Spatio-temporal data analysis has important applications in transportation management, urban planning and other fields. However, spatio-temporal data are often highly dimensional, overly large and contain spatial and temporal attributes, which pose special challenges for analysts. In this paper, we propose a new visual aided mining approach, Visual Fingerprinting (VF) for extremely large-scale spatio-temporal feature extraction and analysis. It adopts a visual analytics approach for spatio-temporal data analysis that can generate fingerprints for temporal exploration while preserving the spatial distribution in a region or on a road. Fingerprinting has been proposed to display temporal changes in spatial distributions; for example fingerprints for a region grid can well display temporal changes in the traffic situations of significant spots of a city. These fingerprints integrate important statistical and historical information related to traffic and can be conveniently embedded into urban maps. The sophisticated design of the visualization can better reveal frequent or periodic patterns for temporal attributes. We have tested our approach with real-life vehicle data collected from thousands of taxis and some interesting findings about traffic patterns have been obtained. The experiments validate our methods and demonstrate that our approach can be used for analyzing vehicle trajectories on road networks.