The purpose of map matching is to calibrate longitude and latitude values measured by GPS device and then project the GPS point onto a road segment. It has been widely used in navigation, data-driven traffic flow analysis and location based services. According to the characteristic that the GPS positioning errors of adjacent areas are similar, this paper presents a map matching model that combines k-nearest neighbor (kNN) algorithm with multi-layer perceptron (MLP) to correct the positioning errors. In this model, a dynamic k value is learned first by using multi-layer perceptron, and then kNN algorithm use it to extract the similar characteristic of errors of k nearest neighbor points according to Euclidean distance. By doing so, the problem arising from applying a global constant k value to the non-uniform distribution of GPS data can be overcome. The proposed model is verified by using real taxi trajectory data of Xi’an City. And the results show that the error correction performance of the model is better than kNN, Support Vector Regression (SVR) and MLP.
A Dynamic K-nearest Neighbor Map Matching Method Combined with Neural Network
2019-10-01
669573 byte
Conference paper
Electronic Resource
English
New Combined kappa -Nearest Neighbor Predictor
British Library Online Contents | 2009
|British Library Online Contents | 2015
|K-Nearest Neighbor Particle Filters for Dynamic Hybrid Bayesian Networks
Online Contents | 2008
|Nearest neighbor search for relevance feedback
IEEE | 2003
|