Study On Video Moving Object Detection And Subtractive Clustering Locating Technology

The most basic issue in an intelligent visual surveillance system is how to detect video moving objects. Only based on the detecting results that research works such as object classification, tracking and content analysis could be carried out. This dissertation mainly discusses the video moving object detection and object locating problems under fixed camera surveillance scenes. Contributions of this dissertation are:1) Considering the video object detecting problem in static scenes, a basic video moving object detecting method based on differences accumulation information was proposed. This method was a basic technology of other detecting and locating method in this dissertation. enhanced Otsu method was used to threshold the background subtracting image. Twice region growing operation was used for marking connective regions. Finally, moving objects were related using minimal Euclidean distance and motion trajectories were painted.2) Considering the situation that moving object detection needs much more detail detecting results, a novel moving object detecting technique was proposed adopting the Knockout technique. The foreground contour region, background contour region and unknown region could be automatically marked.3) Considering the detecting problem to poor spatial connectivity binary images, a grid-based detecting method for moving objects in video sequences was proposed. With the binary result, the image was partitioned using grid. By setting the threshold value for the number of foreground pixels and redefining the frame pixels, grid matrix was defined. And finally adjacent regions were combined together and foreground region was located.4) Object locating is a next step of detection, and it is very important. A novel object locating algorithm was proposed. Subtractive clustering was used for object locating in video sequence. The theory of subtractive clustering was analyzed. Software flowchart and realization steps of the algorithm were presented. Different locating results for different video sequences were studied. Subtractive clustering locating algorithm was also compared with region growing locating method in detail. Taking video object shape into consideration, the original round region of clustering centre of subtractive clustering algorithm was extended to elliptical region, and then a novel object locating algorithm based on elliptical subtractive clustering was proposed.5) Subtractive clustering was used for object locating in video sequences. Seven optimization techniques of subtractive clustering were proposed. When using subtractive method for object locating, the idea that different effective radius in different dimensions was considered and corresponding extended method was proposed. Techniques of using down-sampling, choosing suitable density function and redefining the clustering data set based on certain grid method were proposed. A much more accurate locating method based on fuzzy membership was also proposed. Matrix analysis theory for the calculation of covariance matrix and its eigenvalues and eigenvectors was applied to compute each object's scale and orientation parameters. Comparison of the conventional subtractive method and the proposed subtractive method to different data sets showed the superiority of the proposed approach.