EMBEDDED VOICE ASSISTANCE MODULE FOR THE VISUALLY IMPAIRED BASED ON VISUAL SCENE ANALYSIS

Abstract

The article discusses a hardware module designed to assist visually impaired individuals at public transport stops. The relevance of this research is determined by the increasing number of people with visual impairments in Ukraine, caused both by global trends and by the consequences of military actions that have led to eye injuries among both military personnel and civilians. Since urban infrastructure is predominantly oriented toward visual perception, visually impaired individuals face significant difficulties in orientation, especially at public transport stops. The use of an autonomous device that does not depend on a smartphone or Internet access is of great importance for ensuring mobility and safety in urban environments.The purpose of the study is to develop and analyze the effectiveness of a hardware-software complex based on Raspberry Pi 5, which provides automatic recognition of public transport types (bus, trolleybus, tram) and their route numbers, followed by real-time voice output of the results. The hardware part includes a Raspberry Pi 5 single-board computer with active cooling, a Logitech Web Camera Carl Zeiss Tessar 2.0/3.7 2MP connected via USB 3.0, a Logitech G Pro X external sound card for audio output through headphones, and a portable power bank to ensure autonomous operation.Experimental studies have confirmed high detection accuracy of public transport in daytime conditions and acceptable performance on the embedded platform. At the same time, under low-visibility conditions (twilight, fog, image blurring), a decrease in classification accuracy of transport types and recognition of route numbers was observed. It has been determined that further system development may include expanding the dataset with images collected under real weather conditions, improving video preprocessing algorithms (low-light enhancement and denoising), as well as integrating with mobile applications and GPS navigation. The obtained results demonstrate the practical applicability of the developed solution and its potential for scaling to improve the inclusiveness and mobility of visually impaired individuals.