Conditional random field map matching method facing sparse floating car data

Abstract

The invention discloses a conditional random field map matching method facing sparse floating car data. The method comprises the following steps: (1) training by applying a conditional random field model based on an improved iteration scaling method to obtain characteristic function weight; (2) modeling based on map matching of a conditional random field; (3) analyzing space-time influence factors; (4) predicating based on a conditional random field model of a Viterbin algorithm; finally, obtaining an optimal path of a given GPS (Global Positioning System) observation sequence. According to the conditional random field map matching method disclosed by the invention, influence factors of floating car track map matching are further analyzed from the detail aspect of the flow of a map matching algorithm and the realization of the algorithm, and a low-frequency floating car track map matching method with better precision and robustness is developed; a condition that labeling is deviated is avoided and the matching effect is better.

Description

technical field [0001] The invention relates to a matching method, in particular to a conditional random field map matching method for sparse floating car data. Background technique [0002] Map matching is a positioning correction method, the basic idea of ​​which is to link the vehicle positioning track with the road network information in the digital map, and thus determine the position of the vehicle relative to the map. The map matching application should meet the following two assumptions: (1) The vehicle is always driving on the road; (2) The accuracy of the underlying road network data is higher than that of the vehicle positioning and navigation system. When the above conditions are met, the positioning data and vehicle trajectory can be compared with the road position information provided by the digital map, and the most likely driving section of the vehicle and the maximum possible position of the vehicle in this section can be determined through an appropriate ma...

AI Technical Summary

Problems solved by technology

[0005] ① When the hidden Markov model is used to calculate the joint distribution between the observation sequence and the state sequence, all observation sequences must be enumerable. If there are long-distance dependencies in the sequence, it is very difficult to enumerate all observation sequences of

[0006] ②, Hidden Markov model still has the problem of "label offset"

In this way, no matter what the input is, it will jump to the subsequent state, so that the matching result tends to choose roads with fewer branches. For intersections with high connectivity in the city, misjudgments and omissions occur, which deviates from the actual state. route

Images