Human action recognition using a dynamic Bayesian action network with 2D part models

Authors:
Vivek Kumar Singh;Ram Nevatia
Affiliations:
University of Southern California, Los Angeles, CA;University of Southern California, Los Angeles, CA
Venue:
Proceedings of the Seventh Indian Conference on Computer Vision, Graphics and Image Processing
Year:
2010

Citing 9
Cited 0

Reconstruction of articulated objects from point correspondences in a single uncalibrated image

Computer Vision and Image Understanding
Pictorial Structures for Object Recognition

International Journal of Computer Vision
On Space-Time Interest Points

International Journal of Computer Vision
Detection of Multiple, Partially Occluded Humans in a Single Image by Bayesian Combination of Edgelet Part Detectors

ICCV '05 Proceedings of the Tenth IEEE International Conference on Computer Vision (ICCV'05) Volume 1 - Volume 01
Conditional Random Fields for Contextual Human Motion Recognition

ICCV '05 Proceedings of the Tenth IEEE International Conference on Computer Vision - Volume 2
Discriminative training methods for hidden Markov models: theory and experiments with perceptron algorithms

EMNLP '02 Proceedings of the ACL-02 conference on Empirical methods in natural language processing - Volume 10
3D People Tracking with Gaussian Process Dynamical Models

CVPR '06 Proceedings of the 2006 IEEE Computer Society Conference on Computer Vision and Pattern Recognition - Volume 1
Automatic Discovery of Action Taxonomies from Multiple Views

CVPR '06 Proceedings of the 2006 IEEE Computer Society Conference on Computer Vision and Pattern Recognition - Volume 2
Measure Locally, Reason Globally: Occlusion-sensitive Articulated Pose Estimation

CVPR '06 Proceedings of the 2006 IEEE Computer Society Conference on Computer Vision and Pattern Recognition - Volume 2

Quantified Score

Hi-index	0.01

Visualization

Abstract

This paper presents an approach to simultaneously track the pose and recognize human actions in a video. This is achieved by combining Dynamic Bayesian Action Network (DBAN) with 2D body part models. Existing DBAN implementation relies on fairly weak observation features which affects recognition accuracy. In this work, we propose to use an occlusion sensitive 2D body part model for accurate pose alignment, which in turn improves both pose estimate and action recognition accuracy. To compensate for the additional time required for alignment, we use an action entropy based scheme to determine the minimum number of states to be maintained in each frame while avoiding sample impoverishment. We demonstrate our approach on a hand gesture dataset with 500 action sequences, and show that compared to DBAN, our algorithm achieves 6% improvement in accuracy.