ABSTRACT
This chapter focuses on presenting a computationally simple and distributed algorithm for action recognition. The task involves introducing a novel representation of human actions in terms of their basic building blocks, called primitives. Human action recognition can employ data from either environmental or on-body sensing devices. The fields of computer vision and surveillance have traditionally used cameras as environmental sensors to monitor human movement. Most methodologies for representing human movements map all sensor readings to an identical feature space and then use traditional classification algorithms such as k-nearest-neighbor and naive Bayes classifiers to detect movements. A decision-based distributed classification algorithm uses the signal-processing model to classify an unknown action by integrating information from sensors across the network. The problem of recognizing actions using primitives can be viewed as a decision-tree problem in which each internal decision node represents a sensor node and its branches are the primitives identified within that node.
