The goal of research in semantic analysis of dynamic scenes is to make thecomputer vision system have the visual and cognitive ability to analyze, understand and represent the environment semantically. As an important issue in computer vision, more and more researchers have paid their ebullient attention on this topic, and there emerge a lot of excellent work. Some of these research work have been adopted in many real applications successfully. And this inspirer situation drive research and industry of computer vision to become a very active area. There are some necessary steps in semantic analysis of dynamic scenes. First, all related visual entities should be extracted. Next, all these extracted data should be labeled semantically and represented in a structured form. Then, temporal structured data sequences should be analyzed. After modeling activity and event, semantic representations will be obtained. Though it is become a very hot research topic, there are still a lot of important and di±cult problems in related theories and practices. In this thesis, we provide a general framework of semantic analysis in dynamic scenes to achieve semantic representation and understanding from low-level visual features. This involves many basic problems in image processing, computer vision, data mining and artificial intelligence, which include visual entity extraction and classi¯cation, semantic modeling of dynamic scenes, object detection and tracking, object classi¯cation and recognition, structured representation of information, spatial-temporal data analysis, activity and event analysis and semantic representation and understanding. The main contributions of this thesis include the following: In a word, in this thesis, we have made fruitful attempts and signifincant progresses on semantic analysis in dynamic scenes for our surveillance system.
修改评论