| During the construction of the intelligent transportation,it is the detection for the traffic abnormal events and the extraction algorithm for key frames that is the critical and difficult problem to be solved.Currently,some methods,such as pattern recognition,statistic theory,artificial intelligent,wavelet analysis,dynamic traffic model and so on,are usually used in detecting the traffic abnormal events.However,the complex mathematic model must be established before these methods being used.And the algorithms about key frame extraction,which contain the clustering-based,motion analysis,content-based,shot boundary and so on,need to be analyzed after establishing the mathematic model.However,in some cases,it is difficult and even impossible to establish the related model.Therefore,the research about how to solve these problems without the accurate mathematic model is the hotspot currently.The specified content are shown as below:(1)Aiming at the abnormal parking events,an algorithm for detecting the abnormal events is proposed combining the correlation theory with the k-mediods clustering analysis.Firstly,the analyzable image information should be obtained from the monitored video;secondly,combining the correlation algorithms with the pre-defined threshold,the time section for parking behavior can be recognized by comparing the mean image frame during a period;thirdly,the representative image frame can be achieved via analyzing the image frame sequence in this time frame.Lastly,the method is used in detecting the abnormal parking events on highway.(2)Based on the correlation between different image frames,the thesis researches the algorithm for extracting the key frame of the monitored traffic video.Firstly,the information about video stream should be transformed into the basic information on image frame for analyzing;secondly,in accordance with the correlation between different image frames R(k,l)and self-correlation R(k)for some key frames,discriminant function g(Fk,Fl)is utilized to judge whether the frame is key;thirdly,in order to reduce the frame redundancy,the algorithm is improved on basis of sliding window mechanism;lastly,the key frames extraction experiment is conducted to verify the feasibility and effectiveness for the proposed algorithm.(3)Combining with the correlation theory,the thesis researched the algorithm for key frame extraction.Firstly,the information about image frame is achieved,and then the current frame Fl and its correlation coefficient r(k,l)and the key frame correlation coefficient r(k)are simultaneously used to judge whether the current frame is key via discriminant function f(Fk,Fl),based on which the algorithm is improved via sliding window mechanism.More importantly,the experimental results show that it is able to extract the key frame that meeting the visual characteristics,and the frame redundancy decrease when sliding window mechanism is used into the algorithm.(4)In accordance with the key frame extraction algorithm based on correlation coefficient,an effective evaluation method about the distinction between different video frames is proposed by means of Euclidean distance.Specify speaking,the similarity,i.e.the distance,between two key frames is regarded as the criterion to evaluate whether the selected frame is the best frame.Finally,the feasibility and effectiveness is verified by conducting the experiment.The innovation in this paper can be described as below:(1)The algorithm for traffic abnormal events detection is proposed on basis of correlation theory and k-mediods clustering analysis algorithm;(2)An improved key frame extraction algorithm for traffic monitoring video based on the correlation between different frames;(3)Combining with the theory of correlation coefficient,a key frame extraction algorithm is proposed firstly and improved subsequently for road traffic monitoring video;(4)The method for evaluating the effectiveness about key frame is put forward in accordance with the distinction between different frames. |
PDF Full Text Request