Understanding Through AI
A graduation project that uses AI to help visually impaired individuals understand their surroundings through voice-activated object detection.
The Challenge
Over 2.2 billion people worldwide have visual impairment. Daily tasks like identifying objects, navigating spaces, and understanding surroundings remain challenging despite existing assistive technologies.
Our Solution
Abserny combines computer vision, machine learning, and speech synthesis to provide real-time environmental descriptions. The system detects objects, understands context, and communicates information through clear, natural audio feedback.
Abserny Process Flow
Core Features
Voice Activation
Offline speech recognition using Vosk allows hands-free operation. Just say trigger words like "start" to activate detection without touching any buttons.
AI Object Detection
YOLOv8 nano model provides fast, accurate detection of 80+ object classes including people, furniture, electronics, and everyday items.
Complete Privacy
All processing happens locally on your device. No internet required after setup, no data collected, no cloud services—your privacy is guaranteed.
Natural Language
Converts raw detection data into clear, conversational descriptions. Hear "I see a person and two chairs" instead of technical output.
Lightweight & Fast
Optimized nano model runs on CPU without GPU requirements. Fast inference ensures minimal delay between voice command and response.
Accessible Design
Built with accessibility in mind from the ground up. Screen-reader compatible, voice-first interface, simple controls.
Technology Stack
YOLOv8 (Ultralytics)
State-of-the-art object detection with yolov8n.pt nano model for fast inference on CPU
Vosk Speech Recognition
Offline speech recognition supporting multiple languages with low resource requirements
OpenCV
Camera capture and frame processing with efficient video handling
pyttsx3
Cross-platform text-to-speech engine that works offline without cloud services
KivyMD
Modern Material Design UI framework for Python with accessibility features
Python 3.11
Core language with threading for concurrent audio processing and detection
Get Abserny on Mobile
Real-time, offline object detection with natural voice feedback — coming soon to your smartphone.
Learn More