The paper presents progressive clustering as a method for visually-driven analysis of large trajectory datasets. Progressive clustering proceeds through iterative steps where simple, interpretable distance functions are applied sequentially, with visual assessment guiding refinement decisions. The workflow addresses four successive analytical questions through progressive refinement:
Phase 1: Destination clustering with density adaptation - Progressive clustering by trip destinations with iteratively decreasing sensitivity handles uneven spatial data density, identifying major destination regions across varying densities. Phase 2: Route analysis of selected subsets - Examining major destination clusters by route similarity reveals patterns within destinations, such as prevalence of low-displacement trips. Phase 3: Pattern verification on complete dataset - Observations from subset analysis lead to hypothesis testing: clustering ALL trajectories by start-end proximity verifies pattern prevalence across the complete dataset. Phase 4: Route analysis of refined subset - After excluding verified patterns, clustering remaining trajectories by route similarity identifies route types characterized by spatial structure and frequency, revealing route diversity and repetition. The progressive approach enables sophisticated analysis while maintaining interpretability, showing how the analyst's understanding evolves through exploration. The method was demonstrated on Milan car trajectory data, revealing movement patterns and route organization.