AN OUTDOOR WEARABLE ASSISTIVE SYSTEM POWERED BY CNN FOR BLIND

Abstract

In this research, we suggest a mobility aid to aid the visually handicapped when they are out and
about. The ZED 2 binocular depth camera and the Jetson AGX Xavier embedded system from Nvidia make up
this auxiliary device. The picture of the environment in front of the visually impaired user is divided into seven
equal parts using the CNN neural network FAST-SCNN and the depth map acquired by the ZED 2. The system
calculates a walkability confidence value for each section and then plays a voice cue in the most effective way,
allowing a visually impaired person to safely walk down the sidewalk, avoid obstructions, and utilize the
crosswalk. Moreover, the YOLOv5s network described by Jocher, G. et al. recognizes the obstruction in the
user's path. Finally, we gave a visually impaired individual the suggested assistive device to test out in the area
of a Taiwanese MRT station. According to the visually challenged individual, he felt more secure going outside
thanks to the suggested solution. The trial also showed that the technology might help a blind individual
navigate crosswalks and sidewalks with confidence.