Walter Bell (wbell@cs.cornell.edu)
Pedro Felzenszwalb (pff@cs.cornell.edu)
Dan Huttenlocher (dph@cs.cornell.edu)
March 26th, 1999
This example shows some results of the tracking system. For more examples, refer to the paper. This example show multiple cars being detected and tracked for long periods of time. Each target is surrounded with either a green box, or a yellow box, with its unique identifier in the lower right corner. If the target was found at that frame, the box is green, and the edges of the model overlay the target's location. If the box is yellow, the target could not be matched, and the target's position is being predicted by its previous motion.
A tree lined roadway. A lone car travels down an obstructed roadway at high speed. This sequence has significant camera motion between frames, which makes accurate background estimation essential. The car is acquired at frame 206, and just a few frames later at frame 211, the initial model has expanded to the complete car. The car proceeds in and out of several tree obstructed areas, and is picked up after each one as the same target. Note that object 3 is incorrectly identified as a moving object, and is tracked for a few frames before being lost.
|
|
|
|
![]() Frame 152 |
![]() Frame 159 |
![]() Frame 173 |
![]() Frame 180 |
![]() Frame 204 |
![]() Frame 253 |
![]() Frame 298 |
![]() Frame 313 |
© 1999 walter bell