Enhancing Object Detection in Assistive Technology for the Visually Impaired: A DETR-Based Approach
Sunnia Ikram, Imran Sarwar Bajwa, Sujan Gyawali, Amna Ikram, Najah Alsubaie
Abstract
The paper proposes a real-time obstacle detection and recognition system for visually impaired individuals, designed to enhance their navigation using assistive technology. This system integrates a mobile application equipped with a mini camera for real-time image capture, employing advanced deep learning techniques to process and classify objects. By combining YOLOv8, Faster R-CNN and DETR for object detection, the system provides a comparative evaluation based on precision, recall, F1 score, confidence score and processing efficiency. DETR (Detection Transformer) demonstrates superior performance, achieving a confidence score of 99%, a precision of 98% and a processing speed of 40ms/frame. While Faster R-CNN and YOLOv8 provide valuable insights, they offer slightly lower scores, highlighting the balance between accuracy and computational efficiency. The system’s workflow involves real-time image acquisition, preprocessing, innovative data augmentation and optimization for edge devices using TensorFlow Lite for efficient deployment. The application classifies eighty obstacles, such as pedestrians, vehicles and traffic signal and generates immediate audio feedback, facilitating safe navigation. The model trained for 20 epochs and achieved 98% accuracy in 20<sup xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">th</sup> epoch. This study establishes a scalable, efficient, and practical solution that integrates IoT and real-time image processing to empower visually impaired users with enhanced mobility and safety.