

jeremyolsen commented 6 years ago

I'm going to do a test in the morning to determine the distance to the cone when it is first seen based upon the minimum cone area.

merose commented 6 years ago

Modified code in commit aa0ea11d5ed6417c8aeed31e0b6474f6f98909a6 to make configuring the minimum area for normal driving and recovery possible. Also added a configuration to omit cone detections if they are above a configured horizon position.

merose commented 6 years ago

Based on April 20th videos, the cone never had a Y value greater than about 240, about halfway up. However, the terrain was smoother, so there was less bouncing of the front-end. Using a Y threshold of 280-300 should be ample.