Vast amounts of video data render manual video analysis useless while recent automatic video analytics techniques suffer from insufficient performance. To alleviate these issues we present a scalable and reliable approach exploiting the visual analytics methodology. This involves the user in the iterative process of exploration hypotheses generation and their verification. Scalability is achieved by interactive filter definitions on trajectory features extracted by the automatic computer vision stage. We establish the interface between user and machine adopting the VideoPerpetuoGram (VPG) for visualization and enable users to provide filter-based relevance feedback. Additionally users are supported in deriving hypotheses by context-sensitive statistical graphics. To allow for reliable decision making we gather uncertainties introduced by the computer vision step communicate these information to users through uncertainty visualization and grant fuzzy hypothesis formulation to interact with the machine. Finally we demonstrate the effectiveness of our approach by the video analysis mini challenge which was part of the IEEE Symposium on Visual Analytics Science and Technology 2009.

Creative Commons License

Creative Commons Attribution 3.0 License
This work is licensed under a Creative Commons Attribution 3.0 License.