Flipkart Grid Robo 6.0: Smart Quality Test System Table of Contents Project Overview Key Features Use Cases Technologies Used Installation Guide Usage Challenges Recommendations Contributing License Contact Project Overview
Project Overview This project was developed as part of the Flipkart Grid Robo 6.0 challenge to create a Smart Quality Test System using camera vision technology. The system assesses shipment quality and quantity for India's leading e-commerce company, ensuring that products meet high standards before reaching the customer. The project uses OCR (Optical Character Recognition) and Machine Learning models to analyze product details and detect freshness, quantity, and expiry dates.
Key Features OCR Applications for extracting key details such as brand name, product name, quantity, expiry date, and MRP from images. Machine Learning Classification Models for object recognition, counting, freshness detection, and expiry date validation. Seamless integration of models into a graphical user interface (GUI) with real-time or photo upload capabilities. Ability to connect models through APIs for easy implementation in real-world scenarios.
Use Cases The project addresses several specific use cases: Use Case 1: Extracting Brand-Name, Quantity, and Product Name Accuracy: 81.75% (using Easy OCR and Pytesseract) Use Case 2: Extracting Expiry Date and MRP Accuracy: 80.56% (using Pytesseract and OCR Regex) Use Case 3: Object Recognition and Counting Accuracy: 91.74% (using XGBoost) Use Case 4: Freshness Detection Accuracy: 92.79% (using Autoencoders and fine-tuning) Use Case 5: Expiry Date Detection Accuracy: 90.68% (using neural networks and regex patterns)
Technologies Used Programming Languages: Python Libraries: OCR: EasyOCR, Pytesseract Machine Learning: XGBoost, TensorFlow, Autoencoders Regex for pattern matching GUI: Streamlit (or Flask/Django for API integration) Tools: OpenCV, NumPy, Pandas
Usage Real-time Processing: Use a connected camera to feed images in real time. Photo Upload: Upload photos of shipments to analyze them through the system. GUI: Interact with the models through the provided GUI for ease of use. The system provides real-time feedback on shipment quality, including:
Recognized brand and product names Quantity verification Freshness assessment Expiry date detection
Challenges Throughout development, the following challenges were faced:
OCR Accuracy: Achieving a balance between processing speed and accuracy in extracting text details, especially in cases with low-quality images. Real-Time Processing: Handling large amounts of data in real-time for object recognition and freshness detection. Model Fine-Tuning: Optimizing machine learning models to increase accuracy across different product categories and shipment types. Recommendations To further enhance the system:
Use HD cameras: Higher quality images will improve the accuracy of OCR and object detection models. Model Expansion: The system can be extended to incorporate more categories and models for other e-commerce use cases.
Contact For any questions or issues:
Author: Mayank Choudhary Email: [email protected]
API Integration: Each model can be integrated into a broader system using APIs, enabling scalability and adaptability.