Pedestrian Crossing Detection by VGG16 for Visually-impaired Walking Assistance System

The visually impaired, unlike other physically handicapped people, are constantly exposed to dangers such as traffic accidents and falls from high positions due to their inability to obtain visual information. Therefore, guide dogs are currently used to assist the visually impaired in walking. In the future, however, as the number of visually impaired people increases, it will become difficult to provide walking assistance only with guide dogs. Therefore, to realize safe walking for the visually impaired on pedestrian crossings, we present an automatic detection method for pedestrian crossings using images from cameras attached to the user’s body. Classical image processing-based pedestrian crossing detection is computationally inexpensive and has excellent real-time performance. However, it lacks versatility. Therefore, in this study, we use a transfer learning model called VGG16 to detect pedestrian crossings. To further confirm the usefulness of VGG16, we compared it with other transfer learning models. The results showed that the validation accuracy of VGG16 was 97.50%, indicating that VGG16 was the most useful.