What I Learned During the Wiserli AI Internship Workshop

Use Case:

Real-Time Stove Flame Detection Using YOLOv9 for Smart Kitchen Safety Applications

Team Members

  • Nihar Sawant
  • Aadarsh Shrivastav
  • Srujan Poojari
  • Sameer Vishwakarma

Use Case Importance

Stove flame detection plays an important role in smart kitchen safety, fire prevention systems, gas monitoring, and intelligent home automation. Detecting active stove flames in real-time can help reduce fire hazards, prevent gas leakage accidents, and improve AI-powered kitchen monitoring systems.

Our solution uses YOLOv9 for real-time stove flame detection with high speed and reliable accuracy. The system can detect stove burners and flames under different lighting conditions, camera angles, distances, and partially obstructed kitchen environments.

Data Collection and Annotation

We created a custom dataset by collecting stove burner and flame images from online datasets and real kitchen environments under different backgrounds and lighting conditions.

The dataset included:

  • Burning Stove Burners
  • Non-Burning Burners
  • Multiple Stove Types
  • Indoor Kitchen Environments
  • Different Camera Angles & Distances
  • Low-Light and Bright-Light Conditions

We used annotation tools and bounding box labeling techniques for dataset preparation and model training.

Dataset Split:

  • Train: 70%
  • Validation: 20%
  • Test: 10%

We also applied data augmentation techniques such as:

  • Rotation
  • Flipping
  • Brightness Adjustment
  • Scaling
  • Cropping
  • Blur & Noise Augmentation

These techniques improved model generalization and detection stability in real-world kitchen environments.

Model Training and Validation

We trained our model using YOLOv9 for fast and accurate real-time object detection.

The model was evaluated using:

  • Precision
  • Recall
  • F1-Score
  • mAP (Mean Average Precision)
  • Loss Metrics

The model achieved reliable detection performance even in low-light kitchen environments and partially visible stove burner scenarios.

After observing false detections during initial testing, we improved the dataset quality by adding more diverse stove and flame samples and retraining the model for better accuracy and stability.

Sample Outputs

  • Real-time stove flame detection with bounding boxes
  • Confidence score visualization
  • Detection of burning and non-burning burners
  • Multiple burner detection support

Model Deployment and Demo

The trained model was deployed using the YOLOv9 ecosystem for real-time inference and testing.

Deployment Details:

  • Model Used: best.pt
  • Framework: YOLOv9
  • Application: Real-Time Stove Flame Detection System

Performance:

  • Real-time stove flame detection with bounding boxes and confidence scores
  • Good performance under varying lighting conditions
  • Stable inference performance in real kitchen environments
  • Improved small object detection capability
  • Reduced false detections during testing

Applications

  • Smart Kitchen Safety Systems
  • Fire Hazard Detection
  • Gas Stove Monitoring
  • AI-Based Home Automation
  • Intelligent Safety Monitoring
  • Smart Building Safety Systems
  • Real-Time Kitchen Monitoring

Conclusion

Our project successfully demonstrates a real-time stove flame detection system using YOLOv9 for intelligent kitchen safety applications.

The project helped us understand:

  • Object Detection Workflow
  • Dataset Collection & Annotation
  • Data Augmentation Techniques
  • Real-Time AI Deployment
  • Model Optimization & Performance Tuning
  • Computer Vision & Deep Learning Applications

The system achieved reliable real-time detection performance and demonstrated the practical use of AI-powered vision systems in smart safety monitoring applications.

:pray: Special Thanks

We sincerely thank our mentors, trainers, and coordinators for their valuable guidance and support throughout the project development process.

Trainers:

  • Priti Singh
  • Rugvedi Khandekar

Mentors:

  • Mr. Sandeep Dwivedi Sir
  • Dr. Chandrakant Bothe Sir