Active Vision for Human Action Sensing
2010
This paper presents a real-world implementation of a multi-camera active vision system for form/action recognition of time-varying geometry objects (specifically, the human form). The proposed method uses a novel sensing-system reconfiguration methodology to improve recognition performance. Near-optimal view-points are selected by the algorithm, along a prediction time horizon. to maximize target visibility in the presence of multiple, potentially moving obstacles. The proposed reconfiguration method has been tested using controlled experiments.
Keywords:
- Correction
- Source
- Cite
- Save
- Machine Reading By IdeaReader
24
References
1
Citations
NaN
KQI