An intelligent traffic light control system using deep neural network

نوع: Type: thesis

مقطع: Segment: Masters

عنوان: Title: An intelligent traffic light control system using deep neural network

ارائه دهنده: Provider: pouria maleki

اساتید راهنما: Supervisors: Abbas Ramezani (Ph.D)

اساتید مشاور: Advisory Professors: Hassan Khotanlou (Ph.D)

اساتید ممتحن یا داور: Examining professors or referees: Majid Ghaniee Zarch (Ph.D) - Mohammad Amin Ghasemi (Ph.D)

زمان و تاریخ ارائه: Time and date of presentation: 2021-9-28 - pm 6

مکان ارائه: Place of presentation: Online Room - Faculty of Engineering

چکیده: Abstract: Urban population growth has grown significantly in recent years as urban and suburban traffic has become a challenge in life and therefore traffic control and management have become very important. One of the most important causes of urban and sometimes suburban traffic is the lack of proper management of intersections. Therefore, in this thesis, with the help of intelligent traffic systems (ITS) with proper management of the sequence of phases of traffic lights in a four-way intersection with the help of the intelligent agent and with the method of reinforcement learning ( Q-learning using deep neural network) Cross the intersection in a management manner that reduces vehicle downtime and thus improves traffic. In order to train and evaluate the performance of the intelligent agent used to control the traffic lights, the SUMO simulation environment was used and in order to identify the traffic leading to the intersection of each route, the video image processing method using deep neural networks (Yolo algorithm version 4) was used To teach the Yolo algorithm, Google free dataset images have been used in 6 classes of images (cars, buses, motorcycles, etc.). Rush traffic hours with a quick calculation of the reward function, which is defined based on the waiting time of vehicles, gain a quick and at the same time complete understanding of the environment, ie the information received by the agent from the environment includes important and useful information The work done in this area has been reduced and this reduces the computational time of the neural network and this has helped to implement the system more smoothly, and at the same time the results obtained can be implemented in the real environment. This dissertation contrasts with traditional methods. Also, the use of deep neural networks for the two parts of traffic detection and traffic signal control is a new method presented in this dissertation that gives this ability to the system. Which gives the system the ability to be implemented in a practical and real-world environment.

فایل: ّFile: Download فایل