Study Trajectory Prediction And Crowd Density Forecast Based On Position Big Data

As indoor positioning accuracy becomes higher,location data become more reliable,indoor location data mining has gradually become the research direction of many scholars and researchers.The research of pedestrian prediction in indoor environment is becoming more popular.This paper mainly completed the following aspects of the work:(1)Firstly,the foreground and application value of location prediction are analyzed in a comprehensive way.The research value and practical value in this direction are evaluated from the precision pushing of indoor advertisement service providers,indoor navigation service and population density estimation in indoor areas.Survey on the research of trajectory position prediction has been done,and the development of the research in this direction in China and abroad has been analyzed.(2)The source of indoor location data is analyzed and several indoor location algorithms are introduced.Then several widely used position prediction model were analyzed in detail,comparing their respective advantages and disadvantages.Finally,several common methods for evaluating the similarity of trajectories are analyzed and investigated.(3)The links between POI,pedestrian travel motivation and stay points is explained,extracted continuous pedestrian trajectories into discrete trajectories unitted by POI,removing the reduntant data and leaving semantic information of motivation of pedestrian.The method of extracting POI from the cluster of stay points is proposed and verified.A multi-layer method of calculating the similarity of trajectories is proposed,Multisimi.Through the efficient Euclidean distance method and the LCS algorithm,efficient and accurate trajectory similarity calculation is achieved,and frequent access patterns are found.The 2-order Markov model is applied to position prediction in frequently visited trajectory mode,Multisimi-Markov algorithm is proposed and find a solution to the problem of probability conflict,the forecast accuracy is 16.1%higher than 2-order Markov model.(4)By using Multisimi-Markov algorithm,population density estimation is studied.Time-of-arrival estimation methods based on least square method either historical periodicity is proposed,then the population density is estimated,with an accuracy rate up to 62.22%.Later,on the basis of positioning the typical indoor location network-side platform,for the data storage and management,the existing basic data storage module is improved.For the problem that the performance of relational database cannot meet the demand of position big data analyse,we apply HBASE,non-relational distributed database,and Redis cache module.Deploy population density estimation algorithm and do the pressure test experiments on system,the average response time of the system is about 68.7%.