A deep-learning-based SIoV framework in vehicle detection and counting system for Intelligent traffic management

The automatic detection and counting of vehicles can greatly contribute to improving traffic control and management. This is done by utilizing a deep neural network and classifying labeled data. The proposed model has three stages: first, to separate and count the vehicles; second, to determine whether the vehicle is slow or fast; and third, to make a decision about each vehicle. In the first step, vehicle images are analyzed using the YOLOv3 method, and the vehicles detected are counted. In the next stage, the labeled data is classified into two classes, slow and fast, using a recurrent neural network. Finally, considering the road type, number of cars, and the class of each car, a decision is made about the existing traffic situation. The proposed method was evaluated on two datasets, GRAM and HighwayII, and the experimental results show an accuracy of about 98%, which is better than other existing approaches.