Detection of individuals' intentions and actions from a stream of human behaviour is an open and complex problem. There is however an intrinsic need to automatically recognise the activities performed by users of mobility assistive aids to better understand their behavioural patterns, with the ultimate objective of improving the utility of these devices. While discriminative algorithms such as Support Vector Machines (SVM) are well understood, generative probabilistic approaches to machine learning such as Dynamic Bayesian Networks (DBN) have only recently started gaining increasing interest within the robotics community. In this paper, a comprehensive evaluation of these techniques is carried out for human activity recognition in the context of their applicability to assistive robotics. Results show comparable recognition rates, offering valuable insights into the advantageous characteristics of DBN in relation to their dynamic and unsupervised nature for realistic human-robot interaction modelling.
展开▼
