SmartVision: Transformer-Based Real-Time Object and Face Recognition for Assistive Navigation

Abstract

Visually impaired individuals encounter considerable difficulties while navigating indoor and outdoor environments due to their inability to perceive obstacles, objects, and surrounding conditions. Conventional assistive aids, such as white canes and guide dogs, offer only limited support because they primarily detect nearby obstacles and cannot provide detailed information about objects or environmental context. This work falls within the domains of assistive technology and computer vision and aims to enhance independent mobility and environmental awareness for visually impaired users. To address these challenges, an intelligent assistive system named SmartVision is proposed. The system employs the Detection Transformer (DETR) model for accurate real-time object detection and incorporates DeepFace-based face recognition to identify familiar individuals. In addition, a fallback vision-processing mechanism is integrated to ensure uninterrupted operation whenever the primary detection module encounters failure. The system delivers real-time voice feedback, enabling users to better understand their surroundings and navigate safely. By combining object detection, face recognition, and reliable feedback mechanisms, SmartVision improves navigation safety, social interaction, and overall independence for visually impaired individuals.