Vision AI Project Submission
Use Case Title
Tool Bits Detection – Detect and Count Screwdriver Bits
Team Members
Aayush Kushwaha
Shubham Lande
Mackwin Lobo
Bhavesh Mahale
Deepali Mahto
Use Case Importance
In manufacturing and repair environments, missing or incorrect screwdriver bits can lead to product defects or safety hazards. Automating the detection and count of tool bits improves inventory accuracy and operational efficiency.
Data Collection and Annotation
We used an existing dataset of 830 images of screwdriver bits, captured in varied lighting and angles.
- Classes annotated: 06
- Annotation tool: Roboflow
- Annotation format: YOLO format
- Total images: 830
- Train: 650
- Validation: 111
- Test: 69
- Sample images:
Model Training and Validation
- Model used: YOLOv8
- Metrics monitored: mAP50, mAP50-95, Precision, Recall
- Initial training showed lower performance on closely placed bits, so we added more diverse angles in the training set.
- Final training data: 650 images for training, 111 for validation
- Training settings: 50 epochs, image size 640, batch size 16, patience=5, save_period=5
Model Deployment and Demo Video
- Deployment Platform: YOLOvX mobile app
- Model performance:
- Accuracy: ~88% mAP50
- Real-time detection speed: ~24 FPS on mobile
- Number of deployed models: 1 (Tool Bits Detection)
- Demo Video Link: Tool Bits Detection Demo
https://drive.google.com/file/d/1CfqMagxCNWlNaPe2E4CPnc-4DeWBB9WA/view?usp=drivesdk
Conclusion
Our model successfully detects and counts screwdriver bits with high accuracy and good inference speed.
Key challenges included differentiating overlapping or rotated bits. We learned that model performance improves significantly with well-structured annotations and diversity in training data.
