An adaptive road centerline extraction method for different trajectory data scenarios based on combinatorial optimization
Vehicle trajectory data is an important data source for road map update.Extracting road centerlines from the disor-dered trajectory points or trajectory lines,and generating a structured vector map is a key step for road network generation and update based on trajectory data.The existing methods of road centerline extraction mainly use a single curve fitting algorithm,which are not adaptive to different data scenarios,especially for complex road structures and trajectories of different quality.In addition,compared with the professional collected high-frequency trajectory data,road centerline extraction based on the low-frequency trajectory data collected by float cars is still challenging due to the noise,sparsity,and low position accuracy.There-fore,this paper proposes an adaptive road centerline extraction method for different trajectory data scenarios based on combina-torial optimization and divide-and-conquer strategy.Based on preprocessing and clustering of trajectory data,this method clas-sifies the trajectory data according to its distribution characteristics.Then,the optimal fitting algorithm is matched according to different data scenarios,and the ideal road centerline is generated by combinatorial optimization strategy.This method in-tegrates the advantages of different fitting algorithms,and can effectively solve the road centerline extraction problem for dif-ferent data scenarios such as sparse data and complex road structures(e.g.self-intersection overpasses).Experiments on floating car data in Beijing,China,were conducted and results show that the average position accuracy of the roads generated by this method is 1.24 m,which is significantly better than the existing available methods.
万方数据
维普
