Research And Implementation Of Pedestrian Behavior Analysis Algorithm In Video Sequence

As a lot of importance has been attached increasingly to public security and the video monitoring equipment and video processing technologies are developed rapidly,the video monitoring systems are playing a more and more significant role in society.And in the intelligent monitoring field,the pedestrians' behavior analysis in video monitoring technologies is now one of the research hotspots,which has drawn a larger number of researchers' attention.The pedestrian-oriented behavior analysis are conducted mainly in such aspects as analysis of pedestrians' trajectories,identification of pedestrians' movement and prediction of pedestrians' trajectories,etc.Based on the achievements of pedestrians' behavior analysis,this thesis carries out a research into the generating and clustering algorithms of pedestrians' flow,in which the main research work and innovations are concluded as follows:A pedestrians' long-distance trajectories generating algorithm based on tracking is proposed.The existing tracking algorithms in video monitoring environment cannot achieve a balance between efficiency and accuracy,and are not suitable for different crowd density conditions.Therefore,an improved algorithm is proposed in this thesis for an effective obtaining of the pedestrians' complete movement trajectories.This algorithm is divided into two phases.Firstly,stable and reliable fragments of pedestrians' trajectories are generated by using the online tracking algorithm.Secondly,various characteristics of the trajectory fragments are used to calculate their similarities,then fragments with high similarity are linked.In the whole process of this algorithm,we manage the state of the targets precisely.In this way,the pedestrians' long-distance trajectories are generated.And the efficiency and validity of this algorithm is tested by experiment.A density clustering algorithm for pedestrians' trajectories with semantic information is proposed.In existing monitoring systems,it needs a huge memory space to store pedestrians' trajectories data.And in existing clustering algorithms,there is a lack of reasonable and effective measuring methods for the similarities of the pedestrians' trajectories and an uncertainty of the inputting parameters.Therefore,an improved clustering algorithm is proposed in this thesis.Putting the concrete semantic information obtained in the monitoring scenes into consideration,this clustering algorithm based on directed one-way distance calculates the similarities of pedestrians' trajectories and defines the inputting parameters by using data statistics features.In this way,pedestrians' trajectories will be effectively clustered and the modes of pedestrians' movement will be well established.And the rationality and correctness of this algorithm is tested by experiment on the related clustering of data messages.Based on the framework of traditional video monitoring system,an intelligent analysis platform is designed by using two algorithms discussed in this thesis.The system integrates the generating algorithm for pedestrians' long-distance trajectories and the density clustering algorithm for pedestrians' trajectories with semantic information.An overall architecture of the analysis platform is explained,an introduction to its main submodules is made and a visualization presentation of the main submodules is given in this thesis,which shows that the algorithms and the system architecture are both of good value in application.