Vision Assist: Real-Time Guidance System

In order to help visually impaired people, this study proposes a machine learning-based system that provides audio feedback, distance calculation, and real-time object detection. The system uses Faster Region-based Convolutional Neural Network (Faster R-CNN), optimized with Compute Unified Device Architecture (CUDA) and CUDA Deep Neural Network library (cuDNN) for GPU-accelerated performance with low latency. Images captured by the rear camera of a smartphone are processed to identify objects and compute their proximity. The system offers audio feedback using the Pyttsx3 text-to-speech engine, alerting users to nearby objects and potential hazards. The system achieved over 99% accuracy in recognizing everyday objects. Parallelized GPU operations ensure efficient real-time object recognition and distance estimation, improving users' safety and mobility. Experimental results demonstrate the system's stability and reliability in assisting visually impaired individuals during navigation.