Design of traffic management system using machine learning model YOLOv7

• Main Author: Eng, Xin Fang
• Other Authors: Heng Kok Hui, John Gerard
• Format: Final Year Project
• Language: English
• Published: Nanyang Technological University 2023
• Subjects: Engineering::Mechanical engineering
• Online Access: https://hdl.handle.net/10356/167811
• Institution: Nanyang Technological University

Vehicles have become an inseparable part of modern life, providing individual mobility and enabling long-distance travel. However, the increasing ownership rates of cars and motorcycles have resulted in traffic flow management systems becoming unable to handle high traffic density, leading to traffic congestion and longer commute times. Moreover, traditional traffic control systems are ineffective in reacting to unexpected behavior from road users, leading to a rise in road accidents. To address these challenges, this project proposes an intelligent traffic management system based on YOLOv7 machine learning model to replace the existing traffic system. The proposed system focuses on the capability of the traffic system in reacting to environmental changes, optimizing traffic flow control, and improving road user experiences by implementing adjustable traffic based on real-time conditions. By using real-time traffic data, the proposed system can dynamically adjust the control of traffic flow , reducing traffic congestion and minimizing travel times. In addition, the machine learning model can recognize and predict traffic patterns and adjust traffic lights accordingly, further optimizing traffic flow control. The information and data obtained from the detection and classification of YOLOv7 can be used as parameters for designing a traffic management system that generates various control systems for traffic flow, including sequencing the green light at a junction and regulating the duration of green time. The model was optimized to improve accuracy on training data and ability to generalize new data, one of the methods being fine-tuning by changing hyperparameters to better suit the dataset and vehicle detection application. This report presents two traffic control algorithms that use different parameters for managing traffic flow, both of which have demonstrated advantages in controlling traffic and are suitable for various traffic patterns. Testing of the model and traffic management algorithm was conducted on a four-way intersection, targeting both two-phase and four-phase signals, and evaluated under different traffic conditions to test its ability. In summary, the intelligent traffic management system proposed in this study has the potential to improve traffic flow control, enhance road safety, and improve the road user experience.