The Research And Implementation Of Companion Recommendation Based On Trajectory Similarity

With the construction of smart cities and sky-eye projects,video surveillance systems have been installed in crowded areas such as railway stations,airports,residential areas,and urban roads,and video surveillance in key public areas has basically achieved full coverage.The mature face recognition technology enables the monitoring equipment to detect all effective faces in the monitoring in real time,and compare the captured face pictures with the pre-stored face templates to generate recognition records and extract the collection position and collection time,Identity and other information,which provides us with data support for the analysis of pedestrian's spatiotemporal trajectory.Based on pedestrian trajectory data,this thesis conducts a study on pedestrian companionship model,proposes a model of accompanying person discovery method based on trajectory similarity,and builds a practical target-oriented trajectory analysis system based on distributed clusters.The work includes:(1)A CR-TS model based on trajectory similarity is proposed.First,by filtering,simplifying,and changing the original spatio-temporal trajectory data,the potential companion set is obtained,and then the definition of point companion is given based on the fixedness of the monitoring point position,and a similar time trajectory candidate set is generated using a sliding time window,which greatly reduces the intermediate In the result set,cosine similarity is finally selected as the trajectory similarity metric,and the companions in the micro and macro time are simultaneously considered,and the companion set that meets the similarity threshold is output.(2)Designed a CR-TS trajectory analysis system for specific targets based on distributed clusters.In order to simulate the accompanying person discovery process in the real scene as much as possible,according to the CR-TS method model proposed in this paper,the corresponding parallelization mode is designed,and the distributed environment is used to realize the parallelization of the accompanying person discovery process,thereby reducing the query Processing response time.The system is based on the Hadoop cluster and uses HDFS to store pedestrian historical traffic data.The MapReduce programming model is used to achieve parallel computing of the CR-TS method.With the help of the SSM framework,the generated results are stored in the MySQL database.The data is cached,and the Baidu map API is introduced to draw the trajectory of accompanying people.(3)By designing a set of test cases,the efficiency and accuracy of the CR-TS method proposed in this paper are verified.A set of experiments based on a real pedestrian trafficrecord data set is compared and analyzed with traditional methods.The method proposed in this paper is evaluated in terms of performance,parameters,and correctness,and this method can be verified under the premise of ensuring correct results.Effectively improve the efficiency of accompanying personnel discovery.