Lane-level Map Building And Privacy Protection In Crowdsourcing

Lane-level map is a new generation of maps that contain the centerline of each lane.The mainstream lane-level maps building schemes are traversing the target region by expensive lane-level map acquisition vehicles,which is time-consuming and cannot dynamically update maps as the road changes.A kind of low-cost methods is to mine lane-level information from the trajectories of vehicles.This kind of methods uses the habit of civil vehicle driving along the centerline of each lane of the road and combines the corresponding model fitting methods to identify the position of the centerline.However,the existing lane-level information extraction algorithms can only be implemented on pre-processed trajectory clusters.That is,only the lane-level information of a single road can be extracted.However,due to the interlaced nature of roads,the daily trajectories shared by crowdsourcing users are intertwined,so existing methods cannot be applied to trajectory data sets shared by crowdsourcing users.The daily trajectories uploaded by users contain a large amount of personal information,such as the location of their home and the restaurants they often go to.Sharing the daily driving trajectories will lead to the disclosure of sensitive information of users,thus reducing the enthusiasm of users to share their trajectories.In order to solve the above two problems,this project proposes a lane-level maps building scheme under privacy protection,which is composed of two parts:lane-level map building algorithm and privacy-preserving trajectory sharing scheme.In order to process the trajectories shared by users into trajectory clusters that can be directly used in the lane-level map building algorithm,a trajectory classification algorithm based on road shape is designed in this paper.This algorithm firstly presents an improved discrete Frechet distance calculation method to solve the problem of the traditional discrete Frechet distance algorithm when the trajectory length is not equal.Based on the improved algorithm and the prior knowledge of road shape,trajectories can be classified as trajectories clusters according to the roads they belong to.For these trajectory clusters,this study first extracts the road features based on LSE(Least Square Estimation),then uses GMM(Gaussian Mixture Model)constrained by these features to fit the trajectories clusters,and finally uses EM(Expectation Maximization)algorithm to predict the parameters of the constrained GMM to obtain each lane centerline.In order to prevent privacy leakage when users share trajectories,this study designes a k-anonymous trajectory sharing scheme.The existing location k-anonymity technology can only achieve the k-anonymity of a single location,while a trajectory is a sequence composed of a series of locations.The relationship between adjacent locations is constrained by the real world,so the existing algorithms cannot achieve trajectory kanonymity.Therefore,this study takes the probability distribution of historical trajectories as prior information,and designs k-1 dummy trajectories that is consistent with the reality constraint for each real trajectory,so as to achieve k-anonymous trajectory sharing under the circumstance that attackers utilize background knowledge.