Trajectory-based visualization of coordinated movement data within a bounded area, such as player and ball movement within asoccer pitch, can easily result in visual crossings, overplotting, and clutter. Trajectory abstraction can help to cope with theseissues, but it is a challenging problem to select the right level of abstraction (LoA) for a given data set and analysis task. Wepresent a novel dynamic approach that combines trajectory simpliﬁcation and clustering techniques with the goal to supportinterpretation and understanding of movement patterns. Our technique provides smooth transitions between different abstractiontypes that can be computed dynamically and on-the-ﬂy. This enables the analyst to effectively navigate and explore the spaceof possible abstractions in large trajectory data sets. Additionally, we provide a proof of concept for supporting the analyst indetermining the LoA semi-automatically with a recommender system. Our approach is illustrated and evaluated by case studies,quantitative measures, and expert feedback. We further demonstrate that it allows analysts to solve a variety of analysis tasks inthe domain of soccer.