Research On Key Techniques Of Extracting The Number Of Lanes Based On Floating Car Data

High precision road map is the key to the machine-oriented intelligent auxiliary driving and automatic driving,and lane-level road detailed information is indispensable parts of the high precision road map.At this stage,lane-level road detailed information is totally depended on road information acquisition vehicle to collect the information of road,this collection method has problems such as high cost,long production and maintenance period and so on.With the development of floating car data(FCD),if we could extract the lane-level road information form the FCD,we could reduce the cost,shorten production and maintenance cycle that has high relevance and utility.In this paper,FCD of Wuhan is the data source.We use FCD to filter GPS trajectories,use Gaussian mixture model(GMM)with constraints to model FCD after filtering,detect rough lane-width and extract the number of lanes.Main content include:(1)Filtering massive,low-frequency FCD.First,considering the consistency of GPS trajectories,we split and filter GPS trajectories taking account of similarity of trajectories’ angle and distance.Then we adopt density-based clustering based on Delaunay triangulation network to cluster FCD,select high density trajectory points.(2)Modeling FCD using GMM with constraints to extract the lane number.We model FCD using GMM with constraints,select the optimal model through entropy function,extract the number of lanes from Gaussian parameter.(3)Programing to realize the lanes number extracting system.We program with C#and ArcGIS Engine,store and read data with SQL Server database,design Ribbon interface with DevExpress to realize the lane number extracting system.This paper filters and models FCD of Wuhan,detects rough lane-width and extracts the number of lanes.The results of analysis indicate that our method extracts the number of lanes from low-precision FCD,reduce the cost of collecting the number of lanes,shorten production and maintenance cycle.