Research on vision based outdoor blind guiding robot

This paper designs a blind guiding robot working in the outdoor environment. The robot improves the path planning of the artificial potential field method by acquiring the data from ultrasonic radar and inertial sensors. The vision system of the guide robot can obtain the category of obstacles ahead and their distance through binocular ranging and deep learning algorithm and inform the blind in the form of voice. The guide robot uses Mobilenet_SSD detection algorithm to detect six common obstacles, including person, bicycle, motorcycle, car, manhole cover, and stone ball roadblock. The vision system also recognizes the straight-line detection of string obstacles that pose a threat to the blind. The simulation and experimental results show that the blind guiding robot can successfully avoid obstacles and reach the target point by using the improved artificial potential field method. The vision system has realized the function of obtaining the types and distances of obstacles ahead, and the accuracy of binocular ranging in an outdoor environment can reach 93.3%.