Research And Implementation Of Target Tracking And Counting Algorithms Based On Video Sequences

With the development of computer and video technology, video processing technology has beenwidely used in people’s daily life. There is a large number of useful information in video date whichcollects from vast amounts of monitor camera, which will bring accurate theoretical basis forscientific decision-making for emergencies in monitoring area. So it is important to get targetstraffic quickly and accurately in monitoring area. We count the targets in the base of intelligentvideo analysis, achieve an accurate count based on tracking accurately.In this thesis, we research on the real surveillance video and intensively study several key issuesaround video detection algorithm. Finally, we format a relatively complete video detectionalgorithm. First of all we use the difference of neighbor frames to establish the initial backgroundand update the background in pixel level and frame-level. Secondly we use an improved OTSUalgorithm for adaptive image binarization, which makes better binarization results under differentlight conditions. Some post-processing for binary image is necessary, such as filling cavities,connected domain identification, noise removal, and shadow, and the n get a clear image of thetargets. Thirdly, we use Kalman to predict the location of each moving target in the next frame, anduse the matching algorithm to match. Finally we get traffic statistics with the method of statisticsthe trajectory of target in a certain region, also judgment the trend towards of targets in order toachieve accurate counting purposes.