Detection Of Moving Targets In Video Monitoring Research

Moving target detection in video surveillance is the foundation task for target tracking、traffic monitoring、behavior analysis and so on. The works such as positioning the vehicle target accurately in complex environment, real-time measuring the vehicle speed, have play an important role in intelligent and unmanned transportation systems. Moving object detection based on video is not only requires to detect the foreground objects complete and accurate, also need to achieve the real time in terms of computing speed. For this purpose, the thesis use the correlation of the video pixel in image sequence in time domain and space domain to study the moving target detection problems, it proposed two kinds of detection method:(1)moving target detection algorithm based on modified correlation coefficient;(2)moving target detection based on canonical correlation tree weighted belief propagation.The first method mainly used the correlation in time domain of video image, while the object is found, the pixel values in row(column) of two adjacent images corresponding to different physical objects, they have low similarity, however, while there is no object, the pixel values in row(column) of two adjacent images corresponding to same physical objects, they have high similarity, then the size of correlations in row(column) can be used as the similarity indicates to detect the moving target. For more complex scenes, we can combined with the hierarchical block though, it use the correlations in blocks to detect the moving target. It need to calculate the standard deviation between the two variables when calculate the correlation coefficient, because of the image pixels number is more, it had too much workload when calculate the correlation in row(column), it affected the algorithm’s feature in real time. Standard deviation reflected the deviation from the average of the data, then we can use the average value of the deviation from mean to substitute for the standard deviation, the two groups of data trends are basically identical, the values are basically same, and the average vaue of the deviation is calculating simple, it was conducive to improve the operational speed.The belief propagation algorithm used video image correlation in time domain and spece domain at the same time to detect the moving target area. The traditional belief propagation algorithm based on the features to finish prospect target detection. But traditional belief propagation algorithm is a kind of global matching algorithm, the computation is complexity and need long time,so it detect moving target not in real-time. This paper put forward a new algorithm which divide the input image into equal and independent sub-block, then calculate the canonical correlation coefficient between each sub-block,link two sub-block which has the biggest canonical correlation coefficient, occur a new loop,then decomposition the loop to determine the optimal path of the information transfer in belief propagation algorithm,this algorithm by looking for information dissemination path to modify its complexity and make sure its real-time capability.From the methods of this thesis introduced, a large number of numerical experiments indicate that the detected moving targets are complete and accurate,and its’ running time is shorter,they can achieve real-time detection of moving targets.