Prediction of Integrated Accuracy of GNSS/IMU/Speed Sensor by Machine Learning

Real-time kinematic positioning (RTK), one of the precise positioning methods using the global navigation satellite system (GNSS), is expected for intelligent transportation systems (ITS). In open areas RTK provides positioning solutions with centimeter level accuracy. However, in urban areas, the notorious multipath and none-line-of-sight (NLOS) degrades positioning accuracy to meters level. Besides, GNSS outage often happens due to signal attenuation. Considering the complementarity, inertial measurement unit (IMU) is integrated with GNSS to provide continuous positioning service in urban canyons. Developers of selfdriving cars have very high expectations for cm-level positioning with GNSS and other low-cost sensors. In particular, it is said to be very important to indicate where cm-level positioning is possible, and where cm-level positioning is not possible. Based on this background, simulation software that can predict the availability of cm-level positioning is becoming increasingly useful, as long as a 3D map and satellite orbit are input. To solve this problem, this paper proposes to predict the integrated accuracy of GNSS/IMU/Speed sensor by Deep Learning. In this paper, we have focused on the accuracy less than 30cm horizontal error. In particular, the features derived from the measurements produced by the GNSS/IMU integrated system are employed as the input of the DNN, such as the number of satellites, the time since the last fixed solution, etc. Three real vehicular datasets collected in urban canyons of Tokyo are exploited to validate the effectiveness of the proposed method. The results show that an average prediction accuracy of 78.6% and an Area Under Curve (AUC) of the Receiver Operating Characteristic (ROC) Curve of 0.801 are achieved. Additionally, we show application example of RTK Fix/ No Fix prediction using Deep Learning. When LEO satellite was available like GNSS satellite and case of UAV delivery in urban area in different flight height.