Classification of Human Walking Patterns through Singular Value Decomposition Projection
Conference Paper
Overview
Identity
Additional Document Info
Other
View All
Overview
abstract
Sensing of structural vibration provides a rich source of information that can be used in structural health monitoring and impact/fault localization among other applications. In this paper, acceleration measurements from vibration sensors (accelerometers), installed in an operational smart building (Virginia Techs Goodwin Hall), are used to classify footsteps of different kinds from building occupants. Goodwin Hall is a 160,000 square foot five story building instrumented with over 200 accelerometers mounted to the buildings structure. Singular value decomposition (SVD) projection is used to classify measured data into categories seeded with training data. Contrary to the black box machine learning approach, the SVD framework allows classification parameters to be easily modified and their effects visualized to be understood. Better understanding of the classification problem and its dominant parameters will allow the development of more accurate and robust algorithms for classification of a wide variety of signals.
