Implementation of Smart Transportation Application Using YOLO Deep-Learning Neural Network

Parking and running red lights are the major problems in traffic violations. This study presents two application systems for finding parking spaces and controlling traffic lights. An intelligent traffic light system adaptively controls the time of traffic lights so that the period of green lights is adjusted according to the real-time traffic flow. The traffic jam is thus effectively reduced. An intelligent parking system is also implemented to help drivers find parking spaces efficiently. Initially, a web camera is utilized to capture images. Hence, vehicles are identified to compute the remaining number of parking spaces. A website shows the remaining parking spaces. This system can help a user to find suitable parking spaces quickly. Therefore, parking violations are reduced. The YOLO (You Only Look Once) neural network is employed to identify vehicles, where the networks are trained through transfer learning. Only a small quantity of training images is required. The experimental results show that the vehicle identification accuracy rate can reach 93%, the recall rate reaches 83%, and the overall F measure reaches 87.7%. Therefore, the vehicle identification system can be applied to intelligent traffic light control and roadside parking spaces.