Quote:
Originally Posted by NotInControl
It would be a lot easier, and cheaper to implement a LPS system instead of trying to extrapolate global position via a camera.
The camera would need to be fixed, so mounting it on a quad copter is a no-go if you want accuracy, unless you have some way to track the position of the quadcopter relative to the reference point on the field. A single camera will skew the image so distance will only be if the camera was directly overhead. Plus lighting conditions, and reflective materials unique to each venue will make each site have different behavior.
|
The calculations on the image are not distance, so it doesn't really matter where the camera is as long as it is high enough up. Pardon my crude paint skills:
incase the image doesn't show up:
http://imgur.com/U4V1cxm
Imagine that black rectangle is exactly containing the field. All I would be doing is finding the position of the robot with respect to the black rectangle. So it doesn't really matter where the camera is on top. Yes, it would alter the values some, but not by much. I do agree with the lighting conditions comment, that could be a problem.
As for your other idea, that would be most ideal, but it requires other teams to participate in it. I want to do this project without have to ask other teams to alter their robots or do any extra work. I could easily see your idea be implemented and used to great success, but it requires other teams to play along.
Quote:
Originally Posted by NotInControl
Consider you are programming an autonomous car, you need to track the other cars around you, their position, and velocity, and lets say you want to change lanes, well how do you determine that another car is not switching into that same lane at that moment in time as well. No aprior knowledge can tell you if that car instantly changes course. There is no way to do this with 100% accuracy, unless there is communication between all the cars. If you don't have this communication, the best you can do is predict with some level of certainty less than 100%.
|
It would be something to have access to the inputs of other robots on the field, such as their joystick position(s). At this state though, it would seem that a machine learning algorithm (deep learning) could be used to solve this task (given that each and every team has the exact same inputs for every action, which isn't the case).