|
|
|
|
 |
|
|
|
| |
| |
|
Tracking Isolated Objects using Affine Structure |
|
| |
Description:
We detect corners (point features) in the images and
track these using Kalman filtering and affine structure. Our work
attempts to unify feature tracking and structure
computation and provide simultaneous solutions to both, each of which aids the
other. We use a constant image velocity Kalman filter based method
(proposed by Reid and Murray) to
obtain the initial correspondences and subsequently impose the structure
based constraints to improve the correspondence results.
Such an approach improves upon the performance of the Kalman filter
based correlation tracker with minimal overheads and enhances
the quality of
correspondences making the structure and motion
computations more accurate. Using the structure and motion we localize
the fixation point in each frame more accurately, and the
gaze demand generated is
smooth and immune to occlusion or disappearance of the
corners. We have real-time (25 Hz) implementations of the tracking
algorithm on a transputer based multiprocessor system, SUN Ultra Sparc
and Pentium's. 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.
Tracking a toy train (what the camera sees):
|
|
 |
| Clip 1 (1.0 MByte Mpeg) |
Clip 2 (1.6 MByte Mpeg) |
Clip 3 (1.1 MByte Mpeg) |
How the camera moves while tracking the toy train:
|
| Clip (1.3 MByte Mpeg) |
Head tracking:
 |
|
Clip (0.4 MByte Mpeg)
|
Reports:
-
Object Tracking using Affine Structure for Point Correspondences.
Gurmeet Singh Manku, Pankaj Jain, Amit Aggarwal, Lalit Kumar and
Subhashis Banerjee.
IEEE CVPR'97, June 19-21, San Juan, Puerto Rico, 1997.
- B. Tech. Project Report of Amit Aggarwal and Lalit Kumar.
-
Finding Point Correspondences in Motion Sequences Preserving Affine
Structure.
G. Sudhir, Subhashis Banerjee and Andrew Zisserman.
Computer Vision and Image Understanding, Vol. 68, No. 2, pp. 237-246,
November, 1997.
|
| |
|
|
|