Detection And Tracking Of Moving Targets In Intelligent Video Surveillance System

Since the 2010,the rapid development of the Internet in these years,the application of intelligent monitoring system become more and more scenes,Hikvision,Dahua and foreign Honeywell and other security companies have launched a variety of monitoring equipment products,from the previous local hard disk cameras to the current network camera,to household Smart cameras and intelligent video surveillance systems play a great role in our safety,and are also conducive to public security and social stability.Detection and tracking of moving targets is the key technology in the application of intelligent video surveillance system.Based on the detection and tracking of mobile targets,the system can not only obtain motion features,such as motion direction and speed of moving targets,but also provide corresponding data letters for further classification of moving targets,behavior understanding and so on.Interest plays a key role in the intellectualization of video surveillance system.It has a very important research significance.The target and simulation object of this thesis are video based on the video surveillance system under the static background,and the video can be regarded as the image sequence on the time axis.Therefore,this thesis first analyzes the theoretical knowledge of target detection and tracking,including the overall framework of target detection and tracking,the preprocessing technology in digital image and the extraction of target features,which lays the foundation for the research of target detection and tracking algorithm in the following chapters.In addition,the main object of this thesis is the technology and algorithm of moving object detection and tracking in video sequences.The target detection is to confirm whether there is a moving target in the video sequence collected by the video surveillance system,while the target tracking is to track and locate the location of the moving target in the video sequence in the last step.In the mobile target detection part,this thesis analyzes and compares the detection and recognition algorithms of the moving objects detection,which are widely used at present,such as optical flow field,frame difference algorithm and background difference algorithm.According to their respective algorithm principles and simulation results,their advantages and disadvantages are analyzed.Based on the analysis and study of the above algorithm,a target detection method based on three frame difference method and adaptive mixed Gauss model is proposed in view of the traditional inter frame difference method and background subtraction method which affects target detection in complex background and illumination change.This method first uses adaptive Gauss mixing.The model dynamically constructs the background model and detects the moving target area by the background subtraction method.Then the moving target area is obtained by the difference operation between the three consecutive frames of the adjacent frames by the inter frame difference method.Finally,the candidate target regions obtained by the two methods are integrated and finally the motion order is extracted.The mark position.The method combines the advantages of the three frame difference method and the adaptive hybrid Gauss model detection method.It can quickly and accurately establish and update the background model,overcome the illumination change and background noise and interference,and accurately detect the moving target in the video.In moving target tracking and positioning,this thesis first analyzes the commonly used moving object tracking algorithms.The mean shift tracking algorithm,Calman filter and particle filter algorithm based on target feature tracking are mainly analyzed.A moving target tracking method combining Calma filter and mean shift algorithm is proposed to improve the accuracy of system tracking when the target is obscured by means of the mean drift tracking algorithm when the tracking result is offset and the accuracy is reduced when the target is blocked.The result is much better than that of the mean shift algorithm.