2020 年 25 巻 3 号 p. 185-195
We aim to develop assistive technology for visually impaired persons using computer-vision-based object detection. Many researchers report using vision-based assistive technologies in recent years, but developers could not develop deep-learning vision-based systems without a dataset. Therefore, we developed a dataset that developers can use to create deep-learning vision-based applications for visually impaired persons. The dataset contains 7,122 images and 83,319 instances covering 39 object classes. We implemented a navigation system with the dataset to prove the validity and got some positive feedback from participants.