DMCA
Space-Time Shapelets for Action Recognition (2007)
Cached
Download Links
| Citations: | 3 - 0 self |
BibTeX
@MISC{Batra07space-timeshapelets,
author = {Dhruv Batra and et al.},
title = {Space-Time Shapelets for Action Recognition},
year = {2007}
}
Share
 |
 |
 |
 |
||
OpenURL
Abstract
Recent works in action recognition have begun to treat actions as space-time volumes. This allows actions to be converted into 3-D shapes, thus converting the problem into that of volumetric matching. However, the special nature of the temporal dimension and the lack of intuitive volumetric features makes the problem both challenging and interesting. In a data-driven and bottom-up approach, we propose a dictionary of mid-level features called Space-Time Shapelets. This dictionary tries to characterize the space of local space-time shapes, or equivalently local motion patterns formed by the actions. Representing an action as a bag of these space-time patterns allows us to reduce the combinatorial space of these volumes, become robust to partial occlusions and errors in extracting spatial support. The proposed method is computationally efficient and achieves competitive results on a standard dataset [5].
Keyphrases
action recognition space-time shapelets bottom-up approach partial occlusion mid-level feature special nature local motion pattern temporal dimension local space-time shape standard dataset combinatorial space spatial support space-time pattern 3-d shape achieves competitive result dictionary try space-time volume recent work volumetric matching intuitive volumetric feature
