Research On Interested Target Extraction Of Video Surveillance Based On H.264

Video surveillance system has been widely used because it can provide so much useful information for social security management,law enforcement and people's daily life.However,the shooting device of this system collects and monitors the data all the time,so it will be a great challenge to store the huge amount of data.H.264 is a kind of video encoding and decoding technology specially developed for digital video compression.Because of its efficient compression coding ability,it has been greatly used in the field of video compression.It can greatly improve the capacity of compression and storage if using H.264 encoding technology as the storing and coding technology of videos.With the appearance and adoption of intelligent video surveillance system,more and more researchers have been exploring a more intelligent system,which can understand and deal with some simple events in a certain extent.Target extraction is a part of intelligent video surveillance system,so this paper mainly studied how to extract a interested target in the H.264 compressed domain.In this research,we studied the extraction and detection of abnormal events towards interested targets in H.264 compressed bit stream.The main work includes:(1)we studied some researches of target extraction using the area growing algorithm,used the macroblock of 8x8 encoding mode as the seed area,and its motion vector and neighborhood information as the basic criterion of area growth,finally we obtained the moving target extracted from the H.264 compressed domain.The experimental result showed that this method could extract target information effectively.(2)we studied the target tracking in a compressed bit stream using the particle filtering method.The residual DCT coefficient of macroblock was gotten by partial decoding in the compressed bit stream,then the information of moving target was obtained after squaring and summarizing the residual coefficient.Through filtering,the image of macroblock's residual with little noise was acquired and transformed into binary image.Finally we used the particle filter to track the binary image.According to the target tracking diagram and trajectory information recorded,we extracted out the speed and direction of the moving object,and input it into the generalized regression neural network,which would check if the moving target is an abnormal event,and extract the possible moving objects.(3)by decoding a few I-frames,we obtained the scene information of monitoring videos,and through the summation of various frames,we got a relatively clean background image,and then we used the color histogram clustering method to cut apart the image,calculated and summarized the trajectory information in a certain area,and marked the area as active area or inactive area according to the analysis.Finally the targets falling into inactive areas were labeled as abnormal targets and were extracted.(4)in video surveillance,there were some key monitoring areas,which were named interested areas too.Through the establishment of observation points,we calculated the distances between observation points and targets.In the defined range of distance,we observed the change of tracking diagram to the target,and if the variation was beyond a certain threshold,the target would be considered to be abnormal and need to be extracted..