Fast and Precise Neural Network-Based Environment Detection utilizing UWB CSI for Seamless Localization Applications
Kia, Ghazaleh; Plets, David; Van Herbruggen, Ben; Fontaine, Jaron; Verloock, Leen; De Poorter, Eli; Talvitie, Jukka (2023)
Kia, Ghazaleh
Plets, David
Van Herbruggen, Ben
Fontaine, Jaron
Verloock, Leen
De Poorter, Eli
Talvitie, Jukka
Teoksen toimittaja(t)
Nurmi, Jari
Sospedra, Joaquin Torres
Lohan, Elena-Simona
Huerta, Joaquin
Ometov, Aleksandr
IEEE
2023
2023 International Conference on Localization and GNSS, ICL-GNSS 2023 - Proceedings
This publication is copyrighted. You may download, display and print it for Your own personal use. Commercial use is prohibited
Julkaisun pysyvä osoite on
https://urn.fi/URN:NBN:fi:tuni-2023112210172
https://urn.fi/URN:NBN:fi:tuni-2023112210172
Kuvaus
Peer reviewed
Tiivistelmä
Seamless localization, navigation, and tracking applications can be realized utilizing different sensors and cameras, radio frequency signals such as WiFi, ultra-wideband, and global navigation satellite system, each of which is better suited for different types of environments. As such, awareness of the environment is crucial for the system to efficiently utilize the most relevant resources in each scenario and enable seamless transition between different environments. For example, when vehicles are moving from an open area such as open highway to crowded urban streets, or the opposite, they experience a considerable environment transition, which triggers opportunities for wide-range environment-specific device and algorithm optimization. In this paper, a novel infrastructure-free method utilizing channel state information of ultra-wideband signals and a convolutional neural network is proposed. This method enables a fast detection of the environment type, including crowded urban and open outdoor, reaching a detection latency of only three milliseconds. The experimental data is collected in the real environments of the city of Ghent, Belgium. The test data set, used for numerical performance evaluations, is collected from areas different from those used in the training set. The results show that the proposed method provides an average environment detection accuracy of 90% in the considered test setup.
Kokoelmat
- TUNICRIS-julkaisut [17147]