AI-Powered Simultaneous Multi-Vehicle Speed Estimation for Intelligent Traffic Monitoring in Developing Regions Using YOLOv7 and DeepSORT

The development of intelligent, cost-effective vehicle speed monitoring systems is critical for enhancing road safety, improving traffic regulation, and enabling efficient law enforcement particularly in developing regions with limited infrastructure. This paper introduces a robust, AI-based framework for real-time speed estimation of multiple vehicles from monocular video streams. The proposed system integrates two advanced deep learning models -YOLOv7 for high-precision vehicle detection and DeepSORT for consistent multi-object tracking- ensuring accurate localization and identity preservation across frames. Speed estimation is performed by measuring the time it takes for each vehicle to travel a predefined distance between two virtual reference lines. The elapsed time is calculated based on frame count, and speed is derived using the basic motion formula. Experimental results show that the system achieves 100% detection and tracking accuracy, with an average speed estimation error of less than 3%, outperforming comparable methods in terms of efficiency and precision. The study also identifies and discusses key factors affecting estimation accuracy, such as frame rate variation, distance measurement error, and line placement precision. The approach’s simplicity, accuracy, and use of open-source tools make it well-suited for deployment in resource-constrained environments. Future directions include bidirectional speed tracking, integration with vehicle classification systems, and the use of license plate dimensions for dynamic calibration—offering a scalable foundation for intelligent traffic surveillance.