Indoor Positioning Based-on Images Aided by Artificial Neural Networks

With the springing up of smartphones, indoor navigation becomes more and more popular. One of the algorithms in the domain of indoor navigation is Pedestrian Dead Reckoning (PDR), which has the good potential to confront the challenge of the blocked satellite signal. Moreover, the error of inertial sensors accumulating with time can be solved by updating geospatial information steadily. This study adopts a method based on the built-in sensors combining with the camera. In order to reduce the image processing, the study further adopts the marker self-designed to aid in carrying out indoor positioning. Then, the Artificial Neural Network (ANN) is applied to estimate the distance between the marker and the camera. Because the marker is also georeferenced, the position of camera is calculated through the detected georeferenced marker, estimated distance. Afterward, the result of PDR can be updated. In this study, the result shows that the accuracy using Multi-Layer Feed-Forward Neural Networks (MFNNs) is higher than traditional techniques. However, the architecture still can't overcome the catastrophic forgetting in the neural network. For this predicament, this study proposes using Cascade Correlation Networks (CCNs) and adding the key data to improve accuracy. As a result, based on the same training data, trying to add some key data makes the accuracy can achieves 0.5 meters. © 2019, Chinese Institute of Civil and Hydraulic Engineering. All right reserved.