|M.Sc Thesis||Department of Civil and Environmental Engineering|
|Supervisors:||Prof. Doytsher Yerach|
|Full Thesis text|
Cellular technology is an integral part of everyday life throughout the world. Since the adoption in the U.S. of the E911 Standard in 2003, the accuracy of cell phone location has remained within 75 meters at 67% of cases and 150 meters at 95% of cases. By using different statistical methods with the significant amount of data available, accuracy certainly can be improved.
This study presents algorithms which we developed for mapping linear networks, such as transportation networks, precisely with the use of multiple tracks of cell phone location data that exist within the cellular networks which support location functionality.
The cell phone location data suffers from three kinds of statistical errors: random, bias and outlier. Filtering the outlier errors and smoothing the random errors is essential for the first step of mapping. A Geometric-Statistic (GS) method was developed to deal with the errors. After using the geometric-statistic method, filtered locations of the cell phones and an approximate route in about ?10 meters accuracy were acquired.
In order to improve the level of accuracy, we relied on matching polynomials curves techniques. The approximate route is cut into shorter segments and the linear and the non-linear segments are separated. A polynomial curve is matched for each segment in a fitted degree to its geometric type by using the filtered locations data of the cell phones. The matched polynomial curves are simply connected by linear lines to fill the gaps between the segments into a continuous route. At the end of the process the route is smoothed in order to receive a better representation of the mapped route which is achieved in about ?5 meters accuracy.
This study presents an innovative approach to cell phone location data. Such technology is being used to calculate real time traffic information and other location based applications, but has yet to be used to actually acquire the coordinates and nodes of linear networks. This study presents for the first time algorithms which can precisely and rapidly map wide areas without the need for more energy investment in data collection.
The range of the effectiveness of these algorithms includes areas which have cellular coverage but no mapping infrastructure (developing countries) or constantly updated mapping infrastructure in near real time (developed countries). These algorithms can also be applied on more accurate location data available in devices which have GPS receivers, and thus allow even higher mapping quality.