Adaptive traffic signal control using image analysis and AI techniques

Abstract

Traffic jam is a significant obstacle that makes travel within the city very inconvenient. The systems, such as traffic lights, that are currently in place do not solve this problem. In usual scenario, traffic lights have a fixed pattern of changing from red to green to yellow with fixed timing. But, traffic in roads can be very unpredictable, which is a key point that we must introduce automated traffic signals. In this project, we proposed and designed a system where traffic lights change according to vehicle density on each road. We created an interface to demonstrate the simulation. At first, we trained our data with the YOLOv7 model but the results were not satisfactory so later on we trained with later YOLOv8 model. YOLOv8 model is a deep learning model to recognize objects for computer vision. Then we implemented OpenCV with Streamlit to complete the simulation. Streamlit is a Python based library to build web applications. We used Streamlit to create the simulation interface which takes video sources and the video feeds are handled by OpenCV. The YOLOv8 object detection model is an object detection model for real-time applications. The model has three primary sections input, process and output. If the proposed system is applied in real life then it can reduce problems of fixed timed traffic signal system.