FGRCNN: A Newly Developed Hybrid Network for Multiple Object Detection Using Traffic Images

Detection of more than one object from traffic images in real time using a Social GAN-based Fast Regional Convolutional Neural Network is presented in this paper. To detect the objects, the images obtained from the Multi-Object Tracking (MOT) dataset had undergone pre-processing by employing a Gaussian filtering technique, and enhanced images were subjected to object detection by merging two networks, Social GAN and Fast Region-based CNN (FRCNN). To assess the efficiency of the approach, experimental tests were applied on diverse image datasets, and the results were validated. The Social-Gan-based FRCNN provided better accuracy of 86.5%, TPR of 94.7%, TNR of 90.9%, PPV of 88.8% and NPV of 55.8%.