A rule-based approach to indoor localization based on WiFi signal strengths
by Qiuxia Chen
Ph.D. Computer Science and Engineering
xiii, 100 p. : ill. ; 30 cm
Location plays a very important role in location-aware computing systems, in which objects are retrieved based on their physical locations. For example, finding the nearest objects around a person requires knowledge about the locations of the objects and the location of the person. The identification of the location of an object is known as localization. GPS (Global Positioning System) is widely used for localizing outdoor objects. Unfortunately, it does not work indoor because GPS signal cannot penetrate into buildings....[ Read more ]
Location plays a very important role in location-aware computing systems, in which objects are retrieved based on their physical locations. For example, finding the nearest objects around a person requires knowledge about the locations of the objects and the location of the person. The identification of the location of an object is known as localization. GPS (Global Positioning System) is widely used for localizing outdoor objects. Unfortunately, it does not work indoor because GPS signal cannot penetrate into buildings.
This thesis investigates localization methods in indoor environments. Since GPS is not available, a sensor infrastructure must be available to make indoor localization possible. This thesis focuses on approaches based on the Received Signal Strength (RSS) of WiFi signals because WiFi is widely available in indoor spaces. The main application scenario of this research is to identify the location of a user inside a building. To achieve this goal, RSSs are measured at each location of a space and stored in the server. The measurements are called location signatures of the space. When localization is performed, the user obtains the RSS signature at her (unknown) location, and compares it with the location signatures at the server. The location with signature matching the user’s signature the best is returned as the location of the user.
Traditional localization methods aim to improve localization accuracy, i.e., the error between the estimated location and the actual location. However, they assume that the location signatures are accurate. Unfortunately, RSSs are unstable due to noise, obstacles and environmental changes, causing localization accuracy to deteriorate quickly. Thus, regular calibration on the location signatures, which is prohibitively expensive, is required to maintain high localization accuracy.
This thesis aims to improve both the accuracy and the stability of indoor localization. Instead of using absolute RSSs in comparing the location signatures, we propose a rule- based approach to achieve high localization accuracy and stability. The main idea is to maintain the relations (i.e., “less than”, “equal to”, and “greater than”) of the RSSs of the access points (APs) received at a location and to set up rules to match the RSS signatures based on the relations. The rule-based approach enhances stability because the relation between two RSS signals could remain stable even when their values are changing constantly.
To further address the stability problem, we introduce two important notions, the stability and sensibility of an AP, at a location. Although the RSSs from APs change over time, some APs change less than the others, thus having higher stability, and some APs have stronger signals than the others, thus having higher sensibility. We introduce methods to estimate the stability and sensibility of APs. We present an effective and simple approach to create the relations and rules, as well as heuristics to select the rules for use in localization. We develop a suite of rule-based localization methods based on different combinations of the techniques, including pure matching of location signatures, rule-based system with and without AP stability, and rule-based systems with and without rule stability. We implemented the location methods and tested them in the Department’s Lab area and the results show that rule-based systems with stability consideration perform better than those without stability consideration, which in turn perform better than methods based on pure signature comparison.