|Figure 1. Example event
detections in crowded videos.
2. Volumetric video
segmentation. We use mean shift to
segment the video in space-time. Volumetric segmentation leads to
more consistent regions over time, versus segmenting individual frames.
3. Example volumetric model
of a handwave action.
4. Our shape matching
algorithm is based on region intersection between two shapes. The
shaded area represents the distance between the template and the
video. We are able to match the template with over-segmented
regions and the running time is linearly proportional to the surface
area of the three-dimensional template.
-- Whole Template
-- Parts Model
|Figure 5. We break the template into parts for more
robustness and improved generalization ability.