Support Vector Machines for Improving Vehicle Localization in Urban Canyons

Abstract

Since the middle ages, the need to identify the vehicles position in their local environment has always been a necessity and a challenge. Today, GNSS-based positioning systems have penetrated several field, such as land transport, emergency systems and civil aviation requiring high positioning accuracy. However, the performances of GNSS-based systems can be degraded in harsh environment due to non-line-of-sight (NLOS), Multipath and masking effects. In this paper, for improving vehicle localization in urban canyons, we address a very challenging problem related to GNSS signal reception state detection (LOS, NLOS or Multipath). A SVMbased system for GNSS Multipath detection using the fusion of information provided by two GNSS antennas is proposed. In this work, we aim to explore the potential of machine learning, and more precisely, Support Vector Machines (SVM) to identify GNSS signals reception state. The SVM-based system developed in this work has used the C/N0 of signals provided by RHCP and LHCP antennas, and satellite elevation as classification criteria. The training data set is constructed by several experimental studies done in real environments, Calais, France . Furthermore, four SVM kernel functions are tested, namely, Linear, Gaussian, Cubic and Quadratic. A GNSS signal reception state detection by applying the proposed SVM-based classifier is demonstrated on real GPS signals, and the efficiency of the system is shown. We obtain empirically an accuracy of signal detection about 93%.

A Survey of Vehicle Localization: Performance Analysis and Challenges

“…In order to improve the accuracy of GPS-based localization by GNSS signal reception state detection, the authors of [72] designed a multi-path detection system based on support vector machines (SVM). To enhance the effectiveness of the map-matching method, the authors of [73] proposed a spatio-temporal-based matching algorithm (STD matching) which considers the spatial features of roads (including road topology and detailed road information), vehicle speed constraints on different roads, and real-time vehicle movement during low-sampling rate GPS trajectories.…”