Localization and Navigation with Imagery and Pedestrian Inertial Measurements Utilizing Artificial Neural Networks

David Ellis and Joseph Curro

Abstract:	The field of deep learning research has exploded in recent years creating new potential techniques using artificial neural network (ANN) to solve a wide variety of problems including navigation problems. This paper analyzes a method combining two ANNs using an extended Kalman Filter (EKF) for pedestrian navigation in Global Positioning System (GPS) challenged or denied environments. The technique utilizes sensors in cell phones including the camera, magnetometer, gyroscope, and accelerometer to train two ANNs, for position localization and dead reckoning. The absolute position solution utilizes a convolutional neural network (CNN) to determine location within an environment. Additionally, a gated recurrent unit (GRU) network uses inertial measurement unit (IMU) and magnetometer sensor data to provide a pedestrian dead reckoning solution. The image localization and pedestrian dead reckoning solutions are fused via an EKF to determine the final navigation solution.
Published in:	Proceedings of the 2020 International Technical Meeting of The Institute of Navigation January 21 - 24, 2020 Hyatt Regency Mission Bay San Diego, California
Pages:	672 - 682
Cite this article:	Ellis, David, Curro, Joseph, "Localization and Navigation with Imagery and Pedestrian Inertial Measurements Utilizing Artificial Neural Networks," Proceedings of the 2020 International Technical Meeting of The Institute of Navigation, San Diego, California, January 2020, pp. 672-682. https://doi.org/10.33012/2020.17170
Full Paper:	ION Members/Non-Members: 1 Download Credit Sign In