Back
to indexDon Murray and Jim Little, 2000
Summary
This paper presented a robust method to build a 2D occupancy grid, in real time, based on stereo vision. The occupancy grid contains the probability that grid space is occupied. The approach uses a 2D plane even though 3D disparity analysis is possible. Since the robot is fundamentally moving about in a 2D plane, with the camera located at the highest point, it allows the robot to move under obstacles that are higher than the cameras. "Spikes" in data were observed, corresponding to areas where insufficient contrast between objects existed. The concluding paragraphs identify how the spikes were constructed and apply a simple algorithm to reduce this noise.
Methods
First, the camera position was interesting because the authors used 3 camera in an "L" shape. Using this placement reduced errors since it was unlikely errors would occur in all three cameras based on their angle. The authors speculate that even more cameras (4 seems like a good number) would further reduce the error.
The authors apply a simple algorithm which measures the disparity between pixels in a stereo image. When the disparity between pixels at a particular depth exceeds a threshold, the conditional probability that an object exists at that position is noted. A map is updated based on the simple rule if a grid square is occupied, than increase the grid value at that location. Otherwise, decrease the grid value by the same amount.
Keywords
stereo vision, disparity surfaces, occupancy grid
Rating
8
Bibtex Entry
@article{ murray_little00,
author = "Don Murray and Jim Little",
title = "Using real-time stereo vision for mobile robot navigation",
journal = "Autonomous Robots",
volume = "8",
number = "2",
pages = "161--171",
year = "2000 ",
url = "citeseer.nj.nec.com/article/murray00using.html"
}