비디오 내 움직임 객체의 의미적 표현을 통한 궤적 검색

Abstract

The recent proliferation of multimedia data necessitates the effectively and efficiently retrieving of multimedia data. These research not only focus on the retrieving methods of text matching but also on using the multimedia data features (color, sound, form) of Audio, Video, Image. Especially in the video research field, varieties of compound multimedia features have been exploited. There are two distinct research directions that related to the indexing and retrieving of video data. One research direction is by using the visual features of video objects such as color, shape and texture, and auditory features such as audio and music. The other one is by exploiting the spatio-temporal feature of objects to represent, index and retrieve in the video data. In the former area, there has been a lot of research on some specific video data, such as soccer and billiard video. but it is impossible to generalize the visual and auditory features, for every video differs with each other; In the latter one, however, temporal relation, topological relation and directional relation of the moving object in the video are able to acquire and use for making a definition of semantics for moving object Therefore, in this paper, It will specify how to give definitions to the motions of general moving objects by using the spatio-temporal relation, and match between the definition of moving objects and motion verbs in the vocabulary. To achieve this, first the spatio-temporal relation models using topological relation and directional relation have been made, then a matching of the proposed models with the directional motion verbs proposed by Levin’s verbs of Inherently directed motion is applied, finally the synonym and antonym are inserted by using WordNet. For the similarity measuring between proposed modeling and trajectory of moving object in the video, first extracted the objects, and then compare the trajectories of moving objects with the proposed modeling. Because of the different features with each proposed modeling, finally the rules that have been generated will be applied to the similarity measurement by TSR(Tangent Space Representation).