Neural Network-Based Indoor Positioning Using Virtual Projective Invariants

Indoor positioning techniques has become a key research issue for future smart services because of the high market value in providing location-based services via smartphones and ondemand services. In image sensor communications (ISC) case, one of the main advantages of the indoor navigation system is that the LED itself can transmit its location information using visible light communication. In addition, the camera usually has an angle of arrival sensor that facilitates the precise determination of not only user position but also user orientation. However, because of the nonlinear and highly complicated relationship between 3D scenery and a pictured 2D image, the development of a complex mathematical model is needed to estimate user position using a camera. Neural network is a good approach for minimizing this complicated relationship. Hence, it is possible to develop a precise positioning technique without any complicated mathematical model between the 3D world and 2D image coordinates. This paper proposes a neural network-based novel positioning technique. The proposed method exploits the projective invariant properties of a line that is virtually constructed with the help of ISC. Then, a neural network scheme is used to extract the camera orientation information from that virtual line. Next, a simple mathematical equation is used to estimate user position. Simulation results show the proposed method has better performance than the previous methods.