Toolbox Detection Model
Team Members
Mohit Kattungal , Varad Kulkarni , Parth Vibhute , Kunal Kadam , Pranay Kamble
**Use Case Importance **
Detecting common tools like hammers, screwdrivers, and pliers in real time can enhance workplace safety, prevent tool misplacement, and optimize workflow in industrial environments. Our solution helps identify tools using computer vision on mobile devices for faster, on-the-go insights.
Data Collection and Annotation
We collected a total of 5204 images from Kaggle, Roboflow, and personal image contributions. The dataset was annotated into five classes: Drill, Hammer, Pliers, Screwdriver, Wrench. Annotation was done using Roboflow. The dataset was split as follows: 3640 training images, 1034 validation images, and 529 test images.
Model Training and Validation
We trained the YOLOv8 model for object detection. Key metrics monitored were Precision, Recall, mAP@0.5, mAP@0.5–0.95, and loss values (box, class, and DFL loss). During training, we analyzed visual metrics like confusion matrices, PR curves, and F1-score curves. Our results were logged in results.csv and visualized in plots like results.png, confusion_matrix.png, and more.
Model Deployment and Demo Video
The trained model was deployed successfully on the YOLOvX mobile app by Wiserli, providing accurate and real-time detection with satisfactory FPS and performance.We deployed only one model that was detection with highest accuracy.
Demo video link: https://drive.google.com/file/d/1DmxQlYZ56tGPFLY8OAatHhC1QQk3Iczv/view?usp=drivesdk
Conclusion :
The model performed well in live mobile tests, accurately detecting tools with good precision and recall. We observed that lighting conditions and background clutter affected performance, and adding a diverse dataset helped improve results. A key learning was optimizing annotation consistency and leveraging YOLOv8’s detailed metrics to fine-tune the model effectively.
Thanks to Dr.Chandrakant Bothe sir for this Internship workshop , which led to upskill myself in terms of using computer vision.
