An Investigation and Application of Biology and Bioinformatics for Activity Recognition

Abstract

Activity recognition in a smart home context is inherently difficult due to the variable nature of human activities and tracking artifacts introduced by video-based tracking systems. This thesis addresses the activity recognition problem via introducing a biologically-inspired chemotactic approach and bioinformatics-inspired sequence alignment techniques to recognise spatial activities. The approaches are demonstrated in real world conditions to improve robustness and recognise activities in the presence of innate activity variability and tracking noise.