Real-Time Object Detection and Navigation Assistance for the Visually Impaired

Abstract

Visually impaired individuals were encountering serious issues in achieving safe and independent mobility both indoors and outdoors. All the assistive technologies available in the current state of the art, including white canes, guide dogs, smart canes, and advanced AI-based systems, were providing only partial assistance due to their respective limitations such as poor obstacle detection, high cost, low adaptability, and improper usage in continuous real-time applications. An assistive computer vision-based obstacle detection and navigation system for visually impaired individuals utilizing deep learning is proposed to address these problems. It employed an SSD-MobileNet model for accurate obstacle detection and identification of commonly found objects such as chairs, beds, humans, and vehicles. Audio and haptic feedback will be employed in real-time to assist in safe mobility. The proposed solution will be economical, adaptable, and efficient for real-time applications. The experimental outcomes revealed that the system improved the dependability of the navigation, user safety, and independence in terms of mobility of the visually impaired towards a better quality of life.

Introduction

This paper proposes a real-time assistive navigation system for visually impaired individuals to improve safety, independence, and environmental awareness. Existing tools like white canes and guide dogs have limitations, while many AI-based solutions are costly, bulky, or not suitable for real-time use in dynamic environments.

The proposed system uses computer vision and deep learning (SSD MobileNet trained on the COCO dataset) to detect common obstacles such as people, vehicles, furniture, doors, and stairs from live smartphone camera input. It also estimates the distance of objects (near, medium, far) using bounding box information to prioritize threats.

To reduce user cognitive load, the system provides selective audio feedback using text-to-speech (pyttsx3), informing users about nearby obstacles and their distance. The system is designed to be lightweight, low-cost, and real-time, using Python, OpenCV, and TensorFlow.

Experimental results show strong performance with 92% detection accuracy, 89% F1-score, 87% mAP, and real-time processing at 20–22 FPS with <100 ms latency. Distance estimation error remains below 5%. Overall, the system effectively combines object detection, distance estimation, and voice alerts to enhance mobility and safety for visually impaired users.

Conclusion

The proposed system is able to perform object detection, real-time distance estimation, and audio feedback. This would be made possible by the combination of computer vision and deep learning using an SSD MobileNet model that has been trained on visualizations of the COCO dataset to ensure the correct identification of the objects surrounding the system and real-time proximity information. The experimental results demonstrate the effectiveness of the system with high accuracy in the detection, low latency, and correct distance estimation, thus suitable for practical applications. It is a lightweight, cost-effective, and user-friendly framework that can be applied both indoors and outdoors. The real-time audio feedback enables the user to get clear and actionable information without being cognitively overloaded. In conclusion, it can be said that the system is a correct example of the combination of technology innovation and usability, thus providing an efficient system for environmental monitoring.

References

[1] B. Li et al., \"Vision-Based Mobile Indoor Assistive Navigation Aid for Blind People,\" in IEEE Transactions on Mobile Computing, vol. 18, no. 3, pp. 702-714, 1 March 2019. [2] F. Shaikh, M. A. Meghani, V. Kuvar and S. Pappu, \"Wearable Navigation and Assistive System for Visually Impaired,\" 2018 2nd International Conference on Trends in Electronics and Informatics (ICOEI), Tirunelveli, India, 2018. [3] H. Son, A. Vora, G. Pandey and J. Weiland, \"Infrastructure Enabled Guided Navigation for Visually Impaired,\" in IEEE Transactions on Intelligent Transportation Systems, vol. 26, no. 5, pp. 6764-6777, May 2025. [4] I. N and V. Loganathan, \"Comprehensive Assistive Mobility System for the Visually Impaired: The Smart Blind Stick with Ultrasonic Sensor Solution,\" 2024 International Conference on Inventive Computation Technologies (ICICT), Lalitpur, Nepal, 2024. [5] J. R. B. Guevara et al., \"Mobility Assistive Technology with Artificial Intelligence: Indoor Navigation Assistance for the Visually Impaired Using Arduino-Based Assistive Goggles,\" 2023 IEEE 15th International Conference. [6] M. Kulkarni, M. Chitale, S. Chitpur, A. Chivate, P. Chopade and S. Deshmukh, \"Smart Assistive Navigator for the Blind using Image Processing,\" 2023 International Conference on Sustainable Computing and Smart Systems (ICSCSS), Coimbatore, India, 2023. [7] N. Al Mahmud, T. Hossain, R. V. Biswas, M. Foysal Ahammad Joy, M. Abu Raihan and D. Banik, \"Assistive System and GPS Tracker for the Visually Impaired,\" 2023 26th International Conference on Computer and Information Technology (ICCIT), Cox\'s Bazar, Bangladesh, 2023. [8] O. Gamal, S. Thakkar and H. Roth, \"Towards Intelligent Assistive System for Visually Impaired People: Outdoor Navigation System,\" 2020 24th International Conference on System Theory, Control and Computing (ICSTCC), Sinaia, Romania, 2020. [9] Salvador Mar-nez-Cruz 1 , Luis A. Morales-Hernández 1 , (Member, IEEE), Gerardo I. Pérez- Soto 2 , Juan P. Benitez-Rangel 1 , And Karla A Camarillo-Gómez 3 “An Outdoor NavigaRon Assistance System for Visually Impaired People in Public TransportaRon”. 26 [10] Z. Yang et al., \"SeeWay: Vision-Language AssisRve NavigaRon for the Visually Impaired,\"

Copyright

Copyright © 2026 Mr. B. Vamshi, B. Ashritha, K. Hithasree, S. Sanjana, E. Sumanjali. 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.