Survey on classifying human actions through visual sensors
Artificial Intelligence Review
| Hi-index | 0.00 |
We present a method for 3D monocular kinematic pose estimation and activity recognition through the use of dynamical human motion vocabularies. A motion vocabulary is comprised as a set of primitives that each describe the movement dynamics of an activity in a low-dimensional space. Given image observations over time, each primitive is used to infer the pose independently using its expected dynamics in the context of a particle filter. Pose estimates from a set of primitives are inferred in parallel and arbitrated to estimate the activity being performed. The approach presented is evaluated through tracking and activity recognition over extended motion trials. The results suggest robustness with respect to multi-activity movement, movement speed, and camera viewpoint.