A novel device-free Wi-Fi indoor localization using a convolutional neural network based on residual attention

Area of Science:

• Computer Science
• Electrical Engineering
• Signal Processing

Background:

• Location-based services (LBS) are increasingly used for indoor localization.
• Wi-Fi-based indoor localization using channel state information (CSI) fingerprinting is gaining traction.
• Existing deep learning methods, like CNNs, have limitations in exploring CSI data due to restricted receptive areas.

Purpose of the Study:

• To develop a novel localization method combining high accuracy and generalizability for Wi-Fi indoor localization.
• To address the limited receptive area issue in existing CNN fingerprinting algorithms.
• To enhance the generalizability of tracking systems across different CSI environments.

Main Methods:

• Proposed an attention-augmented residual CNN (RACNN) to fully utilize CSI data and global context.
• Recasted the tracking problem as a denoising task solved by a deep network.
• Investigated and addressed the impact of inertial measurement unit precision variance on tracking performance.

Main Results:

• The RACNN model achieved 99.9% localization accuracy.
• Reduced average localization error to 0.35 m, outperforming traditional methods by 14-15%.
• Demonstrated practical applicability and ability to handle complex indoor environments.

Conclusions:

• The proposed RACNN model offers a significant advancement in device-free Wi-Fi indoor localization.
• The attention mechanism effectively captures crucial CSI data for improved accuracy.
• The method shows high generalizability and robustness in real-world indoor scenarios.