AN IMPROVED OBJECT RECOGNITION SYSTEM FOR VISUALLY IMPAIRED PEOPLE WITH AUDIO FEEDBACK

This research presents the development and evaluation of an improved object recognition system designed to assist visually impaired individuals in identifying objects in their surroundings using deep learning. The background to this study lies in the growing need for cost-effective assistive technologies that leverage artificial intelligence to improve independence and navigation for the visually impaired. Existing systems often suffer from limitations in speed, accuracy, and accessibility, especially in real-time scenarios or on low-powered devices. This project aimed to address these challenges by implementing a lightweight yet effective object detection system using the YOLOv11-nano model pre-trained on the COCO dataset. The primary objectives included deploying a real-time detection system capable of recognizing multiple object classes, generating accurate audio feedback, and optimizing performance for offline and mobile use. The system was developed using Python and OpenCV, with live camera input processed through the YOLOv11-nano model. Audio feedback was provided using text-to-speech integration to inform users of detected objects. To evaluate the system, a small dataset of real-world images was used for testing. Key performance metrics such as precision, recall, F1-score, and Intersection over Union (IoU) were calculated. The results showed a precision of 95%, accuracy of 92%, recall of 90%, F1-score of 92%, and an average IoU of 0.79, demonstrating reliable and accurate performance across varying environments. This system offers a meaningful step toward affordable and accessible visual assistance tools, particularly for offline or mobile deployment. The study concludes that YOLOv11-nano provides a strong balance of speed and accuracy, making it suitable for real-time assistive applications. Future improvements include training on local datasets to reduce misclassification and integrating more advanced audio and navigational feedback features. The research successfully fills a gap in deploying efficient, real-time object recognition systems tailored to the needs of the visually impaired community. Keywords: Object recognition, visually impaired, YOLOv11-nano, deep learning, assistive technology, real-time detection, computer vision, audio feedback, object detection, accessibility. Word Count: 300