With the continuous increase in the number of visually impaired individuals, it has become urgent to address the traffic challenges faced by this group. A blind guiding robot based on speed adaptation and visual recognition was designed to address this problem. The speed adaptation of the robot and the blind person is achieved through feedback control of the distance and speed. Traffic signals are identified using optimized visual recognition method based on YOLOv5 transfer learning, and man-machine interaction is realized by applying multi-module units such as real-time image, speech, and positioning. The experimental results show that the deviation between the relative distance between the man and machine and the set distance was controlled within 13.1%, the relative velocity deviation was controlled within 0.3 m/s, the accuracy of identifying traffic signals reached 91.88%, and when the man-machine distance gap is large, the robot can timely control the man-machine distance to the set distance within 0.7 s, which effectively ensured the travel safety of blind people and provide the groundwork for the practical application of guiding blind robots.