Concept & Overview
Sightner is an assistive hardware-software ecosystem designed to grant spatial independence to visually impaired individuals. By transforming raw visual feeds into natural language descriptive spatial cues, the device overcomes critical challenges in obstacle detection, object recognition, and distance judgment.
Technical Architecture & Neural Stack
- Real-Time Object Detection: Integrated lightweight YOLOv8n models optimized for real-time edge detection of obstacles, pedestrians, and everyday objects.
- Monocular Metric Depth Estimation: Implemented depth estimation utilizing
Intel/dpt-beit-base-384to calculate distance vectors relative to the user. - Contextual Language Processing: Fine-tuned IndoBERT to structure detected objects and proximity readings into concise, context-aware Indonesian spoken feedback.
Project Team
- Collaborators: Farrell Sudjatmiko, Nabila Mutiara Susetio, and Hasan Palito.
- Milestone: Completed as an Engineering Capstone Design Project at Institut Teknologi Sepuluh Nopember (ITS).