Research On Outlier Detection Method Of Trajectory Data For Road Fitting

Trajectory data contains a large amount of road information with practical application value,which directly or indirectly describes the characteristics of the road network,and can be used to achieve road fitting and urban road network extraction.However,there are a large number of outliers in the trajectory data with inconsistent precision and density,which seriously affects the accuracy of road fitting and is not conducive to road network updating and urban traffic planning.Based on road fitting for the application purpose,in order to track data outliers detection as the pretreatment process,this thesis proposes a road fitting trajectory data outliers detection method,first define volatility changes in nuclear density factor description data object,based on the average nuclear density change factor to identify outliers in trajectory data and to eliminate the influence of outliers on the fitting.Secondly,the method of track point clustering is improved,the track data is mapped to the high-dimensional feature space,and multiple clustering constraints are comprehensively measured to make the track points of each class cluster have stronger similarity,so as to improve the accuracy of road fitting.The main contents are as follows:(1)In view of the large number of outliers in the trajectory data,this thesis improves the LOF algorithm by adopting kernel density calculation instead of traditional density calculation,and considering the overall distribution characteristics of the data from the data itself.The kernel density change factor is defined to describe the difference of kernel density values of data objects,reflect the density characteristics of data objects themselves and their density changes compared with other data objects,and is used to discover local anomalies in data sets.The mean kernel density change factor is defined to describe the difference of the kernel density change of data objects,which is used to discover the collective anomaly in the data set.The experimental results show that the proposed algorithm has significantly improved the detection effect compared with the comparison algorithm,and has a high recall rate for the detection results of trajectory data,which meets the requirements of practical application.(2)Too many constraints on the clustering of track points lead to too many parameters of the clustering algorithm,which affects the clustering effect of track points.Aiming at this problem,the clustering constraints of DBSCAN algorithm are improved.First by kernel function to track points in lower dimensional feature space distance measurement is converted into a high dimensional feature space by nuclear distance measurement method,the trajectory point distance and direction Angle at the same time as the constraint condition,is put forward based on the distance and direction Angle of N-DBSCAN clustering algorithm,realize the track points in position and direction under the constraint of clustering.Then,based on the similarity clustering results,the path fitting of the track data in the class cluster is carried out by the method of multiple adaptive regression spline.Experimental results show that this method can effectively divide the track points according to their characteristics,and the accuracy of road fitting results is high.The experimental results show that the outlier detection method for road fitting can effectively identify outliers in the track data and achieve more accurate road fitting for the track data,which has certain reference value for modern urban traffic planning and map update.