Wireless Indoor Localization Using Convolutional Neural Network and Gaussian Process Regression.

Area of Science:

• Computer Science
• Electrical Engineering
• Signal Processing

Background:

• Accurate indoor localization is crucial for various applications.
• Received Signal Strength Indication (RSSI) fingerprinting is a common Wi-Fi localization technique.
• Existing methods face challenges in complex environments with multipath effects.

Purpose of the Study:

• To develop a novel hybrid localization model combining Convolutional Neural Network (CNN) and Gaussian Process Regression (GPR).
• To improve indoor positioning accuracy using Wi-Fi RSSI fingerprinting data.
• To address limitations of CNN models, such as overfitting, and enhance feature extraction.

Main Methods:

• A pre-processing algorithm converts RSSI values into readable vectors for the CNN.
• A CNN model is trained to extract local features from sequences of RSSI vectors.
• Gaussian Process Regression (GPR) is applied to refine target point coordinates and mitigate CNN overfitting.

Main Results:

• The hybrid CNN-GPR model demonstrated superior performance over the k-nearest neighbor (KNN) algorithm, achieving a 61.8% improvement.
• The CNN model alone improved performance by 45.8% compared to baseline methods.
• GPR further enhanced localization accuracy, validating its effectiveness in refining CNN predictions.

Conclusions:

• The hybrid CNN-GPR model offers a significant advancement in Wi-Fi-based indoor localization.
• The integration of GPR effectively complements CNNs by improving accuracy and reducing overfitting.
• The study confirms the positive impact of GPR kernel functions on localization performance.