Real-time Smart Traffic Monitoring System Using YOLOv8 & Deepsort
Real-time Smart Traffic Monitoring System Using YOLOv8 & Deepsort
Authors:
Putangar Mahesh1, Juturu Mounish Reddy2, Sairamanajaneyulu Reddy3, Gollamandala Yesudas4, Asst. prof Ayush Panday5
1,2,3,4 CSE Department of Computer Science, Sandip University, Nashik, Maharashtra, India.
5 CSE Department of Computer Science, Sandip University, Nashik, Maharashtra, India.
Abstract - This paper presents the design, optimization, and empirical evaluation of a low-latency and scalable artificial intelligence (AI) pipeline for real-time vehicular traffic monitoring. The system integrates the YOLOv8 object detector with the DeepSORT multi-object tracking (MOT) framework to generate reliable, fine-grained traffic intelligence suitable for Intelligent Transportation Systems (ITS). The pipeline autonomously detects and classifies four major vehicle categories—cars, buses, trucks, and motorcycles—from high-resolution CCTV streams. Persistent identity assignment is achieved using DeepSORT, enabling accurate estimation of vehicle count, lane occupancy, congestion density, and instantaneous speed. Experimental evaluation shows that YOLOv8n achieves 92.3% mAP@0.5 at 31 FPS, while YOLOv8m achieves 95.1% mAP@0.5 at 23 FPS on an NVIDIA RTX 2060 GPU. DeepSORT attains an IDF1 score of 84.7%, confirming robust identity consistency. The overall system performance demonstrates its suitability for real-time deployment in large-scale ITS environments.
Key Words: YOLOv8, DeepSORT, Traffic Monitoring, Smart Cities, Multi-Object Tracking, Re-Identification, Kalman Filter, Adaptive Traffic Control.