|
|
|
|
 |
|
|
|
| |
| |
|
Collision Detection using Image Divergence |
|
| |
Description:
We make a linear approximation to the optic flow (image velocity)
field and decompose the linear field in terms of the differential
invarinats: divergence, curl and deformation.
The time to collision is inversely related to the
image divergence. We have three different implementations
of collision detection, all running at frame rate (25 Hz),
on a multi-processor transputer based system, SUN Ultra-Sparc and Pentium.
We have also explored the possiblity of
HW-SW codesign using FPGA's for higher frame rates. Currently
we are developing a real-time operating system for such
real-time vision systems as a part of the
embedded systems project.
We show some Mpeg clips below.
The robot runs away:
 |
|
 |
| Clip 1 (0.7 MByte Mpeg) |
|
Clip 2 (0.8 MByte Mpeg) |
What the camera sees:
 |
 |
| Clip 1 (0.4 MByte Mpeg) |
Clip 2 (0.5 MByte Mpeg) |
Does not react to fronto-parallel motion:
 |
|
Clip (0.3 MByte Mpeg)
|
Reports:
-
Real Time Vision System for Collision Detection.
Alok Mittal, Aditya Vailaya, Subhashis Banerjee and M. Balakrishnan.
Journal of Computer Science and Informatics, Special
Issue on Robotics and Automation, 25(1), pp. 174-208, March 1995.
-
Real Time Collision Detection and Avoidance: A Case Study for
Design Space Exploration in HW-SW Codesign.
S. K. Lodha, Shashank Gupta, M. Balakrishnan and Subhashis Banerjee.
CSI/IEEE Intl. Conf. on VLSI Design, Chennai, India, pp. "97-102", January 1998.
|
| |
|
|
|
