Robust Recognition and Segmentation of Human Actions Using HMMs with Missing Observations
- PDF / 1,111,541 Bytes
- 17 Pages / 600 x 792 pts Page_size
- 43 Downloads / 142 Views
Robust Recognition and Segmentation of Human Actions Using HMMs with Missing Observations Patrick Peursum Department of Computing, Curtin University of Technology, GPO Box U1987, Perth, Western Australia 6845, Australia Email: [email protected]
Hung H. Bui Artificial Intelligence Center, SRI International, 333 Ravenswood Avenue, Menlo Park, CA 94025-3493, USA Email: [email protected]
Svetha Venkatesh Department of Computing, Curtin University of Technology, GPO Box U1987, Perth, Western Australia 6845, Australia Email: [email protected]
Geoff West Department of Computing, Curtin University of Technology, GPO Box U1987, Perth, Western Australia 6845, Australia Email: geoff@cs.curtin.edu.au Received 30 December 2003; Revised 17 August 2004 This paper describes the integration of missing observation data with hidden Markov models to create a framework that is able to segment and classify individual actions from a stream of human motion using an incomplete 3D human pose estimation. Based on this framework, a model is trained to automatically segment and classify an activity sequence into its constituent subactions during inferencing. This is achieved by introducing action labels into the observation vector and setting these labels as missing data during inferencing, thus forcing the system to infer the probability of each action label. Additionally, missing data provides recognitionlevel support for occlusions and imperfect silhouette segmentation, permitting the use of a fast (real-time) pose estimation that delegates the burden of handling undetected limbs onto the action recognition system. Findings show that the use of missing data to segment activities is an accurate and elegant approach. Furthermore, action recognition can be accurate even when almost half of the pose feature data is missing due to occlusions, since not all of the pose data is important all of the time. Keywords and phrases: human motion analysis, action segmentation, HMMs.
1. INTRODUCTION The goal of this work is to develop and analyse a method to automatically segment and classify subactions in a trainable manner within a typical indoors environment.1 This paper considers motion arranged into a two-level hierarchy of events ranked by complexity, where the lower level contains shorter motions (dubbed actions) that chain together to form higher-level events (activities) which are longer and more abstract. The challenge is to segment a higher-level activity into its constituent subactions. This is desirable since it allows the activity to be examined in finer detail, such as determining exactly when an actor manipulates an object so that the position of the object can be localised [3]. 1 Some
ideas presented in this work have also appeared in [1, 2].
In order to perform limb-level action recognition, it is necessary to extract features from an estimation of the human actor’s pose. However, pose estimation in realistic environments will inevitably suffer from the problem of incomplete data. For example, a person’s limbs will often be l
Data Loading...
