Blind Guidance System Using Floor Segmentation and Distance Estimation

Visually impaired individuals face several challenges in mobility-related activities, including navigating directions, identifying objects and avoiding obstacles. Several systems have been developed to address these challenges. However, these systems are expensive, less wearable, and provide less effective responses. In this study, we propose a system that can assist visually impaired individuals in avoiding obstacles and walking in safe areas, especially on floors with a grid-like pattern. This paper uses deep learning-based image segmentation to differentiate between walkable floor areas and areas with obstacles. Additionally, a perspective transformation or homography method is also applied to estimate the obstacle's distance. The system is implemented on Android smartphones and can provide a safe area response in the form of a clock position using text-to-speech audio. The segmentation model can achieve average Intersection over Union (IoU) accuracy of 91% with an average total processing time of 173.96 ms. The average distance estimation error is 5.1 centimeter at a 45 -degree camera angle. With these results, the system can offer an accurate, wearable, and cost-effective guidance system for visually impaired individuals.