What is it about?
Reducing Data Collection for Better Indoor GPS Training AI models for indoor localization requires massive amounts of data from different smartphones, which can be costly and time-consuming. SANGRIA introduces a novel data augmentation technique that helps models adapt to device differences without requiring extensive new data collection. This makes it easier to build accurate indoor positioning systems without gathering Wi-Fi fingerprints from every possible smartphone. By using stacked autoencoder neural networks and gradient boosting, SANGRIA creates synthetic but realistic training data that mimics signal variations across devices. This reduces the burden of data collection while significantly improving location accuracy. With 42.96% lower localization error, SANGRIA paves the way for more scalable and efficient indoor navigation systems.
Featured Image
Photo by Austin Distel on Unsplash
Why is it important?
Indoor localization models often struggle with smartphone heterogeneity, where different devices interpret Wi-Fi signals differently, requiring large-scale data collection for each new phone. SANGRIA introduces a novel data augmentation method that generates synthetic yet realistic training data, reducing the need for extensive real-world data collection. By combining stacked autoencoder neural networks with gradient boosting, SANGRIA learns to adapt to signal variations across devices, achieving 42.96% lower localization error while making indoor positioning systems more scalable, efficient, and device-agnostic.
Perspectives
As indoor navigation becomes essential for smart cities, retail, and emergency services, reducing data collection efforts while maintaining high accuracy is crucial. SANGRIA’s novel augmentation method offers a scalable solution that adapts to device heterogeneity, making indoor localization more practical and widely deployable. This approach could pave the way for future AI-driven positioning systems that require minimal real-world training while delivering consistent and reliable location accuracy across diverse environments.
Danish Gufran
Colorado State University
Read the Original
This page is a summary of: SANGRIA: Stacked Autoencoder Neural Networks with Gradient Boosting for Indoor Localization, IEEE Embedded Systems Letters, January 2023, Institute of Electrical & Electronics Engineers (IEEE),
DOI: 10.1109/les.2023.3279017.
You can read the full text:
Contributors
The following have contributed to this page
