Spatial model-aided indoor tracking

Abstract

In order to address the problem of indoor pedestrian tracking, this thesis reports a research on spatial models' ability to reduce tracking error of a WiFi positioning system. There are three main objectives in this research. First, it is to build a suitable spatial model for tracking. Second, it is to develop a tracking algorithm that can make full use of the spatial model. Last, the tracking algorithm should be tested in a live environment. Based on literature study, a grid-based spatial model is chosen to be built because it is easy to design and maintain, has high flexibility, has accurate location data and is powerful for computation. The thesis explores various geometric, topological and semantic features of the grid model and select out the most useful features upon tracking purposes. Among geometric features, coordinate, buffer, orientation vector and Euclidean distance are used. Among semantic features, space, obstacle, and door are employed. Among topological features, the difference between straight-line distance and shortest path distance is chosen. We develop the tracking algorithm combining multiple tracking techniques. In addition to the spatial model and WiFi positioning system, the algorithm also includes magnetometers and grid filters. The former one measures the orientation of a pedestrian. The latter one allows integrating all selected features of the grid model with the measurements from both the WiFi positioning system and the magnetometer to compute the location recursively. To test the algorithm's performance, we built a tracking system with database, web service and mobile client. Several experiments are carried out using the system in a real environment. The experiment results show that the algorithm is able to determine locations at reasonable places (in the correct space, outside obstacles and connected to the previous location) and derive the accurate moving direction of a pedestrian.