Key Frame Extraction And Object Segmentation In Video Surveillance Files

With the social and economic development, video surveillance technology is increasingly widely used, such as in the residential intersection, banks, large warehouses, transportation crossroads, and so on. However, facing massive amounts of video surveillance data, it can no longer rely solely on the monitoring of people's day duties and field operations. Therefore, intelligent video surveillance and video retrieval technology is receiving more attention from the academic research community and the using departments. As a video retrieval technology, there are two key technologies:key-frame extraction and object segmentation techniques which are also increasingly affected by the concerns. This thesis mainly examines the key-frame extraction and object segmentation techniques. The main research contents of this article are as follows:1. For surveillance video files in a steady background, it is necessary to divide the moving targets into invent units.2. According to the analysis of the key-frame extraction algorithm, in the process of studying key-frame extraction algorithm which is based on the shot boundary method, another key-frame extraction algorithm which is based on the invent unit boundary method is proposed. However, the data extracted by this method is redundancy, and thus this thesis proposes another key-frame extraction method based on the area of a moving object, that is, by comparing the size of the moving object to determine the key frame.3. Through the deep analysis of the segmentation method, this thesis proposes an improved Background Differential Object Segmentation Method and an automatic segmentation method for objects based on symmetrical comparison respectively.4. Based on the methods proposed above, this thesis makes experimental system designs for the extraction of the key-frame and the object segmentation separately. In this thesis, it not only contains the basic idea of experimental system designs, but also includes the overall structure of the experimental system and the design and the accomplishment of functional modules in the experimental system. Therefore, it is easier to make a test and comparison for the different algorithms. Through the analysis and the comparison experiments, it is easier found that the extracted key-frame which is based on the area of the moving object can be a better representation of the contents of invents units, which has less redundancy of the data so that it is more convenient for video retrieval. Moreover, the split objects which are spitted by using the automatic segmentation method for objects based on symmetrical comparison can be better to satisfy the requirements of the target classification with a certain value. The profile of these objects is relatively clear and the hole is less.