Enhancing GNSS Performance and Detection of Road Crossing in Urban Area Using Deep Learning

Abstract

In urban areas, signals from Global Navigation Satellite System (GNSS) satellites often arrive at ground receivers with distortion due to non-line-of-sight (NLOS) propagation, and measurements from these distorted signals can cause large positioning errors. When signals arriving at the receiver through an NLOS path are excluded from the position calculation process, the receiver can achieve significantly improved positioning performance. In this paper, we propose a recurrent neural network (RNN)-based NLOS classifier that discriminates LOS and NLOS satellites in urban environments. In addition, as an important application of the proposed classifier, we utilize the proposed technique to detect pedestrian road crossing. Using measurements collected in urban environments, the proposed classifier shows about 90% accuracy in NLOS classification and about 20% better discrimination performance in comparison to the conventional SVM-based NLOS classifier. The proposed positioning technique for road crossing detection using the proposed classifier was demonstrated to achieve positioning accuracy about 45% higher better than that of conventional techniques.