Pedestrian Behavior Pattern Recognition And Analysis Of Indoor Location Data

The indoor space area is the place where human activities are most concentrated and active,and it is also the space where most people produce and live.The layout of the interior space contains a reasonable division of the geospatial location of each functional area,so that its spatial distribution characteristics meet its functional requirements.Therefore,with the advancement of social science and technology,indoor positioning technology has become more and more widely used in our daily life.At this stage,it has a good application prospect in daily life,business services and public safety.However,the indoor environment is often very complicated.The methods and materials used in constructing the indoor space are different.Therefore,due to the influence of building materials and internal signal interference devices,the sensor signals we receive cannot always maintain a stable status.Therefore,how to achieve accurate positioning in a complex and indoor environment is an important direction in the field of indoor positioning technology research,and it is also the current hot research direction.However,after obtaining higher quality position information,there are still many problems in the field of indoor positioning,which are worthy of research and analysis.For example,the analysis of indoor pedestrian behavior patterns is based on the actual application needs of the continuous enrichment of indoor venues.How to use indoor positioning data to make us obtain deeper and more diversified information,and to dig out the rules behind location information will have important application value and theoretical significance.According to the development of indoor positioning technology in recent years,the three directions that need to be studied in depth are:(1)the accuracy of indoor positioning data in the actual environment;(2)the research and analysis of pedestrian behavior patterns in indoor places;Application calculation of prediction models that meet different needs according to indoor locations.This paper starts from these three directions,and proposes an optimization algorithm for the important part of each problem,and applies it to the clustering of pedestrian patterns and the prediction of pedestrian behavior patterns.A large number of experiments The results verify the validity and applicability of the proposed method.The main research contents of this paper are as follows:1)Data preprocessing and data model organization By calculating the indoor positioning data and the indoor map coordinates into a consistent coordinate axis,and eliminating the non-meaningful positioning data in the original data set that does not have a valid active time length and a large geographical deviation,the data abnormal value check is performed.The outliers in the raw data are detected by means of data statistics histograms,Voronoi mapping,and semivariogram/covariance clouds.A storage model that combines time and spatial data through an efficient retrieval.According to the characteristics of indoor positioning data,the data structure,data coordinates and data content are analyzed,and the positioning accuracy index,attribute integrity and consistency are selected as the basic attribute standards of indoor positioning data,and the pedestrian status feature should include the characteristic attribute FA.(Feature Attribute),Feature Relationship FR(Feature Relationship),and Feature Behavior FB(Feature Behavior)establish a data model for indoor positioning data.2)Feature extraction and error analysis of indoor pedestrian trajectory data According to the positioning data coordinate drift phenomenon in which there is a high frequency in the indoor positioning data,and often occurs in the positioning data set with long duration and small position change,this paper first needs to define related to these stay points.After the dwell point is established,the obtained distance threshold and time threshold are applied to the trajectory recognition with higher credibility according to the relevant definition algorithm.The drift phenomenon often occurs in the dwell trajectory,so there are some in the process of extracting the drift point.Obstacles and difficulties,such as whether the drift of the positioning data is a real existence behavior,so by extracting all the trajectories containing drift points,the trajectories are divided into three types,one of which is used as research data.And the standard deviation and the average value are respectively counted,1)the dwell track with drift point,2)the dwell track of the drift point,3)the centroid of the drift point to the centroid of the dwell track of the drift point,and then combined with the AP sensor The distribution structure is used to verify the experimental results.From the perspective of the drift point,the factors caused by the error and the error distribution state are deeply discussed.Based on the experimental results,it can provide strong support for the reasonable solution to improve the error in the future.3)Clustering analysis of pedestrian behavior patterns based on indoor positioning data According to the various attributes of pedestrians and indoor shops,a new algorithm is proposed for the behavioral similarity of pedestrians.The relationship between pedestrians and shops is used to establish the weight matrix,and based on SOM(Self-organizing Maps)algorithm.The pattern clustering analysis of pedestrian trajectory is very sensitive to the SOM algorithm for the similarity measure.The improved similarity algorithm is used to optimize the input number of SOM algorithm,which will make up for its own shortcomings.The advantages of SOM algorithm can be very good.Apply to isolated points.Through the clustering results,indoor pedestrian trajectories can be divided into six categories,and an interactive recommendation algorithm is proposed for each type of pedestrian trajectory,thereby enhancing the interaction between indoor pedestrians.4)Pedestrian behavior prediction based on time series trajectory data The predictive model established by LSTM(Long Short-Term Memory)can not only nonlinearly map time series data,but also predict the input information obtained over a period of time.Pedestrians' indoor behavior patterns are often subject to various external factors.Impact.For example,as an external influence factor,holidays have a significant increase in indoor pedestrian traffic,but the degree of influence on specific time periods varies greatly.The indoor traffic flow data is used to predict future indoor traffic,in the future indoors.Targeted contingency plans can be implemented in the venue based on the pattern of pedestrian behavior obtained.The indoor pedestrian behavior trajectory data is applied to the CPT(Compact Prediction Tree)algorithm to predict the pedestrian behavior pattern,and the accuracy of the obtained prediction result cannot meet the actual application requirements.Therefore,the data source is optimized by compressing the pedestrian behavior trajectory,and then Optimization of the CPT model algorithm.The prediction tree structure is optimized and the stay duration attribute is introduced as an important reference for trajectory prediction.The improved prediction tree algorithm has more accurate predictions than the same type of prediction model.In order to further study whether there is a relationship between the next behavior destination and the previously viewed store after the pedestrian performs activities in different types of stores,and the difference between the store having a strong relationship and the store having a weak relationship.Through the obtained prediction results,the store sequence in the pedestrian trajectory sequence is discussed and analyzed.