Improving indoor positioning system using weighted linear least square and neural network

Abstract

Indoor positioning has grasped great attention in recent years. Many of those technologies are related to the problem of determining the position of an object in space, such as the robot, people, and so on. In this paper, we combine a range-free method, i.e., fingerprinting, and a range-based method, i.e., multi-lateration, to propose a novel indoor positioning system using the received signal strength indicator (RSSI). First, we apply multi-layer perceptron neural network (MLP-NN) on a time series of RSS readings to coarsely estimate the target location. From the knowledge of the coarse location, we select reliable beacons and apply least square-based multi-lateration to their estimated distance to finely estimate the target position. We also proposed a novel weighted least square method based on uncertainty propagation to improve localisation accuracy. Experiments have shown that our proposed system, which is implemented on Raspberry Pi (RPi), is highly precise and deployable.