Design And Research Of The Model For Pedestrian Counting Based On Video

Pedestrian counting,as a frontier topic in the field of computer vision,is very useful in the application of real life and public security.For example,in the case of passage evacuation,human configuration of mall,and security checks at the customs,the intelligent video surveillance can take the place of the manual surveillance to analysis and understand the movement of pedestrians.Therefore,the research of the pedestrian counting based on video is getting more and more attention from public.The topic of this thesis is studying the key technology of pedestrian counting based on video which was taken in the complex outdoor scenes.Putting the videos as the experimental object,the core algorithms of moving target detection and tracking was improved,which can be used as a basis for people counting,and a model of pedestrian traffic counting was designed,which was proved that it has good real-time performance and applicability by amounts of simulation experiments.In terms of pedestrian detection,this paper used the method of background subtraction based on the adaptive Gaussian mixture model,which overcame the problem that pedestrian detection could be influenced by the slight changes in the background,and add the function of shadow detection and removal to avoid the mistaken identification of pedestrian shadow.In terms of pedestrian tracking,the mean shift tracking algorithm with Kalman filter was used,which can predict the obscured pedestrian effectively and realize the continuous tracking,and enhance the system's robustness of pedestrian tracking.Additionally,in terms of pedestrian counting,well-line trajectory method was proposed on the basis of the line method and the trajectory method,this algorithm was able to count number of people accurately under the stationary background of the video monitoring,and solve the problem of repeat count caused by pedestrians stay in video.Through a large number of videos tested on the system,the system was proved that it is adaptive to complex scenes and has high accuracy.