Traffic management using two deep neural networks YOLO and Mask R-CNN

Human traffic monitoring can be a difficult and wearisome task. Because the human operator in trafficmanagement center deals with the videos of a large number of cameras installed on the network. The use of automation canhelp reduce the workload of human operators and facilitate traffic management that reduces the frequent accidents andcongestion on the roads. In this paper, we have proposed an automatic traffic monitoring system that uses two deep learningalgorithms YOLO and Mask-RCN for traffic management. We used a large database of traffic camera images to train deeplearning-based models for congestion detection, traffic nodes detection, and vehicle count detection. For traffic monitoring, wehave presented a Mask R-CNN neural network. In this method, the traffic node is predicted using pixel-level segmentationmasks in classified areas. Also, this model is used to accurately extract information about traffic queues with video camerasinstalled on the street. Thus, initially, a YOLO convolutional neural network model, is used for real-time vehicle detection andclassification. The identification model is then used with a multi-object tracking system (based on IOU) to search and examinevarious traffic scenes for possible anomalies. The obtained results show that the proposed framework performs satisfactorily indifferent conditions without being affected by environmental hazards such as camera blur, low light, rain or snow.

Keywords:

Traffic Management , Deep Neural Network , YOLO , Mask-RCN.