Another way to recognize human action

Understanding human action can be posed as a pattern recognition problem. Actions are complex entities, possessing several representations - linguistic, visual, cognitive, and motor. The ability to recognize the actions, the ability to understand, and to react to, human actions is crucial for cognitive systems of the future. Recognition of human action is divided into two main problems. First is the problem of getting whole body motion data. For this, various techniques such as stereo vision or motion capture can be considered. Second is the problem of interpretation of the human motion, which includes modelling of action, feature extraction, classification, and detection of action. The focus of this research is attended to the second problem. This paper is to create a framework that is able to segment and classify individual actions from a stream of human motion using video data. Based on this framework, a model is trained to automatically segment and classify an activity sequence into sub-actions during inferencing. Our goal is to make our system adaptable to different events in different domains. Hidden Markov models (HMMs) originally emerged in the domain of speech recognition. In recent years, they have attracted growing interest in the area of computer vision as well. This report presents a method for determining action scene using a multi-dimensional hidden Markov model (HMM). Instead of using geometric features, actions are converted into sequential symbols. HMMs are employed to represent the action and their parameters are learned from the training data. Based on the most likely performance criterion, the action can be recognized through evaluating the trained HMMs. We will be developing a prototype system to demonstrate the feasibility of the proposed method. The proposed method is applicable to any action represented by a multi-dimensional signal, and will be a valuable tool in human computer interfaces.

Text L.N.AbdullahUPM.pdf Restricted to Repository staff only Download (92kB)

Actions (login required)

View Item