Abstract: Braille is a tool for visually impaired individuals to learn knowledge and skills. But normal individuals often have little knowledge of Braille, resulting in numerous communication barriers between normal and visually impaired individuals. In this context, a natural scene Braille segment image dataset was constructed, and a natural scene Braille recognition method was designed based on the SSD framework. This method first analyzed that braille in natural scene images is usually small in size and tightly arranged, and the aspect ratio of braille characters is basically fixed; Subsequently, combining the characteristics of Braille, a CNN structure and recognition feature layer selection strategy, default box size, Braille character labels and loss functions, as well as an image input strategy were designed, and a natural scene braille recognition method was proposed. On the above Braille segment image dataset, the proposed Braille recognition method has gotten an Hmean value of 0.796 and an FPS value of 66.22; Compared with classic algorithms SSD and Faster RCNN in the field of object recognition, Braille recognition performance has been significantly improved.