Research On Indoor Trajectory Prediction Methods Based On WiFi Data

Researches show that people tend to spend around 80% of their lifetime in indoor environments such as shopping mall,office building,airport terminal,subway station and so on.As an important issue in location-based services,accurate indoor trajectory prediction has become a crucial research problem in many areas such as public security,convenient service,and city management.However,most of the work to date are only carried out on outdoor trajectory data,overlooking the study of indoor trajectory.Therefore,it has been a burning issue to accurately and efficiently predict the further whereabouts of indoor users that has important practical significance in driving consumption,alleviating space congestion,public safety,and other fields.With the maturity of WiFi positioning technology and the prevalence of WiFi-enabled devices,it is easier to obtain indoor trajectories.Based on a large-scale realworld indoor WiFi trajectory data,and combined with the theory of Hidden Markov model,Markov model,indoor trajectory similarity and users’ moving moving intent recognition,this thesis carried out the indoor trajectory prediction problem from spatial,temporal,semantic and user’s web behavior perspective respectively.The main contributions of this dissertation are summarized as follows:Firstly,traditional trajectory prediction methods ignore the impact of semantic features in trajectory leading to poor performance of trajectory prediction in the indoor environment.Based on the historical WiFi trajectories of indoor users,this paper focuses on the trajectory prediction problem by employing the Hidden Markov model.The optimal hidden states and hidden states sequences are first investigated,then a trajectory point clustering algorithm is proposed to reduce the size of the transition matrix.From the spatial perspective,this paper puts forward the concept of frequent sub-trajectory,which tackles the data sparsity problem in the transition matrix.After that,the influence of semantic aspect is studied and further integrated with indoor trajectory prediction.Experiments on real-world data show that the proposed method performs well in indoor trajectory prediction.Secondly,the temporal factor such as detention time is ignored by traditional methods,and high space complexity and zero frequency problem always impact the prediction accuracy of traditional methods.Therefore,this paper tries to discover the interrelationship between the temporal aspect and the accuracy of trajectory prediction.From the spatial perspective,the transition matrix size and data sparsity problem are first investigated and tackled through the sub-trajectory and partial match based variable order Markov model,an improved FP-growth algorithm is presented to mine the frequent patterns in the user specified time window.Then from the temporal perspective,the impact of detention time on trajectory prediction accuracy is investigated.Finally,the impact of the location that the indoor user currently locates on the accuracy of trajectory prediction is well studied.Experimental results show that the proposed method solves the data sparsity problem and achieves high prediction accuracy.Furthermore,traditional prediction methods overlook the similarity between trajectories,which leads to the problem of being time-consuming.This paper assumes that people’s movement behaviors are likely to be affected by the crowds,then a sequential similaritybased trajectory prediction approach is proposed.A revised Longest Common Sub-Sequence(LCSS)algorithm to compute the spatial similarity of the indoor trajectories is first presented,then a novel algorithm considering the indoor semantic R-tree is proposed to compute the semantic similarities between trajectories.After that,a unified algorithm is put forward to group the trajectories,then the clustered trajectories are used to train the prediction models.The results of the simulation experiments verify the accuracy of the method.Finally,traditional prediction methods,which mainly focus on the trajectory data itself,overlook the moving intent of users and lead to poor prediction accuracy.This paper mines users’ moving intentions behind indoor trajectories to better understand human behaviors in the indoor environment.It first proposes the concept of semantic categorization,which is used to semantically label a physical space and find the correlation between user’s open text query and the physical semantics;then the relationship between cyber queries and physical activities is studied.After that,this paper proposes the shopping intent recognition method to analyze users’ shopping intents,and the effect of semantic context on future location prediction is studied.Experimental results show that the proposed approach can accurately identify users’ shopping intents and provide ideal results for indoor trajectory prediction.