Development of an Object Detection System for the Visually Impaired

This study presents the development of an Android-based object detection application designed to assist visually impaired persons in recognizing and interacting with objects in their immediate environment. The system operates on an image processing and deep learning framework implemented using Google’s TensorFlow API and Convolutional Neural Networks (CNN). The application allows users to capture real-time images through a mobile device camera, where detected objects are processed and conveyed to the user through synthesized voice output. Android was selected as the development platform due to its open architecture and built-in accessibility features, including speech recognition and text-to-speech services, which support intuitive and hands-free user interaction. The methodology adopted in this work included the review and definition of relevant concepts in pattern recognition and computer vision, evaluation of existing object detection applications to identify gaps, design of an improved assistive detection architecture, and implementation of the proposed system. Additional assistive modules such as Optical Character Recognition (OCR), voice assistance dialogue support, and image labeling using Firebase ML Kit were integrated to expand user awareness and usability. The model was evaluated using the TensorFlow Object Detection API, where a confidence threshold was applied to interpret detection reliability. Detection outputs with a confidence level of 60% and above were classified as accurate, while results below 59% were regarded as unreliable, ensuring dependable real-time feedback. Overall, the resulting system, referred to as Vocal Vision, enhances environmental awareness and independence for visually impaired users by providing real-time object detection, voice-guided navigation, and contextual understanding of their surroundings through a portable and user-friendly mobile interface.