Previous Abstract Return to Session E4b Next Abstract

Session E4b: Smartphone Decimeter Challenge (Co-sponsored by Google)

Integration of SVM-Based NLOS Classification and Solution-Separation Test for Accurate Smartphone RTK and PPP-RTK Positioning in Dense Urban Areas
Yang Jiang, Zhitao Lyu, Yan Zhang, Yang Gao, University of Calgary
Date/Time: Thursday, Sep. 19, 2:58 p.m.

High-precision positioning with smartphones using Global Navigation Satellite Systems (GNSS) has potential to support many applications including vehicle navigation and sports monitoring. In dense urban areas, it remains difficult to achieve such accuracy due to the Non-Line-Of-Sight (NLOS) effect of GNSS signals. In a traditional approach, statistical testing techniques such as Solution Separation (SS) and Receiver Autonomous Integrity Monitoring are applied to detect and exclude outlier measurements based on estimation models, but their performance is poor when with a low number of healthy measurements. Although machine-learning-based NLOS classification and multipath estimation techniques can improve positioning performance in the measurement domain, it does not consider the estimation model which contains useful information such as test statistics for the NLOS classifier.
This research presents a novel approach to improve smartphone GNSS positioning accuracy in dense urban areas. This is achieved by integrating a Support Vector Machine (SVM)-based NLOS classification system with a binary-tree-based outlier searching scheme of the SS-test. First, the SVM classification is used to obtain the probability of NLOS occurrences, which is utilized to guide the direction for the outlier searching scheme. Second, after identifying outliers through statistical testing, their probabilities of NLOS are adjusted to increase the classification accuracy.
To evaluate the performance of our integrated method, we conducted a vehicle experiment in the dense urban area of Calgary City, collecting smartphone GNSS data using Xiaomi MI 8 and Google Pixel 7 Pro models, where a NovAtel ProPak 7 system is used to provide a reference trajectory. After training the SVM classification model, the performance of the proposed method is evaluated based on RTK and PPP-RTK tests. The results demonstrate that our method not only improves NLOS classification accuracy but also significantly enhances GNSS positioning accuracy for smartphone users, marking a substantial advancement in urban navigation technologies.

Previous Abstract Return to Session E4b Next Abstract