WiFi Indoor Location Based on Area Segmentation.

Area of Science:

• Computer Science
• Signal Processing
• Machine Learning

Background:

• High-precision, low-cost indoor positioning is crucial for future location-based services.
• Indoor data exhibits uneven distribution and high correlation, posing challenges for traditional algorithms.
• WiFi signal instability introduces noise, degrading indoor positioning system performance.

Purpose of the Study:

• To propose a novel regionalized indoor positioning system leveraging the DBSCAN algorithm.
• To address challenges of data sparsity and noise in indoor environments.
• To enhance positioning accuracy and reduce latency for indoor localization.

Main Methods:

• Region segmentation using Density-Based Spatial Clustering of Applications with Noise (DBSCAN) to identify core, edge, and noise points.
• Dimensionality reduction via Stacking Auto-Encoders (SAE) for core regions.
• Localization using Deep Neural Networks (DNN) with adaptive learning rates in core regions.
• Localization using Random Forest (RF) algorithm for edge regions.

Main Results:

• The proposed system achieved high positioning accuracy on the UJIIndoorLoc dataset.
• Edge points: positioning accuracy within 1.5m achieved with >87.2% probability in 32ms.
• Core points: positioning accuracy within 1.5m achieved with >98.8% probability.

Conclusions:

• The DBSCAN-based regionalized positioning system effectively handles noisy and uneven indoor data.
• The hybrid approach combining SAE-DNN and RF offers superior performance compared to traditional methods like Multi-Layer Perceptron and K-Nearest Neighbors.
• The system demonstrates a promising solution for high-accuracy, low-latency indoor positioning.