Research On Activity Semantic Recognition Method Based On Spatial Temporal Trajectories

The wide application of smart mobile terminals provides a large amount of spatial temporal data related to individual locations.Compared with the original spatial temporal trajectory data,the semantic activity trajectory data can more intuitively reflect the specific human activities.Recognizing the activity semantics of the original spatial temporal trajectory helps to understand the life pattern of the individual,provide customized services for the individual,identify the common characteristics and needs of the group,and optimize the allocation of social resources.In the past,the way to obtain the semantics of user activities was to conduct questionnaire surveys on users,which would consume a lot of manpower and material resources.The largescale popularity of GPS devices and the large-scale GPS data generated provide a new solution for activity semantic recognition.In the past,some researches identified the user's activity semantics by analyzing the spatial or temporal information of GPS trajectory data.These studies ignore the close connection between time and space of GPS trajectory data and the strong periodicity of human activities.Based on the above problems,this paper extracts the user's spatial and temporal characteristics from the original GPS trajectory data and social media check-in data,and finally digs out the user's periodic pattern characteristics,constructs a classifier and a network model to identify the user's activity semantics.The research content of this article includes:(1)Aiming at the close connection of spatiotemporal trajectories,a method for expressing spatiotemporal characteristics of trajectories based on periodic pattern mining is proposed.First,according to the time and space distance of the trajectory,density clustering is used to cluster the original GPS trajectory data and social media check-in data to obtain the activity trajectory cluster.Then,the reference spots of each active trajectory cluster are extracted,and a sequence of active reference spots is obtained.Finally,the Lomb Scargle period diagram is used to mine the user's activity semantic period for the activity reference point sequence.(2)On the basis of expressing the characteristics of individual periodic patterns,extract the spatial and temporal characteristics of the user's trajectory,and propose the use of random forest and gated recurrent unit networks to construct an individual activity semantic recognition model,and then identify the user's activity semantics.Through comparison with previous research methods,the method in this paper has better recognition accuracy.(3)Designed and implemented an activity semantic recognition system,integrating GPS trajectory data collection and simulation,data cleaning,activity semantic recognition demonstration and other functional modules to visually display the semantic recognition process and results of individual spatiotemporal trajectories.The experimental results verify the effectiveness and better recognition accuracy of the activity semantic recognition method in this paper.In the comparative experiment with or without periodic pattern features,adding periodic pattern features can effectively improve the recognition accuracy by more than 20%;In the comparative experiments with other activity semantic recognition methods,the recognition method in this paper shows better classification performance.The recognition accuracy on the two public check-in datasets of NYC and TKY increased by 10.8% and 23.7%,respectively.