Moving Object’s Denoise, Segmentation And Detection Algorithm Based On Video Sequence

Recently, the public security become more and more rigorous. As an essentialtechnique in the field of security, intelligent surveillance system is widely applied in manysituations such as school, hospital, band and prison, et al. Currently, the intelligentsurveillance system can only record the events that happen in the scene. It can not processand analyze the signal effectively. Hence, in this thesis, we pay our attention to the filter,segmentation and detection of video signal. The main contains are listed as follows:The video signal is frequently corrupted by noise in the process of capturing,transmitting, and store, and therefore, the quality of video decrease dramatically. Toimprove the quality of video signal, two improve video filtering algorithm are proposed inthis thesis. First, an adaptive filtering algorithm is presented. In the presented algorithm,the shape, size and orientation of filtering window can adjust adaptively. The presentedalgorithm overcomes the shortcoming of traditional filtering algorithm. Second, a mixedfiltering algorithm is proposed. The mixed filtering algorithm first identifies the pixel iscorrupted by Gaussian noise or Salt&Pepper noise, then an appropriated filteringalgorithm is selected to filter the noise. Compared to traidontonal filter algorithm, theproposed mixed filtering algorithm can effectively deal with the filtering problem ofvidieo signal, whichi is corrupted by mixed noise.Segmentation is an important issue in video processing. In this thesis, two videosegmentation algorithms are proposed. First, many artificial objects can be detected ifbackground subtraction method is used when the contrast is low in the video signal. Toovercome this problem, an algoirthm for removing artificial objects is proposed based onTsallis entropy. The porposed algrithm considers both the information difference andcorrelation between object and backgroud, and it effectively solves the problem of videosegmentation in the case that the contrast is low. Second, an efficient algorithm forsegmenting multi-object is proposed. The proposed algorithm is baded on the multilevelthresholding method utilizing exponential smoothing.In process of object detecting, the W4algorithm can not update the backgroundeffectively when an object moves into background. To overcome this shortcoming, an new moving object detection algorithm is proposed based on adaptive background subtraction.In the proposed algorithm, the background classification and background updating rule isdefined in a different way. Experimental results show that the performance of the proposedalgorithm is superior to W4algorithm. On the other hand, the SIFT feature based objectdetection algorithm is very slow. In this thesis, a novel matching algorithm is proposedbased on regional covariance using the invariant moments and affine differential invariantsas features.Finally, the above research results are applied to a resident’s monitering system. Theresults show that these algorithm can get a better performance.