Obstacle Detection for the Visually Impaired Using Deep Learning and Audio Cues

Обнаружение препятствий для людей с нарушениями зрения с использованием глубокого обучения и аудиосигналов
Rayes Jordan Pradana, Ammar Yaafi Adhinegoro, Owen Bernhard, Rhio Sutoyo
2025-11-06

instance segmentationmonocular depth estimationobstacle detectionspatial audio feedbackvisually impaired navigation
Sight is one of the five senses that plays a crucial role in our everyday lives. Individuals with visual impairments frequently encounter challenges in their day-to-day activities. This research aims to assist visually impaired individuals in navigating roads, whether unfamiliar or with numerous obstacles. Rather than creating a system that requires specialized hardware, as previous works have done, this research opted to develop a mobile app that can run on devices most people already own. The system is created using a combination of an instance segmentation model, a monocular depth estimation model, and spatial audio feedback to assist users. To assess the system’s actual effectiveness, live testing was conducted with participants who have visual impairments. The research findings indicate that while the proposed system demonstrates promising potential and several positive attributes, such as accurate obstacle detection, user intuitiveness, ease of learning, and others, there remain key areas requiring improvement. These include the inability to differentiate obstacles at higher elevations than the user, feedback sounds that can overwhelm ambient noise, and several other limitations. These shortcomings provide valuable lessons and highlight key areas for improvement in future development iterations.
1
A mobile application was developed to support visually impaired navigation without requiring specialized hardware.
2
Feedback sounds could overwhelm ambient environmental noise, highlighting an important usability limitation.
3
Live testing with visually impaired participants indicated accurate obstacle detection, intuitive interaction, and ease of learning.
4
The system combines instance segmentation, monocular depth estimation, and spatial audio feedback for obstacle awareness.
5
The system could not reliably distinguish obstacles located at elevations higher than the user.
6
User testing identified promising potential but also several areas requiring improvement in future development iterations.

A mobile navigation app for visually impaired individuals operating in road environments with obstacles

The app’s obstacle-detection accuracy and usability of spatial-audio guidance for navigation, including limitations in detecting elevated obstacles and competing with ambient noise

Publication Details
Publication Date
2025-11-06
Journal
Publisher
ISSN
Cited by
0
Access Type
Author Information
Authors
Rayes Jordan Pradana
Ammar Yaafi Adhinegoro
Owen Bernhard
Rhio Sutoyo
Explore further
Open the scid.ai AI chat with a ready-made request: it will find papers on a similar topic and help build a literature review.
Find similar papers in the chat →
Make a presentation
100%