Vision-Based Indoor Navigation System for Assistive Applications Using Deep Learning

Abstract

Significant amounts of work have been devoted to the development of indoor navigation systems due to GPS deficiencies in indoor navigation. In this paper, the analysis and application of the smart indoor navigation system have been described. The suggested system is capable of working with video feeds and uses machine learning algorithms. The proposed methodology is based on computer vision algorithms, sensor fusion, and machine learning algorithms, such as convolutional neural networks (CNN). It solves issues concerning weak signals, obstacles, and the dynamic nature of the environment. It is a reliable methodology that applies CNNs. The system has been developed by using the Python programming language, OpenCV, and TensorFlow/Keras libraries. The application of the proposed methodology was carried out in various indoor environments, such as corridors, laboratories, and hospitals at universities. The test results showed improved accuracy of 0.60 against 0.24 during ten training sessions.

Introduction

The study focuses on developing an intelligent, video-based indoor navigation system using machine learning and computer vision. Indoor navigation is increasingly important in smart buildings, hospitals, and educational institutions. However, GPS performs poorly indoors because of signal attenuation and multipath interference. Existing alternatives, such as Wi-Fi, Bluetooth beacons, and inertial measurement units (IMUs), also have limitations related to signal instability, positioning errors, and accumulated drift.

The primary objective of the research is to design and evaluate a smart indoor navigation system that can recognize its surroundings, detect obstacles, identify navigable paths, and provide real-time directions. The system is also intended to support assistive navigation, particularly for visually impaired users.

The literature review examines several existing technologies, including Wi-Fi fingerprinting, Bluetooth Low Energy (BLE) beacons, IMU-based dead reckoning, and vision-based systems such as simultaneous localization and mapping (SLAM). It also discusses machine learning techniques such as Convolutional Neural Networks (CNNs), Support Vector Machines (SVMs), and reinforcement learning. Although these methods can improve navigation accuracy and adaptability, challenges remain, including computational costs, limited training data, environmental changes, and signal interference. Combining multiple sensors can help overcome the limitations of individual technologies.

The proposed system uses a CNN-based architecture integrated with computer vision and sensor fusion. Video recordings are collected using smartphone cameras in indoor environments such as university corridors, laboratories, and hospitals under different lighting conditions and with dynamic obstacles. The videos are converted into image frames, which are labelled and organized into datasets for supervised learning. The frames undergo preprocessing, including resizing, normalization, and noise reduction. The CNN then extracts visual features to identify paths and obstacles, while the broader architecture incorporates IMU readings and Wi-Fi/Bluetooth positioning information. Based on the processed information, the system generates instructions such as moving forward, turning left or right, and warning about obstacles.

The prototype is developed using Python, OpenCV, and TensorFlow/Keras, with Google Colab used for cloud-based training. The dataset is divided into 70% training, 15% validation, and 15% testing data. Model performance is evaluated using accuracy, precision, recall, F1-score, and categorical cross-entropy loss. Training over 10 epochs reportedly improves accuracy from approximately 24% to 60%, indicating that the model learns to recognize navigation-related patterns. User feedback and tests under varying lighting and obstacle conditions are also used to assess satisfaction and robustness.

The system is designed to provide real-time navigation through smartphone or webcam input, with visual overlays or audio instructions. Its modular architecture allows individual components to be improved independently and supports potential deployment on smartphones and embedded devices.

Conclusion

In our research, we present a smart indoor navigation solution that takes advantage of the combination of machine learning and videos to tackle the weaknesses in the GPS-based navigation system within the indoor environment. In the proposed technique, CNN, sensor fusion, and frames from videos are applied to provide efficient guidance in the indoor environment. Our experiments have shown the efficiency of our proposed solution, and its application to people, even blind individuals. Other future research opportunities could include other perspectives on the topic, such as real-time processing through compression and quantization, increasing the diversity of the training dataset, applying an embedded IMU sensor fusion model, and conducting scene analysis by means of trans-formers. Also, future research can consider implementing our solution on edge devices, like Raspberry Pi, and mobile devices

