Road users detection for traffic congestion classification

One of the important problems that urban residents suffer from is Traffic Congestion. It makes their life more stressful, it impacts several sides including the economy: by wasting time, fuel and productivity. Moreover, the psychological and physical health. That makes road authorities required to find solutions for reducing traffic congestion and guaranteeing security and safety on roads. To this end, detecting road users in real-time allows for providing features and information about specific road points. These last are useful for road managers and also for road users about congested points. The goal is to build a model to detect road users including vehicles and pedestrians using artificial intelligence especially machine learning and computer vision technologies. This paper provides an approach to detecting road users using as input a dataset of 22983 images, each image contains more than one of the target objects, generally about 81000 target objects, distributed on persons (pedestrians), cars, trucks/buses (vehicles), and also motorcycles/bicycles. The dataset used in this study is known as Common Objects in Context (MS COCO) published by Microsoft. Furthermore, six different models were built based on the approaches RCNN, Fast RCNN, Faster RCNN, Mask RCNN, and the 5th and the 7th versions of YOLO. In addition, a comparison of these models using evaluation metrics was provided. As a result, the chosen model is able to detect road users with more than 55% in terms of mean average precision.