Research On Trajectory Privacy Protection Method Based On Dummy Trajectory

With the rapid development of GPS smart mobile devices and mobile computing,Location Based Service(LBS)is sought after in the application of the Internet of Things.The updating and upgrading of intelligent terminal and positioning technology is aimed at providing communication carrier for information exchange between users and service providers.When people are immersed in the convenient input continuously brought by the Internet era,they are also unknowingly exporting their personal privacy.The most direct way to disclose users' privacy is to release trajectory data directly without processing.The attacker can not only obtain the location distribution and travel trajectory of the moving object,but even carry out data mining by associating external knowledge and infer the hidden private information,such as users' life preferences and social relations.Therefore,the research on trajectory privacy protection is of great significance to the development of people and society.This thesis studies the privacy protection in the process of trajectory data release.Based on the purpose of privacy protection from different angles,two methods of trajectory privacy protection based on dummy trajectory are proposed.1)In view of the problem that the current dummy trajectory generation method does not consider the multiple characteristics of trajectory and is easy to produce unreasonable position,this thesis puts forward a kind of based on sparse representation of the dummy trajectory generation method.In this method,the original trajectory data set is preprocessed and transformed into a trajectory matrix with multiple features.Then,we use sparse representation to construct the expressions of test samples and training samples,and L2 regular terms is added to construct a new target loss function.Finally,the optimization iteration was carried out based on the method,and Top-M training samples with the highest corresponding weight coefficient were selected for each test sample.Experimental results show that this method has a better effect of generating dummy trajectory.Compared with the existing MLN(Moving in a Limited Neighborhood)and MN(Moving in a Limited Neighborhood)methods,based on different evaluation indexes,the effectiveness of the proposed method is verified based on different evaluation indexes.2)In view of the problem that most of the generated dummy trajectories can be detected by CNN network model,caused the problem of lower efficiency of privacy protection,this thesis puts forward a kind of dummy trajectory generation based on trajectory sampling points clustering method.Firstly,the original trajectory data set is preprocessed and transformed into a trajectory matrix with multiple features.Then we calculate the trajectory similarity according to the newly defined distance function,and the size of the clustering set is reduced by setting the threshold value.After that,we use the K-means clustering method to screen out the nearest neighbor set of the target again based on the set obtained above.After traversal of each sampling point,a new dummy trajectory is reconstructed.The method can ensure that the locations of sampling points in the dummy trajectory set are close at any time in the time interval.Similarly,when different thresholds are set,different dummy trajectory segments will be formed and divided into the cluster set of real trajectories.Experimental results show that: compared with the listed five dummy trajectory generation algorithms,the probability of the dummy trajectory generated based on this algorithm to be successfully detected is less than 10% when the detection rate of the above dummy trajectory is 97.2%,94.1%,86.8%,83.3% and 16.5% respectively,which achieves a better trajectory privacy protection effect.