Research on Indoor 3D Positioning Algorithm Based on WiFi Fingerprint.

Area of Science:

• * Computer Science
• * Electrical Engineering
• * Geomatics Engineering

Background:

• * Accurate indoor 3D positioning is crucial for applications in multistory buildings like shopping malls and airports.
• * Existing methods often require additional hardware or lack precision in complex environments.
• * Leveraging wireless access points (APs) for positioning offers a cost-effective solution.

Purpose of the Study:

• * To develop an indoor 3D positioning system using wireless APs without supplementary hardware.
• * To propose a deep learning-based 3D dynamic positioning model utilizing temporal fingerprints.
• * To enhance positioning accuracy and robustness by incorporating temporal information.

Main Methods:

• * A novel 3D dynamic positioning model based on temporal fingerprints is proposed.
• * A sliding time window is used to create temporal fingerprint chips as input for the positioning model.
• * A temporal convolutional network (TCN) with causal convolutions, dilated convolutions, and residual connections extracts spatiotemporal features, which are then fed into a deep neural network (DNN) regressor.

Main Results:

• * The proposed model effectively estimates 3D position coordinates by learning complex nonlinear relationships.
• * Experimental validation using an open-source dataset demonstrated the model's effectiveness.
• * Comparative analysis showed the proposed model yields more accurate 3D position coordinates than existing methods.

Conclusions:

• * The temporal fingerprint-based deep learning model significantly improves indoor 3D positioning accuracy and robustness.
• * The integration of temporal convolutional networks (TCNs) enables effective spatiotemporal feature extraction.
• * This approach offers a practical solution for high-accuracy indoor localization in large-scale, complex environments without additional hardware.