EnsembleNet: a hybrid approach for vehicle detection and estimation of traffic density based on faster R-CNN and YOLO models

Due to static traffic management regulations on roadways, traffic flow may become congested as it has been growing on roads. Estimating traffic density impacts intelligent transportation systems as it helps build efficient traffic management. Vehicle recognition and counting are two main steps to estimate traffic density. Vehicle identification systems can use motion, handcrafted features, and convolutional neural network (CNN)-based methods. The utilization of deep learning technologies is increasing daily with the popularity of CNN. Different classification and detection models have been developed using transfer learning. In this study, data are collected from several open-source libraries, including MB7500, KITTI, and FLIR. Image annotation has been done to classify vehicles into different categories. Various data augmentation methods are implemented to increase the dataset size and to reduce class imbalance problem. Image quality has been enhanced by performing the sharpening process. Then, a hybrid model of Faster R-CNN and YOLO using majority voting classifier has been trained on processed data. The proposed model’s findings have been compared with its base estimators on the collected datasets. The proposed model has demonstrated detection accuracy of up to 98%, whereas YOLO and Faster R-CNN provide 95.8 and 97.5%, respectively. Additionally, compared to YOLO and Faster R-CNN, experimental results show that the proposed model performs better at estimating traffic density. Hence, the proposed approach can effectively enhance road traffic management.