Uncalibrated camera approach for vehicle speed estimation using detection and tracking

Abstract The availability of cheap and high-quality digital cameras has fostered their deployment on the roads for traffic monitoring as part of Intelligent Transport System. Implementing accurate vehicle speed estimation solution exploiting traffic surveillance footage from such widespread camera represents a credible and cost-effective alternative to traditional methods. In this work, we present an uncalibrated camera approach for vehicle speed estimation based on detection and tracking using YoloV8 and ByteTrack. The proposed method considers an average speed over a given distance by means of intrusion lines and the implementation of vector product to monitor the arrival of the traveling vehicle at expected reference locations. The experimental results demonstrated average Root Mean Squared Error and Mean Absolute Error of respectively 1.78 km/h and 1.41 km/h for vehicles traveling in the range 10 km/h – 60 km/h. These are well below the legal tolerance values for actual speed estimation systems and demonstrate the potentiality of the presented method that paves the way toward future improvement for deployment as a simple and robust solution of vehicle speed estimation.