A novel video data model for moving object description based on spatio-temporal relations

摘要

Video processing has been more and more concentrated on moving objects in the video. Video objects refer to semantic real-world entity definitions that are used to denote a coherent spatial-temporal region and be automatically computed by the continuity of spatial-temporal low-level features, such as color and motion. So, in this paper, we propose a video data model to describe events and actions performed by moving objects. This model is flexible, to support the moving objects spatio-temporal relations for semantic concepts description and low-level feature extraction. In order to cater to the users' query by video content instead of raw data, we decompose the video object action (VOA) at semantic level into an elementary video object motion (EVOM) for extracting the low-level features. The video data model, based on moving objects, can bridge the gap between semantics and low-level features.