Evaluation of Motion Data with Uncertain 2D Vector Fields
Mathias Otto, Mykhaylo Nykolaychuk, Tobias Germer, Klaus Richter, Holger Theisel · 2010
We present a novel approach for automated video analysis of motion data. Our goal is to discover the most frequent activities in a recorded scene containing motion data with uncertainty. In this poster we show how to model such real-world data with uncertain 2D vector fields and extract their topology by segmenting the domain into regions of homogeneous motion. These regions describe the characteristic object activities in the scene. Our approach combines topological methods from flow visualization with robust object detection and tracking.