Portable Audio-Based Navigation Aid for the Visually Impaired

Navigational aids for individuals with visual impairments have evolved with advancements in technology. Yet, many systems still fail to address critical requirements such as portability and ease of use without extensive training. In addition, current solutions often lack detailed information about detected obstacles, hindering real-time decision making. This paper presents an assistant for navigation using a smartphone which employs deep learning methods for detecting and classification. The system provides real-time information on the type, position, distance, motion status, and contextual scene data of obstacles, delivered through audio feedback to ensure accessibility. By leveraging a compact model with optimized inference times, the system can run efficiently on portable devices, offering seamless integration in dynamic environments. Furthermore, the assistant features an adjustable confidence threshold, which allows for dynamic filtering of detected obstacles based on user-specific parameters, thereby enhancing the accuracy and relevance of the feedback.