Collision Prediction

The task of estimating the future trajectories of surrounding pedestrians and obstacles and determining whether any of them will intersect with a user's own future position within a short prediction horizon (typically 2–4 seconds). In assistive technology for blind travellers, collision prediction differs from simple obstacle detection because it must account for dynamic, human movement: other pedestrians change speed and direction, and their trajectories must be tracked frame to frame with techniques like Kalman filters and matched across frames with algorithms like the Hungarian method. The output of a collision predictor feeds a collision-warning interface that alerts the user early enough to adjust walking speed or take an alternative path.