Vehicle Detection and Classification using YOLOv5 on Fused Infrared and Visible Images
H Shihabudeen, J Rajeesh · 2023
Visual surveillance systems in intelligent transportation systems need to detect and classify vehicles in order to create an ambient environment in smart cities. The presence of data imbalances reduces the performance of these systems. The suggested model utilizes deep learning techniques for effective identification and classification of vehicles. The YOLO v5 architecture is utilised for detection and can identify very small vehicles like bikes and scooters on the road. A fusion model is designed to generate a composite image with more complementary features for detection purposes. The fusion of infrared and visible modalities can create images of improved resolution that are suitable for night-time vehicle recognition and suited for the surveillance of remote locations. The quality of the fused image was also verified with similar models, and structural similarity is very good. In comparison with similar models, the novel algorithm produces good metric values like 0.96 for accuracy, 0.97 for sensitivity, 0.97 for specificity, and 0.93 for precision.