Real-Time Vehicle Classification and License Plate Recognition via Deformable Convolution-Based Yolo v8 Network

License plate detection (LPD) system identifies and tracks vehicles using their license plate (LP) numbers, making it useful for law enforcement, toll collection, and parking management. In addition to the differences in color, font, and language used in these characters, it becomes increasingly difficult to find similar plates. To overcome these challenges, a novel deformable convolution-based Yolov8 network (DEN-YOLO net) has been proposed to detect cars and their LP in real time. Initially, the input vehicle images undergo several preprocessing steps, such as low-light enhancement, superresolution, and defogging for enhancing the image quality. The You Only Look Once-v8 (YOLOv8) model is employed for detecting vehicles by distinguishing between images containing cars or other vehicles. Then, the detected car images are fed to the Yolo net for recognition of the LP. Subsequently, LP detection and segmentation are performed to extract alphanumeric characters, which are recognized using character recognition algorithms. The proposed DEN-YOLO net achieves an overall accuracy level of 98.9% based on its testing and training accuracy curves. The proposed DEN-YOLO net improves the overall accuracy of 6.05%, 8.02%, 4.82%, 2.26%, and 1.85% better than YOLO v5, YOLO, Otsu’s method, Yolo v7, and YOLOv3, respectively.