Item

Abstract

Currently, people are accustomed to using smartphones that integrate numerous sensors for navigation. In pedestrian navigation, the walking distance is measured by a step counter utilizing the smartphone's built-in Inertial Measurement Unit (IMU). However, it is difficult to calculate the exact distance change in the smartphone-based vehicle navigation with the traditional inertial navigation method due to the low quality of IMU in smartphones. Global Navigation Satellite System (GNSS) modules play a major role in most cases of smartphone-based vehicle navigation. Since GNSS signals are frequently blocked or reflected in tunnels or urban canyons, positions provided by GNSS are unreliable in these areas. A vehicle positioning algorithm with deep learning-driven distances prediction is proposed in this work to continually enhance the location accuracy during GNSS outages. Our deep learning network consists of the Convolutional neural network (CNN) and Gated Recurrent Unit (GRU). The inputs of the network are the raw IMU and barometer data. The labels for the network training are provided by the integrated GNSS/IMU/Barometer solutions in the smartphone. In GNSS-blocked areas, IMU, barometer, and deep odometry are combined to offer precise locations for navigation systems. Results indicated that the suggested technique outperforms the Non-Holonomic Constraints (NHC) assisted IMU in the horizontal and vertical directions by 73.14% and 98.33% respectively. This proposed integrated system is implemented on Android devices to illustrate that the proposed approach function effectively.