Please use this identifier to cite or link to this item:
|Title:||Transfer learning enhanced AR spatial registration for facility maintenance management||Authors:||Chen, Keyu
Cheng, Jack C. P.
Li, Chun Ting
|Keywords:||Engineering::Electrical and electronic engineering||Issue Date:||2020||Source:||Chen, K., Yang, J., Cheng, J. C. P., Chen, W. & Li, C. T. (2020). Transfer learning enhanced AR spatial registration for facility maintenance management. Automation in Construction, 113, 103135-. https://dx.doi.org/10.1016/j.autcon.2020.103135||Journal:||Automation in Construction||Abstract:||Augmented reality (AR), which requires a spatial registration technique, has proved to greatly improve the efficiency of facility maintenance management (FMM) activities. Being one of the most promising techniques for indoor localization, Wi-Fi fingerprinting has been widely used for AR spatial registration. However, localization accuracy of Wi-Fi fingerprinting decreases over time due to dynamics of environmental factors. Readings from different mobile devices can also affect the accuracy negatively. In this paper, a transfer learning technique named transferable CNN-LSTM is proposed for improving the robustness of Wi-Fi fingerprinting while implementing AR in FMM activities. Convolutional neural network (CNN), embedded with long short term memory (LSTM) networks, is utilized to predict the location of unlabeled fingerprints. Multiple kernel variant of maximum mean discrepancy (MK-MMD) is adopted to reduce the distribution difference between the source domain and the target domain, so that the location of the newly collected unlabeled fingerprints can be predicted accurately. As shown in the experimental validation, the transferable CNN-LSTM can achieve an accuracy of 97.1% in short-term (without significant environmental changes) spatial registration, 87.8% in long-term (with significant environmental changes) spatial registration, and around 90% in multi-device spatial registration, indicating a higher accuracy and better robustness over other conventional approaches.||URI:||https://hdl.handle.net/10356/161072||ISSN:||0926-5805||DOI:||10.1016/j.autcon.2020.103135||Schools:||School of Electrical and Electronic Engineering||Rights:||© 2020 Elsevier B.V. All rights reserved.||Fulltext Permission:||none||Fulltext Availability:||No Fulltext|
|Appears in Collections:||EEE Journal Articles|
Updated on Dec 4, 2023
Web of ScienceTM
Updated on Oct 31, 2023
Updated on Dec 7, 2023
Items in DR-NTU are protected by copyright, with all rights reserved, unless otherwise indicated.