تصميم وتنفيذ نظام رؤية ذكي للأشخاص ضعيفي البصر باستخدام الشبكات العصبية العميقة

Abstract

Visually impaired individuals face significant challenges in their daily lives, particularly with regard to mobility and object recognition. This research aims to develop an intelligent assistive system based on artificial intelligence technologies, specifically deep neural networks, to recognize various objects and provide real-time auditory feedback to the user. This contributes to enhancing the user’s awareness and understanding of their surrounding environment.

The proposed model is trained using a deep learning algorithm on a dataset of images containing objects directly relevant to the needs of visually impaired individuals, such as medications, banknotes, and locations. In addition to object recognition, the system estimates the distance between the user and surrounding obstacles or people by applying the principle of triangle similarity, eliminating the need for traditional distance sensors. This approach reduces costs and improves the system’s efficiency.

The system relies on low-cost hardware components, namely the ESP32-CAM for video streaming and the Raspberry Pi 3 Model B for data processing, while leveraging the Google Colab platform to perform complex computational tasks efficiently.

Evaluation results demonstrate outstanding performance of the model in recognizing objects within the three target categories. The precision values for banknotes, locations, and medications are 96.2%, 97%, and 99.3% respectively, while the recall values are 97%, 97.5%, and 99.5% for the same categories. The mean average precision (mAP) reaches 98.2% for banknotes, 98% for locations, and 100% for medications.

These results reflect the model’s high reliability and classification accuracy, confirming its effectiveness and significance in practical applications that require robust and accurate object recognition.