Mitigating location and speed errors in floating car data using context-based accuracy estimation

Although the floating car data contains detailed behavior of vehicles, most researches utilizing the floating car data mainly focus on providing link-level, large-scale traffic estimation because raw floating car data usually contains location and speed errors due to the characteristics of on-board sensors such as GPS and tachometers. Such errors are usually vehicle-dependent and considered hard to be eliminated. To tackle the challenge, in this paper, we propose a method to mitigate location and speed error in raw floating car data. The method takes two different information sources from GPS and tachometers respectively, and relies more on “more dependable” source. The dependability is assessed based on the locations and situations where the data are observed (e.g. in urban canyons the accuracy of GPS decreases and tachometers are more reliable). By analyzing and modeling such characteristics, we eliminate errors contained in those information sources to obtain more accurate traces and behavior of vehicles. We have evaluated our method using the real floating car data obtained over 7 days from 4777 vehicles that have installed commercial on-board navigation systems. We have shown that the relative distance errors have been reduced from 3.84% to 2.86% in 81.6% of vehicles.