Abstract
| This paper presents a method for video information retrieval considering the frames extracted from the video sources through the segmentation of each key frame, the tracking of the salient objects and the domain ontologies. Based on disambiguation algorithms are constructed domain ontologies for annotation of the video contents. An object oriented database is used to store all specific dataextracted by segmenting, tracking, and annotation of the salient objects. The presented technique combines the visual and semantic descriptors with spatial information for a good indexing of data. The object oriented approach used it was imposed by the complex structure of physical objects you can find in the videos as well as the addition of semantic meanings, which must relate to the objects detected. The annotation based on domain ontologies and the pseudonatural query language close to human understanding are a proper contribution of this article, which has led to fulfilling the final goal, the extraction of semantics-based syntax and retrieving the frames with relevant information within a pseudo-natural query language close to human understanding. The results obtained for the sport domain validate the research carried out and represents a model for applying this method on videos from multiple domain of activity. |
 |