The player will show in this paragraph |
|
Depth Capture (25 fps) from SwissRanger SR4000. (Top Left) Intensity as reflected from infrared light. (Top Right) Depth of the scene at each pixel. (Bottom Left) Plotted 3D points (front view). (Bottom Right) Plotted 3D points (side view). |
| For more information on the camera: go to mesa imaging |
The player will show in this paragraph |
| After simple depth segmentation |
The player will show in this paragraph |
The player will show in this paragraph |
| Jittering the location of the box in 3D. For each jittered location, we score how well the observed image matches the appearance of the head. The idea then is to keep the locations that have the highest scores and discard locations with low scores. |
Jittering the location of the box in 2D |
The player will show in this paragraph |
The player will show in this paragraph |
| Tracking the head in 3D. Here we display the location using the weighted average of scored locations. Because we don't update the appearance of the head in each subsequent frame, the tracker loses track of the head when the person turns away. |
Tracking the head in 2D |