Data interpretation algorithms for sensor networks in buildings

Abstract

The purpose of this research was to create methods to determine important information from large amounts of data collected from a sensor network in a building. This project focused on data in the form of counts of people in areas of the building, with the goal of determining the movement that occurred. The project was also largely focused on the case when sensors cover the building completely, and the sensors' ranges do not overlap, with the use of this solution being the ability to resolve ambiguity that will exist in all versions of the problem, and the foundation that it provides for solutions to the complex scenarios. A method is given to solve the problem using linear programming and a cross prediction method. Three sources of ambiguity are identified: crossing, move extensions, and flow around cycles. A second approach using movement along predefined paths is given, and this approach is shown to be accurate when all people follow the predefined paths, as long as the paths selected meet certain conditions. An application of this method to the case when sensors do not cover the building is discussed. The algorithms were tested by using simulated people to generate movement data. Crossing prediction was shown to have high accuracy when people move at a constant speed of one node per time step, have less accuracy when people move at variable speeds slightly slower than one, and have very low accuracy when a wide range of speeds both below and above one are allowed. The linear programming method with and without crossing prediction was shown to handle cycle ambiguity well when the number of people in the building was not too large compared to the number of sensors.