基于排名学习和多源信息的地图匹配方法

融合多源信息能有效提高地图匹配的准确率。已有的地图匹配方法依赖于数学模型,当引入新类型的数据时, 需要重新设计数学模型或调整模型参数。为解决该问题,提出了一种端到端的数据驱动地图匹配方法。该方法不需要建立具体的数学模型，只需从匹配结果已知的数据中学习候选道路的评分函数：选出某GPS点的候选道路，利用评分函数对所有候选道路进行打分，选择分数最高的道路作为地图匹配结果。实验结果表明,该方法能直接利用新类型的数据提高地图匹配的准确率，能在数据缺失时避免准确率急剧降低。此外，具有与基于HMM方法相近的准确率和与基于夹角特征和距离特征方法相当的速度。

An information fusion map matching method based on ranking learning

Fusion of multi-source information can effectively improve the accuracy of map matching results. Existing map matching methods rely on mathematical models. When new types of data are introduced, it is necessary to design a new mathematical model or adjust the complex parameters of those model in order to obtain the best map matching results. To solve the problem, this paper proposes an end-to-end data-driven map matching method. Instead of establishing a specific mathematical model, our method uses a neural network to learn a scoring function of the candidate roads from the data with the matching results by the ranking learning method. During matching, the method first selects all the roads near the GPS point as candidate roads; then uses the scoring function to score and sort all the candidate roads; finally selects the road with highest score as the map matching result. The experimental results demonstrate that the proposed method can effectively use new types of data and adapt to data missing conditions so as to achieve improvement or avoid sharp drop in accuracy. On the other hand, with the accuracy close to the HMM-based method, and the run time close to the angle- and distances-feature-based method.