Integration of CSI and Camera Image for Indoor Wireless Localization
Open Access
- 30 March 2026
- journal article
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Access
- Vol. 14, 49961-49974
Abstract
Research in integrated sensing and communications is advancing toward next-generation systems beyond 5G and 6G, with significant progress in highly accurate indoor localization methods, particularly those leveraging channel state information (CSI) and deep learning (DL) techniques. However, most existing approaches assume access points equipped with a large number of antennas, which limits their applicability in current communication systems that typically use fewer antennas. Furthermore, DL-based localization methods often require extensive training datasets, posing a significant challenge in terms of the cost of generating training samples. This study addresses these challenges by proposing a camera-assisted localization framework designed to reduce both the number of required antennas and the cost associated with generating training samples. The proposed method integrates object detection from camera images to identify users among multiple detected objects and combines the estimated user coordinates from images with CSI data to achieve precise localization. This approach achieves high localization accuracy with a limited number of antennas and training samples. Furthermore, to minimize the effort required to acquire labeled data, the framework incorporates a semi-supervised learning strategy that uses a small set of labeled data together with a larger set of unlabeled data. This method enhances localization accuracy while reducing the cost of training data acquisition. Numerical simulations demonstrate that the proposed method achieves superior localization accuracy with fewer antennas and labeled data compared with conventional methods that rely solely on CSI.Keywords
Funding Information
- Japan Science and Technology Agency Support for Pioneering Research Initiated by the Next Generation (JPMJSP2151)
- Japan Society for the Promotion of Science London (25K01232)
This publication has 50 references indexed in Scilit:
- AI-Based Indoor Localization Using RSSI: CNN-LSTM Based Approach2025
- Improving Passenger Detection With Overhead Fisheye ImagingIEEE Access, 2024
- Indoor Localization With an Autoencoder-Based Convolutional Neural NetworkIEEE Access, 2024
- YOLOv1 to v8: Unveiling Each Variant–A Comprehensive Review of YOLOIEEE Access, 2024
- DNN-Based Indoor Localization Under Limited Dataset Using GANs and Semi-Supervised LearningIEEE Access, 2022
- Non-cooperative Personnel Tracking with Cross Modality Learning in 5G-enabled Warehouse Application2021
- End-to-End Object Detection with Transformers2020
- Deep learning methods for fingerprint-based indoor positioning: a reviewJournal of Location Based Services, 2020
- An Overview on Edge Computing ResearchIEEE Access, 2020
- ML estimation of time and frequency offset in OFDM systemsIEEE Transactions on Signal Processing, 1997