| Computer vision is a great challenge in computer science of an important research direction.With the computer has a more extensive role in the military,physiotherapy,security and video surveillance,and other fields,the impact of computers on people growing to give priority and attention more and more.Application of this technology is becoming increasingly feasible.People's living standards gradually improved,fixed camera surveillance video with its rich,intuitive and specific information in the form of expression has been more widely used,tracking and monitoring in the military field,the application of robot vision and intelligent parking lot security video surveillance,public places,shopping parks,shopping malls,schools,etc.to get good results.At the same time,uninterrupted access to vast amounts of video data,storage space is very large,according to the relevant requirements,save video files monitored date for more than fifteen days,the video surveillance system to monitor if a mall full required day monitoring,8-channel video surveillance storage space required is approximately 3.4TB,so a lot of data is simply a database generated by the mall.Thus,surveillance video data of a city,a region and the country arising from monitoring sites can't imagine.Security personnel look at these large amounts of video data to see exactly to each object in every scene,there are still some places,some places is in motion,a long time watching these massive surveillance video,fatigue is also possible miss important information.While this information seems inconspicuous,there may be the key to solve the case.Worse,in most cases,these vast amounts of video surveillance in the case has not been forced to delete the browser,because the duration of the 15-day surveillance video view again takes enormous human and material resources.So,so vast amounts of video data directly lead to the following significant issues:widespread deployment of video camera,and 24-hour continuous video recording video data showing explosive growth,artificial browsing difficult to quickly query the historical record.In addition,the current video surveillance system is limited storage space,most of the video memory will be deleted after 7 days,which means that some of the evidence appears in the video monitor also had to remove such evidence is lost.From the above,we propose a camera-based network video algorithm concentrated on big scene collaborative monitoring of the use of a camera having a partially overlapping,on enrichment panorama video show complete motor behavior.Since the projection error,time synchronization and object extraction error and other reasons,extraction between the camera target trajectory is likely to cause a mismatch.To overcome the above problems,the use of sub-image pasting,combined with effective features,the camera motion between the target trajectory match,and then match the good tracks overlap fusion.Concentrate on the algorithm,taking into account the loss of overlapping objectives and optimized to give the best arrangement,the panorama was concentrated superimposed video.For this design,the author is mainly responsible for the monitoring of the video track extraction,filling,as well as the trajectory of the interception and paste and other parts of the implementation and testing,optimization.In addition,and the function of the system analysis,from the perspective of the overall introduce the work flow of the whole system and because of the project to achieve real-time processing of video surveillance,on the processing speed is very high demand,the author also responsible for the optimization of the whole project,accelerated. |
PDF Full Text Request