References

[1] T. Samavedula, S. Mohapatra and S. K. Nayak, “Mini Mapper: Cost-Effective Indoor Mapping and Navigation using Nano Drone,” 2025 17th International Conference on COMmunication Systems and NETworks (COMSNETS), Bengaluru, India, 2025, pp. 1377–1379. [2] A. S. Alluhaidan et al., “From GPS to AR: Leveraging Augmented Reality and Grid-Based Systems for Improved Indoor Navigation,” IEEE Access, vol. 13, pp. 55211–55230, 2025. [3] T. Li et al., “Digital Twin-Enhanced Wireless Indoor Navigation: Achieving Efficient Environment Sensing With Zero-Shot Reinforce-ment Learning,” IEEE Open Journal of the Communications Society, vol. 6, pp. 2356–2372, 2025. [4] L. Huang et al., “High-Precision Indoor Positioning Technology Based on Multichannel Array Pseudolite,” IEEE Trans. Instrum. Meas., vol. 74, pp. 1–10, 2025. [5] R. S, A. Yadav and S. B. Pal, “Enhanced Indoor Pathfinding Navigation in Multistoried Buildings Using a Hybrid of RRT-Connect and Dijkstra’s Algorithm,” 2025 Int. Conf. on Multi-Agent Systems for Collaborative Intelligence (ICMSCI), Erode, India, 2025. [6] Y. Feng et al., “A Smart Cane-Based Indoor Positioning System for Visually Impaired People,” IEEE Sensors Journal, vol. 25, no. 7, pp. 12085–12096, 2025. [7] S. K et al., “Autonomous Indoor Navigation Using Nav2,” 2025 3rd Int. Conf. on Intelligent Data Communication Technologies and IoT (IDCIoT), Bengaluru, India, 2025. [8] J. Huang et al., “Pedestrian Navigation Using Waist-Mounted IMU and RTK,” IEEE Sensors Journal, vol. 25, no. 3, pp. 5510–5519, 2025. [9] R. Gayathri et al., “Spatio-Temporal and Semantic Enhanced Context-Aware Navigation for Indoor Robots,” IEEE Access, vol. 13, pp. 36909–36929, 2025. [10] C. Xu et al., “DUNE: Sim2Real Transfer for Depth-based Navigation in Unstructured Dynamic Indoor Environments,” ICASSP 2025, Hyder-abad, India, 2025. [11] S. M. Sheikh et al., “Visible Light Positioning as a Next-Generation Indoor Positioning Technology: A Tutorial,” IEEE Commun. Surveys Tutorials, 2024. [12] A. Ali et al., “A Visible Light-based Indoor Navigation and Localization Solution for Visually Criticized Users,” IEEE Trans. Emerging Topics in Computing, 2023. [13] U. B. Mahadevaswamy and N. L. Chiranth, “Development of a Smart Application for Indoor Navigation (INMaps),” Wireless Personal Com-munications, 2024. [14] S. Manju, S. Singh and S. Kumar, “IoT-Based Smart Indoor Navigation System with Voice Assistance for Museums,” Advances in IoT and Smart Environments, Springer, 2024. [15] X. Lin et al., “Wi-Fi-Based Indoor Localization and Navigation: A Robot-Aided Hybrid Deep Learning Approach,” Sensors, vol. 23, no. 14, 2023. [16] M. Salimibeni and A. Mohammadi, “Hybrid Indoor Localization via Re-inforcement Learning-based Information Fusion,” IEEE ICASSP 2023. [17] S. B. Derkar, D. Biranje, L. P. Thakare, S. Paraskar and R. Agrawal, ”CaptionGenX: Advancements in Deep Learning for Automated Image Captioning,” 2023 3rd Asian Conference on Innovation in Technology (ASIANCON), Ravet IN, India, 2023, pp. 1-8, doi: 10.1109/ASIAN-CON58793.2023.10270020. [18] M. Ghosh et al., ”An Analysis of Deep-Neural-Network Model for the Determination of the Bit-Rate of Optical Fiber Signals,” 2023 11th Inter-national Conference on Emerging Trends in Engineering Technology Signal and Information Processing (ICETET - SIP), Nagpur, India, 2023, pp. 1-4, doi: 10.1109/ICETET-SIP58143.2023.10151480. [19] Shende, D., Thakare, L., Agrawal, R., Wyawahare, N. (2024). Automatic Disease Detection for Various Plants Leaf Using Image Processing Techniques and TensorFlow Algorithm. In: Shaw, R.N., Siano, P., Makhilef, S., Ghosh, A., Shimi, S.L. (eds) Innovations in Electrical and Electronic Engineering. ICEEE 2023. Lecture Notes in Electrical Engineering, vol 1115. Springer, Singapore. [20] S. Paraskar, A. S. Pathan, R. Parteki, R. A. Jichkar, L. Thakare and [21] T. Deotale, ”A Review on Deep Learning Methods and CT Scan Approaches for Covid-19 Detection,” 2022 10th International Confer-ence on Emerging Trends in Engineering and Technology - Signal and Information Processing (ICETET-SIP-22), Nagpur, India, 2022, pp. 1-6, doi: 10.1109/ICETET-SIP-2254415.2022.9791687. [22] S. R. Singh Thakur and L. Thakre, ”Harnessing Artificial Intelli-gence for Real-Time Driver Fatigue Detection,” 2024 International Conference on Artificial Intelligence and Quantum Computation-Based Sensor Application (ICAIQSA), Nagpur, India, 2024, pp. 1-7, doi: 10.1109/ICAIQSA64000.2024.10882453. [23] Das, D., Ghosh, M., Raut, M., Thakre, L., Jichkar, R. (2025). Enhancing Ophthalmologic Infrastructure: Seamless Integration of Deep Convolu-tional Neural Networks for Real-Time Ocular Disease Detection. In: Bajaj, A., Abraham, A., Castillo, O. (eds) Proceedings of the 15th International Conference on Soft Computing and Pattern Recognition (SoCPaR 2023). SoCPaR 2023. Lecture Notes in Networks and Systems, vol 1243. Springer, Cham. [24] Kalbande, K., Patil, W., Deshmukh, A., Joshi, S., Titarmare, A. S., Patil, [25] S. C. (2024). Novel Edge Device System for Plant Disease Detection with Deep Learning Approach. International Journal of Intelligent Sys-tems and Applications in Engineering, 12(3), 610-618. [26] Swati Sorte et al 2021 J. Phys.: Conf. Ser. 1913 012134. [27] S. Sarkar, S. Ahire, S. Rahate, R. Barde, R. Agrawal and S. Sorte, ”Design of Weapon Detection System,” 2022 3rd Interna-tional Conference on Electronics and Sustainable Communication Systems (ICESC), Coimbatore, India, 2022, pp. 1016-1022, doi: 10.1109/ICESC54411.2022.9885601.

Copyright

Copyright © 2026 Shrikant Kharode , Paras Ghime, Rucha Jichkar, Swati Paraskar, Dr. Laxman Thakre, Dr. Swati Patil, Leena Sharma. 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.