Context-aware knowledge extraction for UV scene understanding

In the surveillance systems, Unmanned Vehicle (UV) scene inter- pretation is a non-trivial problem, because UVs need to possess human-like common-sense knowledge to correctly interpret events and situations occurring in the monitored environment. Mobile camera-related issues, such as motion blur, can further complicate scene interpretation, causing a lack of reference points that bad- ly a ects the interpretation of scene entities and situations. To this purpose, this thesis investigates the synergistic combination of video tracking with Semantic Web technologies to enhance UVs at the interpretation of dynamic scenarios. The rst part of the thesis provides a survey conducted on the methods employed to extract knowledge from the acquired structured and unstructured data. When dealing with unstruc- tured data, there is the need to de ne and process contextual data to extract high-level concepts from text. To this purpose, an approach is introduced to mine concepts from texts by building layered contextual knowledge on document terms exploiting a geo- metrical structure, called Simplicial Complex. Then, the focus switches to the knowledge extraction from multimedia data, and more speci cally, video data. To this purpose, an ontology-based approach is presented to represent the video scene as composed of mobile (i.e., people, vehicles) and xed entities (i.e., environ- mental sites and features), along with the spatio/temporal rela- tions among them. The use of the ontology reasoning can support alerting event detection... [edited by Author]