Vehicle speed estimation by license plate detection and tracking

Abstract

Speed control systems are used in most countries to enforce speed limits and, consequently, to prevent accidents. Most of such systems are based on intrusive technologies which require complex installation and maintenance, usually causing traffic disturbance. In this work, we propose a non-intrusive video-based system for vehicle speed estimation. The proposed system detects moving vehicles using an optimized motion detector. We apply a specialized text detector to locate the vehicle’s license plate region, in which stable features are selected for tracking. The tracked features are then filtered and rectified for perspective distortion. Vehicle speed is estimated by comparing the trajectory of the tracked features to known real world measures. For our tests, we used almost five hours of videos in different conditions, captured by a single low-cost camera positioned at 5.5 meters height. The recorded videos contain more than 8,000 vehicles, in three different road lanes, with associated ground truth speeds obtained from an inductive loop detector. We compared our license plate detector with three other state-of-the-art text detectors, and our approach has shown the best performance for our dataset, attaining a precision of 0.93 and a recall of 0.87. Vehicle speeds were estimated with an average error of -0.5 km/h, staying inside the +2/-3 km/h limit determined by regulatory authorities in several countries in over 96.0% of the cases.