An intelligent traffic detection approach for vehicles on highway using pattern recognition and deep learning

Intelligent transportation system (ITS) is widely employed in dynamic traffic management to alleviate roadside congestion and increase traffic efficiency for solving the increasingly critical traffic congestion problems. With the advancement in ITS, real-time data collection of vehicles on the road has become a reality. A vast amount of traffic data ensures that the state of the road network may be analysed and predicted in real time for vehicle detection. Decision support system plays a significant role in the early decision making based on some defined criteria again available options. If the decision is made in the right and precise way, then eventually, it lead to success. This research investigates and designs a vehicle recognition algorithm and the road environment discrimination algorithm, which greatly increase the accuracy of highway vehicle detection, using a deep learning framework. In this work, we collect the highway video surveillance's images in various environments, create an original database, build a deep learning model of environment discrimination, and train the classification model to achieve real-time highway environment as the basic condition of vehicle recognition and traffic event discrimination. The proposed work uses a decision support system which is feed with basic information for vehicle detection and selection. Labeling the vehicle target and sample preparation of various environments are carried out to improve the accuracy of detecting the vehicles on a highway. The vehicle recognition algorithm is investigated in this context, and a vehicle detection technique based on weather environment recognition and a rapid RCNN model is presented. The performance of the vehicle recognition algorithm developed in this study is then verified by comparing detection accuracy with the existing state-of-the-art approaches. The comparison with these approaches in terms of accuracy, sensitivity, and F-score shows that our algorithm outperforms these approaches for detecting and classifying vehicles on the highway.