This paper presents a novel approach to describe traffic accident events at intersections in human-understandable way using automated video processing techniques. The research mainly proposes a new technique for video-based traffic accident analysis by extracting abnormal event characteristics at intersections. The approach relies on learning normal traffic flow using trajectory clustering techniques, then analyzing accident events by observing partial vehicle trajectories and motion characteristics. In first phase, the model implements video preprocessing, vehicle detection and tracking in order to extract vehicle trajectories at road intersections. Second phase is to determine motion patterns by implementing trajectory analysis and then differentiating normal and abnormal events by defining descriptors, and last phase executes semantic decisions about traffic events and accident characteristics.