 |
 |
Project P1:
Autonomous Helicopter Pose and Landing
Project Goal
In class we will discuss a technique called "Structure From Motion" (SFM)
where a 3D rendering of a scene can be generated purely from observations
of a moving camera and intelligent mathematics. Determining the location
of the camera as it moves is an equivalent problem: "Pose From Motion"
(PFM). In practice, building such a system is extremely challenging and
generally unsolved.
Planar homography is a useful special case of the general structure from
motion problem. In planar homography, the scene is well approximated as a
flat plane. The flat-plane assumption is especially appropriate in aerial
robotics. And this special case is readily solvable.
In this project, you will build a pose-estimation system for the Stanford
Autonomous Helicopter (Ng et. al.) that will use only one video camera.
Specifically, using pre-recorded video from the helicopter, you will code
a planar homography pose-from-motion system and a (very simple) simulator
to visualize the output of your algorithm. An extension, time permitting,
is to find the small percentage of the scene that does not behave like a
flat plane and tag it as an obstacle. Adding this functionality would in
effect create an autonomous landing module in addition to pose estimator.
Project Scope
The main deliverable for this project is a visualization depicting both
the input video frames and the positions of the helicopter in as many
degrees of freedom as you can recover (5 is the theoretical maximum).
Tasks
- Acquire Aerial Video (we have it on DVD)
- Perform Correspondence Matching on Video using OpenCV (1 weeks)
- Compute Initial Homography Approximation (1-2 weeks)
- Build Basic Visualization Engine (1-2 weeks)
- --MIDTERM REPORT--
- Normalize and Decompose Homography Approximation (1-2 weeks)
- Smooth Pose Estimation using Kalman or Particle Filter (2-3 weeks)
--IF TIME PERMITS--
- Obstacle Tagging for Landing
- Compare Discrete Homography + Kalman Filter vs. Continuous Homography
Project Status
Andrei Aron (naron at stanford),
Kevin Lee (kevinlrd at stanford),
Julie Tung (jct at stanford),
Alex Li (alexli at stanford)
Point of Contact
David Stavens,
Hendrik Dahlkamp
Midterm Report
not yet submited
Final Report
not yet submitted
|
 |
|
|
