The Role of Task Control and Context in Learning to Recognise Gesture

We are interested in methods for building more intelligent cognitive vision systems in our ActIPret project. The aim of this project is understanding the activities of expert operators for teaching and education. Our approach is to learn models for the components and later the task and context of the visual processing in the ActIPret system. The paper first introduces general issues and some approaches for the example of gesture learning and recognition. Second, aspects of our cognitive vision framework are described as they are relevant to the evaluation of the two approaches tested here. Third, the computational models for the time delay RBF (TDRBF) network and Hidden Markov Model (HMM) are described and results given. Finally, extensions of this work and conclusions for system integration of the results are discussed in the light of task-based control and contextual processing.