Overview
EzyCart is an AI-powered smart shopping cart system that integrates computer vision and sensor-based technology to autonomously identify items placed into the cart. It eliminates manual barcode scanning and checkout by using object detection, weight verification, and fallback scanning mechanisms to ensure robust and reliable billing. Designed to streamline retail shopping, it delivers an intuitive, contactless experience.
Pipeline
1. Real-time object detection via YOLOv5 from a live camera feed
2. Weight measurement using load cells to validate item placement and quantity
3. Barcode scanner fallback to improve robustness for hard-to-detect items
4. Product database lookup and dynamic billing updates
5. Touch-based UI for user interaction and final billing
6. Exportable billing receipts and inventory logs
Hardware & Integration
The prototype is built using an NVIDIA Jetson Nano module with a Pi Camera for vision, load cells for real-time weight sensing, and a barcode scanner module for redundancy. The cart uses a compact embedded frame with a display unit and interactive interface. Weight validation is cross-referenced with vision predictions to prevent mismatches and frauds.
Highlights
• 97%+ object recognition accuracy and real-time billing feedback
• Designed to support 30+ grocery items in prototype
• Integrated weight sensors for fraud prevention and redundancy
• Barcode scanner for fallback when CV confidence is low
• Patent published in India for the system architecture
Tech. Stack
Python, PyTorch, YOLOv5, OpenCV, Jetson Nano, Load Cells, HX711, Barcode Scanner, SQLite, Raspberry Pi Camera, HTML/CSS/JS