Video-based vehicle speed estimation using motion vectors in video streams

摘要

An automated low complexity video-based vehicle speed estimate is described that operates within the video stream to examine video sequences to identify and eliminate unique non-violators and / or to identify and select potential offenders in a multi-layered speed enforcement system Layers provide improved accuracy with respect to selected (too fast moving) candidate vehicles. Video motion vector clusters corresponding to a vehicle are identified and tracked over multiple frames of the captured video. The motion vector cluster motion is converted from pixel per second to real speed (eg, mph) to determine if the vehicle has traveled too fast. Estimated speed data added to the video stream data is metadata, and video segments of over-speed candidate vehicles are stored and / or transmitted for a subsequent check (eg, automated or manual).