AIVA – Artificial Intelligence Vision Assistant for Visually Impaired using YOLO and MiDaS Depth Estimation

Abstract

Individuals with visual impairments often face challenges from limited awareness of their surroundings. This affects their independence and safety in daily life. Our study presents a mobile based Artificial Intelligence Vision Assistant for people with visual impairments. It gives real time scene understanding through audio feedback. The system uses a smartphone camera to grab live video feeds. It relies on lightweight deep learning models like YOLO for quick object detection. EffiecientDet-Lite2 handles feature extraction well. TensorFlow Lite runs inference right on the device. The setup spots objects and picks out obstacles. It estimates distances too. It recognizes known people with a built in face recognition part. Visual details turn into natural speech via Android Text to Speech. This lets users get steady guidance without needing internet. The approach offers low delay and strong privacy. It delivers reliable offline work. All this boosts navigation and awareness for visually impaired folks. It helps their autonomy as well.