Datenbestand vom 13. Juni 2019
Tel: 089 / 66060798
Mo - Fr, 9 - 12 Uhr
Fax: 089 / 66060799
DER VERLAG IST IN DER ZEIT VOM 12.06.2019 BIS 23.06.2019 AUSCHLIESSLICH PER EMAIL ERREICHBAR.
aktualisiert am 13. Juni 2019
978-3-8439-1491-8, Reihe Informatik
Scalable Visual Analytics for Video Surveillance
290 Seiten, Dissertation Universität Stuttgart (2013), Softcover, A5
Today, analysis of video surveillance data is challenged mainly by two issues: the vast amount of recorded video data and the complexity of the entities, actions, and relationships captured by the video sequences. While the first issue renders traditional monitoring impractical, the second prevents the application of automatic analysis methods. The search for solutions to this complex information processing problem is a vivid research topic in computer science. This thesis investigates an approach of dealing with the mentioned dilemma by combining the strengths of traditional monitoring and automatic video analysis. It therefore follows the methodology of visual analytics proposing the integration of human and machine analysis facilitated by visualization and human computer interaction. However, with the users in the analysis loop, additional questions and issues arise that challenge the perceptual capabilities of the human users, their situation awareness, and the appropriateness of their trust in automation. Hence, to enable visual analytics of vast amount of complex video data the challenges of technical and perceptual scalability have to be faced as well as an appropriate human computer interface has to be defined. This thesis tackles these challenges through the consideration of three main topics: i) reliable algorithms for automatic video analysis, ii) coordinated and multiple presentations of different aspects of the data, and iii) technical/perceptual filtering techniques.
Aside from the inspection of individual components, this thesis introduces a complete visual analytics pipeline that is capable of addressing the mentioned challenges of video analysis in general and video surveillance in particular. The pipeline is based on a refined version of the visual analytics model. The stages of this pipeline as well as the design of the particular components of its prototypical implementation are determined by the tasks and challenges of video surveillance compiled from a large body of closed circuit television (CCTV) studies. Evaluation of the individual components with respect to effectiveness, efficiency, and user satisfaction is a substantial contribution of this thesis. Besides such component-based evaluation, the capabilities of video visual analytics on the system level are assessed by challenge-based evaluation. Finally, the typical workflow of video visual analytics is illustrated by two usage scenarios.