An integration of segmentation technique on edge devices for license plate recognition

Typically, vehicle license plate recognition involving large quantities of images is carried out centrally in data centre. This results in high infrastructure and operational costs and presents difficulties for widespread deployment. To address these limitations, we propose a solution that deploys license plate recognition algorithms on edge computing devices, reducing the load on both infrastructure and centralised systems. For the purpose of training the license plate recognition model, we gathered more than 5,000 images of vehicles from various street and parking lot environments. We employed YOLOv8 for segmenting license plates and recognising the characters. Following segmentation, point sets were obtained, and based on these point sets, the license plate was reoriented to a frontal view. This allowed us to achieve a recognition accuracy of 99.21% in identifying license plate characters. Testing results on a Jetson Nano device, using 640x640 resolution data under different lighting and weather conditions, revealed an average processing speed of approximately 2.2 fps. In particular, we successfully segmented and classified license plates at distances ranging from 0.5 to 3 m, with an accuracy of up to 99.53%. This method is highly efficient, with low computational costs, and allows for smooth operation on embedded devices without compromising accuracy when compared to commercial systems.