Distribution-Based Weights Estimation for Map Matching Algorithms

Abstract

With the development of mobile computing, map matching algorithms are utilized to recover vehicles’ traveling routes. Existing map matching algorithms usually infer the traveling routes based on a driving cost model that considers various traveling features and driving preferences (referred to as weights). In general, the weights are estimated either empirically or through data-driven approaches. For empirical setting, the weights are determined based on field studies and must be reestimated when the driving scenarios are changed. Alternatively, in data-driven approaches, the weights are determined from the analysis of historical routes of vehicles. However, estimation bias may be introduced due to the selection of data sources. To decrease the estimation bias, this article proposes a distribution-based weights estimation method. Specifically, the weights are estimated by inferring their distributions through Bayesian inference based on the driver traces collected in a city scale. Next, a sampling-based strategy is adopted to determine the weights values from their distributions. Experimental results show that the proposed approach achieves a higher map matching accuracy than the existing approaches, and it is applicable for both linear and nonlinear driving cost models used in map matching algorithms. Moreover, the estimated weights can help to understand driver preferences in different traveling cases.