Detection and Recognition of Road Sign ROI Based on Dashcam Video Data

This paper studies traffic sign detection and recognition technology based on the Yolov5 model, which is used to extract the category and quantity information of traffic signs in images. Based on this, targeted improvements were made to the model to solve the problem of missed and mistaken detection of small traffic sign targets. Improvement measures include adding a small target detection layer, using an attention module to enhance feature extraction ability of the backbone network, and using Bi-FPN to replace the original FPN + PAN structure of the model. Experimental results show the improved model exhibits higher accuracy and precision in extracting traffic sign information. This paper studies OCR technology for extracting spatial information corresponding to traffic signs in images. PP-OCR was selected to extract spatial information from traffic signs. To improve the accuracy of spatial data recognition, computer vision techniques were used to preprocess images and reduce interference of image noise.