This project focuses on enhancing GNSS positioning accuracy in urban environments by integrating Machine Learning (ML) with a Particle Filter. Traditional GNSS systems face challenges in urban areas due to signal reflections and obstructions, leading to Non-Line-of-Sight (NLOS) signals and positioning errors. Our solution leverages LOS/NLOS classification using a Random Forest machine learning model, combined with particle filtering for more accurate position estimation.
In urban environments, GNSS signals often get obstructed or reflected by buildings, causing NLOS signals that lead to inaccurate positioning. The aim of this project is to classify these signals into LOS or NLOS and use this information to filter out unreliable signals, thereby improving position estimation.
- LOS/NLOS Classification: A Random Forest model classifies satellite signals as either LOS or NLOS based on attributes like signal strength, azimuth, and elevation.
- Particle Filtering: We simulate multiple potential user positions using a Particle Filter. These positions (or particles) are updated based on the machine learning model’s predictions, gradually refining the estimated location.
- Improved Position Estimation: The system improves positioning accuracy by ignoring signals classified as NLOS and focusing on reliable LOS signals.
-
Data Collection:
- We collect satellite data (e.g., azimuth, elevation, signal-to-noise ratio).
- A labeled dataset of LOS/NLOS signals is used to train the Random Forest classifier.
-
Model Training:
- The Random Forest classifier is trained using labeled LOS/NLOS data.
-
Real-time GNSS Simulation:
- During GNSS signal acquisition, each satellite signal is classified as LOS or NLOS.
- The particle filter then refines the possible user positions based on the LOS/NLOS classification.
-
Positioning Accuracy:
- After several iterations, the particle filter provides a final, more accurate estimated location.