ESTIMATING THE TRAFFIC FLOW SPEED ON TRAFFIC VIDEOS USING YOLOV8 AND BYTETRACK

Abstract

Traffic flow is a critical aspect of economic, social, and environmental development. To assess traffic flow, estimating the speed of traffic is crucial. In this research, we propose a model for estimating traffic speed based on data collected from traffic surveillance cameras. The main objective is to count and track vehicles to estimate traffic flow by combining the Yolov8 and ByteTrack models, then calculating the average speed of vehicles. To train and evaluate the model's performance, data collected from the Vinh Thanh Van Police - Rach Gia City, including 10,092 images and over 96,024 labeled objects in various conditions, were used. The study experimented and compared the performance of our model with models combining Yolov8 and DeepSort. The results indicate that the proposed model has the lowest execution time and the capability to estimate traffic flow close to reality, with an accuracy of 91.39%. The dataset used in this research can be explored and utilized as a benchmark for similar problems.