Abstract
In view of the inability of Global Navigation Satellite System (GNSS) to provide accurate indoor positioning services and the growing demand for location-based services, indoor positioning has become one of the most attractive research areas. Moreover, with the improvement of the smartphone hardware level, the rapid development of deep learning applications on mobile terminals has been promoted. Therefore, this paper borrows relevant ideas to transform indoor positioning problems into problems that can be solved by artificial intelligence algorithms. First, this article reviews the current mainstream pedestrian dead reckoning (PDR) optimization and improvement methods, and based on this, uses the micro-electromechanical systems (MEMS) sensor on a smartphone to achieve better step detection, stride length estimation, and heading estimation modules. In the real environment, an indoor continuous positioning system based on a smartphone is implemented. Then, in order to solve the problem that the PDR algorithm has accumulated errors for a long time, a calibration method is proposed without the need to deploy any additional equipment. An indoor turning point feature detection model based on deep neural network is designed, and the accuracy of turning point detection is 98%. Then, the particle filter algorithm is used to fuse the detected turning point and the PDR positioning result, thereby realizing lightweight cumulative error calibration. In two different experimental environments, the performance of the proposed algorithm and the commonly used localization algorithm are compared through a large number of experiments. In a small-scale indoor office environment, the average positioning accuracy of the algorithm is 0.14 m, and the error less than 1 m is 100%. In a large-scale conference hall environment, the average positioning accuracy of the algorithm is 1.29 m, and 65% of the positioning errors are less than 1.50 m which verifies the effectiveness of the proposed algorithm. The simple and lightweight indoor positioning design scheme proposed in this article is not only easy to popularize, but also provides new ideas for subsequent scientific research in the field of indoor positioning.
Highlights
In recent years, location-based service (LBS) has become the basic service of people’s daily work and life [1]
The pedestrian dead reckoning (PDR) algorithm is based on the gait characteristics of pedestrian walking and uses a low-cost self-contained sensor to calculate the relative position of the pedestrian
In the indoor positioning system proposed in this paper, we use the micro-electromechanical systems (MEMS) sensor on the smartphone to implement the PDR algorithm, which includes three modules: step detection, stride length estimation, and heading estimation
Summary
Location-based service (LBS) has become the basic service of people’s daily work and life [1]. Many scholars have introduced deep neural network models into the field of indoor positioning to process data from inertial devices and various signal sources in indoor environments. This paper mainly discusses how to use deep learning methods to process sensor data to achieve the purpose of calibrating the cumulative error of PDR, and build a lightweight indoor positioning system that is easy to promote and apply. We propose an innovative application method that saves a trained deep learning network as a KB-level model and stores it in the root directory of a smartphone This model can respond quickly in real-time positioning and meet the real-time requirements of indoor positioning.
Sign-in/Register to view highlights & summary Sign-in/Register to access full text options 
Free) 
Talk to us
Join us for a 30 min session where you can share your feedback and ask us any queries you have
Schedule a call
