In recent years, as the number of visually impaired people has increased, various walking assistance systems have been developed. One of the most important functions is to create a route to the destination and guide the user. However, visually impaired people are not always able to turn right or left or correct their path as instructed because they have difficulty in understanding their surroundings. Therefore, in this study, we considered it important to give instructions to the user at the appropriate timing and collected three types of data: the set route, his/her own trajectory, and the timing of giving instructions, with the aim of ensuring that the visually impaired person can walk on the proposed route. Using these data and a recurrent neural network (RNN), we developed a system that provides automatic voice guidance at the appropriate timing. The results of the verification showed that the system provided good straight-ahead guidance, but there is still room for improvement in the right/left turn guidance.
