AI-Powered Navigation Assistant for Visually Impaired Users using Real-Time Obstacle Detection

Abstract

Modern assistive technologies, particularly those designed for visually impaired individuals, require efficient real-time perception and response mechanisms that traditional methods fail to provide effectively. Conventional tools such as walking sticks or basic sensor-based systems offer limited environmental awareness and are unable to accurately identify and classify surrounding obstacles. A practical assistive solution must support real-time object detection, reliable processing, and intuitive user feedback to enhance navigation safety and independence. However, many existing systems depend on external hardware or lack the ability to perform efficiently under dynamic real-world conditions. In this paper, we propose UniGen AI, an intelligent mobile-based obstacle detection system that utilizes computer vision and machine learning techniques to identify surrounding objects in real time. The system integrates a smartphone camera with the TensorFlow Lite framework and employs the SSD MobileNet model to perform efficient on-device object detection with low latency.To ensure continuous and reliable operation, the application is designed with a modular architecture that processes input frames, analyzes object information, and delivers immediate feedback through a Text-to-Speech (TTS) module. This enables users to receive clear audio alerts about nearby obstacles without requiring visual interaction. Furthermore, the system operates entirely on the mobile device, eliminating dependency on internet connectivity and additional hardware. Experimental evaluation demonstrates that the proposed approach provides accurate detection, fast response time, and improved usability. The system significantly enhances navigation safety while maintaining portability, efficiency, and real-time performance in dynamic environments.

Introduction

The text presents an AI-based assistive navigation system designed to help visually impaired users overcome the limitations of traditional aids like walking sticks and basic sensors. These conventional tools often fail to provide real-time environmental understanding or accurate obstacle detection, making navigation unsafe in dynamic settings.

To address this, the proposed system (UniGen AI) uses smartphone cameras combined with computer vision and machine learning for real-time object detection. It is built using lightweight models like SSD MobileNet and TensorFlow Lite, enabling efficient on-device processing without external hardware. Detected obstacles are converted into audio alerts using Text-to-Speech, allowing hands-free guidance.

The system is designed to handle challenges such as changing lighting conditions, real-time processing constraints, and limited mobile device resources. Its architecture includes an AI processing layer, safety monitoring module, and decision control system that balances AI predictions with safety rules to ensure reliable navigation.

Additionally, the mobile client supports environment monitoring (camera, GPS, battery) and adaptive performance management to optimize speed and efficiency. Key contributions include real-time obstacle detection, lightweight model integration, voice-based feedback, modular architecture, and real-world evaluation.

The literature review highlights related work such as IoT smart sticks, YOLO-based detection systems, wearable assistive devices, infrared and ultrasonic navigation tools, all aiming to improve safety and accuracy for visually impaired users.

Conclusion

This paper presented an AI-Powered Navigation Assistant for visually impaired users that combines Artificial Intelligence, Computer Vision, GPS navigation, and voice interaction technologies. By integrating intelligent obstacle detection, adaptive navigation support, and security monitoring mechanisms, the proposed system demonstrates a reliable, safe, and efficient solution for improving independent mobility and real-time assistance for visually impaired individuals.

References

[1] A. Langley et al., ”The QUIC Transport Protocol: Design and InternetScale Deployment,” in Proc. SIGCOMM, 2017. [2] L. Li, W. Chu, J. Langford, and R. E. Schapire, ”A contextual-bandit approach to personalized news article recommendation,” in Proc. WWW, 2010. [3] M. Fowler, ”CircuitBreaker,” martinfowler.com, 2014. [4] NIST, ”Guide to Intrusion Detection and Prevention Systems (IDPS),” NIST Special Publication 800-94, 2007. [5] E. Rescorla, ”The Transport Layer Security (TLS) Protocol Version 1.3,” RFC 8446, 2018. [6] S. Bubeck and N. Cesa-Bianchi, ”Regret Analysis of Stochastic and Nonstochastic Multi-armed Bandit Problems,” Foundations and Trends in Machine Learning, 2012.

Copyright

Copyright © 2026 Shivranjani Dedhe, Dr. M. P. Borawake, Bipasha Awghade, Vishal Sudrik , Kartik Padole. This is an open access article distributed under the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.