| Item type |
SIG Technical Reports(1) |
| 公開日 |
2021-05-27 |
| タイトル |
|
|
タイトル |
Evaluation of Anti-Deterioration in Wi-Fi Positioning by Pseudo-Labeling |
| タイトル |
|
|
言語 |
en |
|
タイトル |
Evaluation of Anti-Deterioration in Wi-Fi Positioning by Pseudo-Labeling |
| 言語 |
|
|
言語 |
eng |
| キーワード |
|
|
主題Scheme |
Other |
|
主題 |
位置推定，都市 |
| 資源タイプ |
|
|
資源タイプ識別子 |
http://purl.org/coar/resource_type/c_18gh |
|
資源タイプ |
technical report |
| 著者所属 |
|
|
|
Ritsumeikan University, Graduate School of Information Science and Engineering |
| 著者所属 |
|
|
|
Ritsumeikan University, College of Information Science and Engineering |
| 著者所属 |
|
|
|
Ritsumeikan University, College of Information Science and Engineering |
| 著者所属(英) |
|
|
|
en |
|
|
Ritsumeikan University, Graduate School of Information Science and Engineering |
| 著者所属(英) |
|
|
|
en |
|
|
Ritsumeikan University, College of Information Science and Engineering |
| 著者所属(英) |
|
|
|
en |
|
|
Ritsumeikan University, College of Information Science and Engineering |
| 著者名 |
Junkai, Ge
Ziyufei, Li
Nobuhiko, Nishio
|
| 著者名(英) |
Junkai, Ge
Ziyufei, Li
Nobuhiko, Nishio
|
| 論文抄録 |
|
|
内容記述タイプ |
Other |
|
内容記述 |
In recent years, with the proliferation of smartphones and the deployment of more signal transmitting base stations, indoor Wi-Fi fingerprinting algorithms have been used as the primary method for estimating location. The fingerprint algorithm needs to build a localization model in advance, however, deterioration caused by dynamic environmental changes causes a decline in the accuracy. In general, we use labeled data to train the localization mode, but re-collecting labeled data every once in a while will cost a lot. Compared with labeled data, the collection of unlabeled data is much simpler and can be collected through various methods such as crowd-sourcing. This paper proposes a method to attach the estimated position to the unlabeled data as a label, and make Pseudo-Labeled data for training model. The proposed method can effectively reduce the workload. Using the data collected in Umeda underground street to verify, compared with the original localization model without calibration, the positioning accuracy increased by an average of 33%. |
| 論文抄録(英) |
|
|
内容記述タイプ |
Other |
|
内容記述 |
In recent years, with the proliferation of smartphones and the deployment of more signal transmitting base stations, indoor Wi-Fi fingerprinting algorithms have been used as the primary method for estimating location. The fingerprint algorithm needs to build a localization model in advance, however, deterioration caused by dynamic environmental changes causes a decline in the accuracy. In general, we use labeled data to train the localization mode, but re-collecting labeled data every once in a while will cost a lot. Compared with labeled data, the collection of unlabeled data is much simpler and can be collected through various methods such as crowd-sourcing. This paper proposes a method to attach the estimated position to the unlabeled data as a label, and make Pseudo-Labeled data for training model. The proposed method can effectively reduce the workload. Using the data collected in Umeda underground street to verify, compared with the original localization model without calibration, the positioning accuracy increased by an average of 33%. |
| 書誌レコードID |
|
|
収録物識別子タイプ |
NCID |
|
収録物識別子 |
AA11838947 |
| 書誌情報 |
研究報告ユビキタスコンピューティングシステム（UBI）
巻 2021-UBI-70,
号 10,
p. 1-8,
発行日 2021-05-27
|
| ISSN |
|
|
収録物識別子タイプ |
ISSN |
|
収録物識別子 |
2188-8698 |
| Notice |
|
|
|
SIG Technical Reports are nonrefereed and hence may later appear in any journals, conferences, symposia, etc. |
| 出版者 |
|
|
言語 |
ja |
|
出版者 |
情報処理学会 |
IPSJ-UBI21070010.pdf (805.4 kB)
