An effective fusion positioning methodology for land vehicles in GPS-denied environments using low-cost sensors

It is still a main challenge to achieve accurate and reliable vehicle positioning in Global Position System (GPS)-denied environments using low-cost sensors. Although existing methods have obtained a certain performance improvement, there is still room for further enhancement in positioning accuracy due to the lack of effective observation information during GPS outages. To address this challenge, this paper proposes an effective fusion positioning methodology based on enhanced observation information. Firstly, a data denoising algorithm based on convolutional denoising autoencoder is developed to effectively filter the noises in inertial sensors, so as to provide more clean data for subsequent modeling. Then, a differential fusion strategy is designed to selectively fuse multiple observation information from monocular camera, low-cost GPS and wheel speed sensor with extended Kalman filter, which can further improve the positioning accuracy. Finally, a long short-term memory-based error prediction model is constructed to learn the mapping relationship between the result of data denoising and differential fusion. In the case of GPS outages, the error model provides accurate corrections to compensate for position errors of inertial navigation system. The performance of the proposed methodology was evaluated on real-world data collected in complex urban environments. Experimental results indicate that the proposed positioning methodology can obtain significant accuracy enhancement.