Hi all,
I'm a graduate student at MIT working on autonomous obstacle avoidance using stereo vision on fixed wing platforms. I use a (heavily firmware modified) APM 2.5 as a low level controller, so I've learned a ton from everyone here.
Wanted to share our latest with you guys: https://www.youtube.com/watch?v=_qah8oIzCwk
We have a paper out on pushbroom stereo, which is how the obstacle detection works (onboard at 120fps!)
- Andy
Replies
Great stuff Andy. I think that your method of simplifying the sensory data and analysis by incorporating the movement of the aircraft as a dimensional input is fundamental to creating suitable sensors for small drones. We came to the same conclusion several years ago and have been working on a similar measuring principle using a laser scanner that creates a moving conical detection surface rather than your moving flat image plane. It has a slower update rate (20 fps) but can measure out to 100m with the idea of providing longer reaction times for intelligent navigation, not just obstacle avoidance. Of course the signal processing is much simpler because the range information is directly available and we have tested the ability to track power lines at 60kph from more than 30m away. An active sensor has the singular advantage of working in poor lighting conditions with no concern for shadows or other imaging anomalies. Having now hijacked your post I wanted to let you know that if you are planning to do further work in this area and would like to experiment with laser technology just let me know.