Traffic image text detection via YOLOv8 based on attention mechanism and rotation proposals location

Traffic scene text detection is an important technology in the field of image processing, which has significant practical significance in the applications of assisted driving, automatic driving and intelligent transportation. However, traffic scene text detection faces many challenges, such as background complexity and text Angle diversity. In this paper, the YOLOv8 model is applied to text detection tasks in traffic scenarios. In YOLOv8s model, attention mechanism is introduced to enhance feature extraction ability. At the same time, Circular Smooth Label (CSL) is used for Angle classification, so as to achieve accurate prediction of text with different oblique angles.The experimental results show that compared with DBNet algorithm, F-measure on self-made datasets CTST and public datasets ICDAR2015 is improved by 1.7% and 2.1%, respectively. In addition, the proposed algorithm is 2.03 times faster than the EAST algorithm in terms of image processing speed, and the detection time of a single image is about 0.03 seconds, which effectively meets the needs of real-time detection during driving.