Application of deterministic methods for indoor people localization

Abstract

This paper investigates the application of deterministic methods for indoor people localization, namely, the method of determining the smallest distance from a person to the centers of the rooms and the method of comparing the coordinates of a person with the area of the rooms. The necessary conditions for the operation of each method were analyzed. It is shown that the application of the method of the smallest distance is optimal when adjacent rooms have homogeneous shape and size, because with heterogeneous rooms, a person near a shared wall may be closer to an adjacent room’s center. For the method of comparing the coordinates of a person with the area of the rooms, the necessary conditions are a sufficient number (at least three coordinate points) and quality (adequate distribution and diversity of input points) of input data, because if they are not sufficient, the accuracy of determining the person's location will decrease. For both cases, it is critical to have information about the floor where the person is located, since geodetic coordinates (latitude/longitude) omit elevation, causing ambiguity in multistory buildings. A practical solution is to apply machine‐learning models trained on RSSI measurements from WAPs. In addition to the necessary conditions of deterministic methods, the paper also investigates the influence of the methods used for data processing. The arithmetic centroid and polygonal area centroid methods were compared to obtain the coordinates of the room centers. It is shown that the arithmetic centroid shifts to large data clusters, while the polygonal centroid is more resistant to uneven data distribution. The Euclidean distance and the haversine methods were compared to obtain the distance between two points. It is shown that the influence of these methods on the accuracy of the method of the least distance to the centers depends on the coordinate system in which the data are represented. For the global coordinate system (WGS 84/EPSG:3857), the best result was shown by the haversine method, and for the local coordinate system - by the Euclidean distance method. It is shown that, subject to the specified conditions, both methods provide a sufficiently high accuracy of localization of people in the room.