Shopping Cart and Basket Detection Using YOLOvX
Team Members
Harsh Patil, Tanuja Patil, Vedant Sankalp, Sean Gomes, Aaryan Shaikh
Use Case Importance
Shopping cart and basket detection plays an important role in smart retail systems and automated store management. In supermarkets, malls, and retail stores, monitoring shopping carts and baskets manually can be time-consuming and inefficient. Our AI-powered shopping cart and basket detection system uses Computer Vision and YOLO to detect carts and baskets in real time through CCTV cameras or mobile devices.
This system can help retail stores, supermarkets, and smart shopping centers by automating customer activity monitoring, improving store analytics, reducing manual supervision, and enhancing smart retail operations.
Data Collection and Annotation
We collected a custom dataset of shopping carts and shopping baskets from supermarkets, online image sources, and public retail environments under different lighting and background conditions. The dataset included carts and baskets from multiple angles, distances, and crowded shopping scenarios to improve real-time detection performance.
The dataset contained approximately 1400+ images including:
- Shopping carts
- Shopping baskets
- Different store environments
- Multiple object orientations and scales
All images were manually labelled and annotated using Roboflow, where accurate bounding boxes were created around shopping carts and baskets for effective model training.
The annotated dataset was exported in YOLO format and divided into training, validation, and testing datasets for model evaluation and performance analysis.
Tools & Technologies Used
- Python
- YOLOv9t
- OpenCV
- NumPy
- Google Colab
- Roboflow
- YOLOvX Mobile Application
Model Training & Validation
The model was trained using the YOLOvX framework with the YOLOv9t model for real-time object detection optimized for mobile deployment.
During training, important evaluation metrics were monitored, including:
- Precision
- Recall
- mAP@0.5
- mAP@0.5:0.95
- Loss Curves
We also performed:
- Data augmentation
- Hyperparameter tuning
- Dataset refinement
- Validation testing
- Real-time debugging
to improve detection accuracy and inference performance in real-world retail environments.
The trained model successfully detected shopping carts and baskets in real time with accurate object localization and classification.
Model Deployment & Testing
After successful training, the model was deployed on the YOLOvX mobile application for real-time testing and inference.
The system was tested using:
- Mobile camera feed
- Supermarket-like environments
- Real-time shopping scenarios
- Different lighting conditions
The application displayed:
- Bounding boxes
- Detection labels
- Confidence scores
- Real-time inference results
The model achieved smooth real-time performance and accurate detection on mobile devices.
Challenges Faced
- Collecting diverse shopping cart and basket datasets
- Manual annotation of images
- Handling overlapping objects
- Reducing false detections
- Improving real-time mobile inference speed
- Optimizing model size for deployment
These challenges helped improve practical knowledge in Computer Vision, debugging, dataset preparation, and AI model optimization.
Conclusion
The project successfully demonstrated how AI and Computer Vision can be used for real-time shopping cart and basket detection in retail environments.
Through this project, practical experience was gained in:
- Dataset collection
- Manual image annotation
- YOLO model training
- Real-time object detection
- Mobile AI deployment
- Retail monitoring systems
This project enhanced understanding of Machine Learning, Deep Learning, and Computer Vision concepts while improving implementation, optimization, and problem-solving skills.
Special Thanks
A sincere thanks to Dr. Chandrakant Bothe and the WISERLI Team for their mentorship, guidance, and support throughout the internship journey.
Project Link:
Demo Link- https://drive.google.com/file/d/19jSjCNuPWoKGcajxp8oVub-kuTUQkJW0/view?usp=drive_link