Recognition of Activities in Resource Constrained Environments; Reducing the Computational Complexity

TitleRecognition of Activities in Resource Constrained Environments; Reducing the Computational Complexity
Publication TypeConference Paper
Year of Publication2016
AuthorsEspinilla, M., Rivera Antonio J., Pérez-Godoy M.D., Medina J., Martínez L., and Nugent C.
EditorGarcía, Carmelo R., Caballero-Gil Pino, Burmester Mike, and Quesada-Arencibia Alexis
Conference NameUbiquitous Computing and Ambient Intelligence
Pagination64–74
PublisherSpringer International Publishing
Conference LocationCham
ISBN Number978-3-319-48799-1
Abstract

In our current work we propose a strategy to reduce the vast amounts of data produced within smart environments for sensor-based activity recognition through usage of the nearest neighbor (NN) approach. This approach has a number of disadvantages when deployed in resource constrained environments due to its high storage requirements and computational complexity. These requirements are closely related to the size of the data used as input to NN. A wide range of prototype generation (PG) algorithms, which are designed for use with the NN approach, have been proposed in the literature to reduce the size of the data set. In this work, we investigate the use of PG algorithms and their effect on binary sensor-based activity recognition when using a NN approach. To identify the most suitable PG algorithm four datasets were used consisting of binary sensor data and their associated class activities. The results obtained demonstrated the potential of three PG algorithms for sensor-based activity recognition that reduced the computational complexity by up to 95 {%} with an overall accuracy higher than 90 {%}.