Real-Time Object Identification and Assistance using Deep Learning for the Blind
Sandhya S V, S Sakthi Vinayagam, B Sujitha, D Thirisha, Subash Palaniappan Thirumoorthi, B Praveen Kumar · 2025
Users with visual impairments face challenges in new spaces due to limited perception of surrounding objects. Conventional aids like canes or guide dogs lack full awareness at the object level. The challenge is resolved by this project through real-time object detection via deep learning. The solution uses YOLOv8 and EfficientNet-B3 is a backbone for feature improvement are combined to give speed and high accuracy. The model is executed on live webcam or camera feeds to identify and locate different real-world objects, including humans, vehicles, animals, and furniture, in real time. A Flask-based Python backend supports the detection, while HTML, CSS, and JavaScript support an easy-to-use front end. The innovation lies in the combined form of YOLOv8 and EfficientNet-B3 to achieve improved accuracy, as well as multilingual voice output of detected objects in three languages. The system highly improves real-time situational awareness and autonomy for users with visual impairments, providing a cost-effective, accurate, and easy-to-use assistive technology that outperforms current models in detection.