Abstract
AbstractFor vehicle detection tasks, the accuracy of detection model trained by high-resolution images will be decreased in low-resolution images. To solve this problem, a new object detection network is proposed which cascades super-resolution and target detection networks. SwinIR restores the image from low-resolution to high-resolution which integrates the advantages of both CNN and Transformer. YOLOv5 is improved by replacing its loss function with EIOU to enhance convergence speed and localization accuracy. The experiments are conducted on UA-DETRAC dataset, and the effectiveness of the detection network is verified.KeywordsVehicle detectionSuper-resolutionDeep learning
Sign-in/Register to access full text options 
Free) 
Talk to us
Join us for a 30 min session where you can share your feedback and ask us any queries you have
Schedule a call
