Research On Algoirthms For Moving Objects Detection In Video Sequences

With the fast development of society, the security problem has beenpaid more and more attention, and video surveillance has been wi dely used.However, comparing with the traditional video monitor, which record visualfacts to obtain evidence after the event, intelligent analysis module was addedto intelligent video surveillance system, the modules can real-timely captureand analysis the movement behavior in image sequence. Especially, onunusual behavior, intelligent video surveillance system can give timelywarning. Moving object detection is a process that detect s and segmentsmoving object in video image sequence. The accuracy of the object detectionresults is the major premise of target s tracking, targets classification andbehavior understanding.There are many factors affecting the video images quality during theprocess of collection, transmission and storage, video image pre processingcan effectively improve the accuracy of the moving objects detection.Median filter is a traditional method of video image preprocessing. For thedefects of standard median filtering algorithm on balancing noise removingand preserving image details, this study proposed an improved imagepreprocessing algorithm which was based on the threshold value judgment.The algorithm firstly tests the noise points on picture and then made ajudgment. It then took a smoothing pixel filtering processing for noise points,but not for non-noise points. This algorithm used threshold value judgmentmethod both in testing and dealing with noise point s. Fully considering thesimilarity of gray value of current pixel and neighborhood pixels, it avoidedthe case that it made wrong judgment from signal point which had theextreme value to noise point. This algorithm is especially suitable for theimage where noise of high concentration and many extreme value points exist.Experimental results show that the algorithm we proposed in this study made a great filter performance for each noise levels, and retained edges details ofimage very well.In background subtraction method, it is a key factor whether thebackground model is established and updated accurately. For thedisadvantage that the Gaussian mixture model cannot deal such cases astargets long time staying and light mutations, we proposed a moving objectdetection algorithm that based on the improvement of the Gaussian mixturemodel. The new algorithm combined Gaussian mixture model with framedifference method, and judged whether there were moving objects in picturethrough the frame difference. If there were no moving objects, our algorithmwill not update the Gaussian mixture model, so that even if the objects stayedfor a long time, it could still be detected. At the same time, the new algorithmintroduced the normalized cross-correlation coefficient, which is used todetermine whether there is a dramatic light changes in current frame bycalculating the correlation of gray value between background frame andcurrent frame. If there is a correlation, we will weaken it to reduce theinfluence. The experimental results show that, compared with the originalbackground modeling method of Gaussian mixture, the improved method weproposed can eliminate the above two shortcomings well. Meanwhile, theexperimental results of moving objects detection that combined withimproved video image preprocessing method based on median filtering andimproved Gaussian mixture model show the effectiveness of thepreprocessing.