Traffic congestion is a major challenge in modern cities, often worsened by traditional fixed-timer traffic signals that fail to adapt to real-time conditions. This project presents a web-based deep learning system for intelligent traffic signal management. Using YOLOv8 object detection models, the system processes uploaded traffic images to detect and classify vehicles such as cars, buses, trucks, and motorcycles. Based on vehicle counts and density, adaptive traffic logic dynamically adjusts green light durations (30s, 60s, 90s), reducing waiting times and improving flow efficiency. The application is developed with a Flask backend, Bootstrap frontend, and visualization tools such as Matplotlib and Chart.js, providing an interactive dashboard that displays bounding boxes, animated counts, and recommended signal timings. Unlike traditional sensor-based or simulation-only approaches, this system offers a scalable, cost-effective solution that integrates computer vision, web technology, and adaptive traffic control. By lowering congestion, fuel wastage, and emissions, the project contributes to sustainable urban mobility and aligns with smart city initiatives.