Traffic Surveillance System: Robust Multiclass Vehicle Detection and Classification

In modern times, the identification and categorization of intelligent vehicles have attained significant recognition within the realm of highway management, as it holds paramount importance for effective traffic control and administration. The traditional vehicle detection and classification methods often cause imprecise results because they are restricted to a limited number of perspectives and viewpoints. We came up with a novel deep-learning strategy designed for the segmentation and detection of vehicles. Our innovative methodology for vehicle detection and classification involves the utilization of semantic segmentation using DeepLabv3ResNet101 for the detection of vehicles we have used YOLOv5, followed by the extraction of vehicle features employing the FAST Fourier transform algorithm, blob detection algorithm, and KAZE features and ORB feature. Subsequently, we employ the CNN (Convolutional Neural Network) model for vehicle classification, achieving an impressive classification accuracy of 96%. These experimental investigations are conducted on the widely recognized benchmark dataset known as the " Vehicle-OpenImage dataset". We have used five class datasets including (car, bus, truck, motorbike, and ambulance). The results of our study establish the current innovatory performance in the domain of vehicle detection and classification.