Evaluation of Non-GPS Train Localization Schemes Using a Commodity Smartphone with Built-In Sensors

In this paper, we compare performances of train localiza-tion schemes by the dynamic programming of various sensor information obtained from a smartphone attached to a train, and further discuss the most superior sensor information and scheme in this localization system. First, we compare the localization performances of single sensor informa-tion schemes, such as 3-axis acceleration information, acoustic information, 3-axis magnetic information, and barometric pressure information. These comparisons reveal that the lateral acceleration information input scheme has the best localization performance. Furthermore, we optimize each data fusion scheme and compare the localization performances of the data-fusion schemes using the optimal ratio of coefficients. The results show that the hybrid scheme has the best localization performance, with a root mean squared error (RMSE) of 12.2 m. However, there are no differences be-tween the RMSEs of the input fusion scheme and 3-axis acceleration input scheme in the most significant three digits. Consequently, we conclude that the 3-axis acceleration input fusion scheme is the most reasonable in terms of simplicity.