Artificial Neural Networks Aided Image Localization for Pedestrian Dead Reckoning for Indoor Navigation Applications

Abstract

Recently, indoor navigation has become popular because of the popular smartphone and the growth of Location Based Services (LBS). Pedestrian Dead Reckoning (PDR) has the good potential to confront the challenges in environments lacking a Global Navigation Satellite System (GNSS). However, PDR has inherent errors that accumulated step by step. An image-based localization can as an aiding system, because virtually all mobile devices contain a basic camera sensor. However, the image-based localization requires successive and overlapped images for continuously positioning. In addition, the solutions provided by either image-based localization or a PDR are usually in a relative coordinate system. Therefore, this study proposes a system, which uses space resection-aided PDR with georeferenced images of a previously mapped environment. In order to implement the procedure automatically and reduce the image processing, this study further uses markers in the georeferenced images. After that, Artificial Neural Network (ANN) is novel applied to estimate the distance between the marker and camera. Since the marker is also georeferenced, the camera position is updated through the detected georeferenced marker, estimated distance, and orientation from inertial sensor, and then update the PDR result. The indoor mobile mapping system (IMMS) is used for the effective production of georeferenced images. The result shows the proposed system is able to initialize in indoor without manually given initial position and provides long-term accurate navigation.