Research On Spatio-Temporal Correlation Analysis And Visualization Of Crowd Behavior In Video Scene

Obtaining spatio-temporal correlations of crowd behavior in complex scene videos is essential for studying crowd movement behavior.Complex scene videos contain various areas while the crowd movement in each area has multiple features as the videos change over time.Understanding crowd movement in long videos by manual approach is time-consuming.To obtain the correlation of long video crowd movement patterns in different temporal granularity and various areas,this paper adopts the correlation analysis method based on video scenes and visualization of crowd spatio-temporal behavior with the purpose of improving the effectiveness and accuracy of crowd movement video understanding.The main content of this paper can be listed as follows.1.This paper proposes a crowd behavior description method in complex scenes based on spatio-temporal thematic semantics and builds a spatio-temporal semantic reasoning atlas.First,the video scene is segmented to make the crowd behavior expression reflect the changing scenes better.Then,the crowd movement bag of words is constructed through low-level features and scene features,and the features in the bag of words are combined to form high-level crowd behavior features based on spatio-temporal semantics.The crowd behavior features obtained by this model classify the crowd behavior of complex scenes.Finally,according to the construction principle of the event triad of matter mapping,the spatio-temporal attributes of video scene crowd behavior are constructed into event triad.According to the matter logic of video scene crowd behavior development,the relationship between the entities of the triad is obtained,and the relationship between the crowd behavior patterns in each video scene region is more comprehensively displayed.2.An adaptive time-granularity spatio-temporal correlation analysis method is proposed.The video crowd behaviors of complex scenes are strongly correlated in both time and space,and the spatio-temporal correlations of crowd behaviors presented by different temporal granularity are different.The constructed spatio-temporal semantic crowd behavioral reasoning atlas will be different.In the first place,this method performs adaptive temporal granularity division of video crowd behavior information.Then,we obtain the spatio-temporal data of crowd behavior based on spatio-temporal basket method,and finally use frequent item mining method to obtain frequent item sets,and uses association analysis method to obtain association rules between crowd behaviors to reveal the inner spatio-temporal laws of their occurrence.3.A spatio-temporal crowd behavior visualization method is proposed.The spatio-temporal semantic crowd behavior reasoning atlas constructed by complex scene videos under different temporal granularity has high-dimensional characteristics.This paper proposes a correlation visualization method that includes both single scene and multiple scenes,which can realize the transformation of information from high-dimensional to low-dimensional following the logic of video scenes from whole to local,crowd from individual to the group,and behavior from lowlevel features to high-level semantics,making the way of visualizing crowd behavior of video scenes more diversified.