Ulster University Logo

Indoor Localisation Through Object Detection on Real-Time Video Implementing a Single Wearable Camera

Shewell, Colin, Nugent, Chris, Donnelly, Mark and Wang, Haiying (2016) Indoor Localisation Through Object Detection on Real-Time Video Implementing a Single Wearable Camera. In: 14th Mediterranean Conference on Medical and Biological Engineering and Computing, MEDICON 2016, Paphos, Cyprus. Springer Verlag. Vol 57 6 pp. [Conference contribution]

[img] Text - Accepted Version
232kB
[img] Text - Supplemental Material
Indefinitely restricted to Repository staff only.

94kB

URL: http://dx.doi.org/10.1007/978-3-319-32703-7_237

DOI: doi:10.1007/978-3-319-32703-7_237

Abstract

This paper presents an accurate indoor localisation approach to provide context aware support for Activities of Daily Living. This paper explores the use of contemporary wearable technology (Google Glass) to facilitate a unique first-person view of the occupants environment. Machine vision techniques are then employed to determine an occupant’s location via environmental object detection within their field of view. Specifically, the video footage is streamed to a server where object recognition is performed using the Oriented Features from Accelerated Segment Test and Rotated Binary Robust Independent Elementary Features algorithm with a K-Nearest Neighbour matcher to match the saved keypoints of the objects to the scene. To validate the approach, an experimental set-up consisting of three ADL routines, each containing at least ten activities, ranging from drinking water to making a meal were considered. Ground truth was obtained from manually annotated video data and the approach was subsequently benchmarked against a common method of indoor localisation that employs dense sensor placement. The paper presents the results from these experiments, which highlight the feasibility of using off-the-shelf machine vision algorithms to determine indoor location based on data input from wearable video-based sensor technology. The results show a recall, precision, and F-measure of 0.82, 0.96, and 0.88 respectively. This method provides additional secondary benefits such as first person tracking within the environment and lack of required sensor interaction to determine occupant location.

Item Type:Conference contribution (Lecture)
Keywords:Ageing in place; Ambient Assisted Living; Context-aware services; Machine vision; Wearable computing
Faculties and Schools:Faculty of Computing & Engineering
Faculty of Computing & Engineering > School of Computing and Mathematics
Research Institutes and Groups:Computer Science Research Institute > Smart Environments
Computer Science Research Institute
Computer Science Research Institute > Artificial Intelligence and Applications
ID Code:35661
Deposited By: Dr Mark Donnelly
Deposited On:16 Nov 2016 09:35
Last Modified:16 Nov 2016 09:35

Repository Staff Only: item control page