Algorithm for Extracting Signposts from the Scene and Understanding Text on Them

The level of autonomy of modern vehicles is increasing every day. The vehicle performs more and more driving tasks independently. One of these tasks is recognizing signposts and understanding the text on them. In this paper, an algorithm for this purpose was developed based on a deep learning approach, specifically using You Only Look Once (YOLO) v7 applied to the images captured by the front-view camera. The solution is tested in the PC environment and on the embedded platform Raspberry Pi 4 to examine its performance when operating in limited resources conditions. The proposed solution achieved a high level of accuracy in detecting signposts, interpreting text, and determining directions. The experimental results also show that there is room for improvement, especially in cases of partial obstruction or low image quality. Notably, testing on the Raspberry Pi 4 highlighted the need for additional optimization due to software and hardware limitations, emphasizing the importance of resource-efficient algorithms for embedded platforms.