Smoke Opacity Meter for Open-Burn Monitoring

ISEF Category: Environmental Engineering

Subcategory: Pollution Control  ·  Difficulty: Advanced  ·  Setup: University Lab  ·  Time: Full Year

The Hook

Smoke can look “light” to your eyes and still carry a lot of pollution. That makes open-burn enforcement hard. A camera can help, but only if you teach it to judge smoke the same way every time. This project turns a messy visual problem into a measurable signal.

What Is It?

This project asks you to estimate how dark smoke looks from images or video. The classic reference is the Ringelmann chart, a simple visual scale that rates smoke opacity. Opacity means how much light smoke blocks. In plain terms, darker smoke blocks more of what you see through it.

A Raspberry Pi camera can collect images, and a machine learning model can sort them into Ringelmann-like classes. Think of it like teaching a computer to compare smoke against a ruler, but the ruler is made of visual patterns instead of inches. Your job is to build a system that turns camera images into consistent scores you can test against human labels or known references.

The hard part is not just spotting smoke. It is separating smoke from background sky, lighting changes, and camera angle. That makes this a strong project for learning image processing, classification, and validation.

Why This Is a Good Topic

This is a good science fair topic because you can test it with images, clear labels, and repeated measurements. It connects to real pollution control, since smoke opacity affects air quality rules and open-burn enforcement. You can learn computer vision, model evaluation, and data collection without needing a wet lab. A strong version also lets you compare human scoring against automated scoring.

Research Questions

• How does camera position affect smoke opacity classification accuracy?
• What is the effect of background type on Ringelmann-classifier performance?
• Does adding image preprocessing improve agreement with human smoke ratings?
• To what extent can a Raspberry Pi model distinguish adjacent Ringelmann opacity levels?
• Which lighting conditions cause the largest classification errors?
• How does training data size affect smoke-opacity prediction accuracy?
• What is the effect of using video frames instead of still images on measurement consistency?

Basic Materials

• Raspberry Pi board with camera module.
• Tripod or fixed mount for the camera.
• SD card with operating system installed.
• Laptop or desktop computer for coding and data analysis.
• Free image dataset or your own labeled smoke images.
• Printed Ringelmann chart or digital reference images.
• Notebook for labeling observations.
• Light-colored backdrop or outdoor reference scene for controlled image capture.

Advanced Materials

• Raspberry Pi board with camera module and stable power supply.
• Calibrated reference targets for image correction.
• Tripod with adjustable height and angle markings.
• Portable light meter or lux meter.
• Annotated smoke image dataset with class labels.
• Laptop with Python environment for model training and evaluation.
• External storage for image logging.
• Optional spectral or air-quality sensor for comparison measurements.

Software & Tools

• Python: Runs image processing, model training, and accuracy analysis.
• OpenCV: Handles image capture, preprocessing, and feature extraction.
• TensorFlow Lite: Runs a small image classifier on Raspberry Pi hardware.
• ImageJ: Helps inspect image brightness, contrast, and region-of-interest measurements.
• Jupyter Notebook: Organizes your analysis, plots, and model comparisons.

Experiment Steps

1. Define the exact smoke classes you will predict and how they map to the Ringelmann scale.
2. Plan how you will capture images so camera angle, distance, and background stay consistent.
3. Build a labeled image set and decide how you will split it into training, validation, and test groups.
4. Choose one preprocessing path first, then compare it against a simpler baseline model.
5. Design controls that separate smoke signal from lighting, haze, and motion blur.
6. Plan how you will score performance with confusion matrices, error rates, and agreement with human labels.

Common Pitfalls

• Training on images from one background, then testing on another, which makes the model fail when the scene changes.
• Using auto-exposure without control, which shifts brightness and breaks opacity comparisons.
• Labeling images too broadly, which blurs nearby Ringelmann classes and lowers accuracy.
• Collecting too few examples of dense smoke, which leaves the model weak on the hardest cases.
• Confusing smoke color with smoke opacity, which mixes two different visual signals.

What Makes This Competitive

A stronger project goes beyond a simple classifier. You can compare multiple camera setups, test whether the model agrees with trained human raters, and measure error by opacity class instead of only overall accuracy. You can also study how much preprocessing helps under different lighting conditions. That kind of careful validation makes the project feel like real monitoring research, not just an app demo.

Project Variations

• Use drone video frames instead of a fixed Raspberry Pi camera to test whether motion changes opacity scoring.
• Compare Ringelmann-based classification with a regression model that predicts continuous smoke darkness.
• Test the same model on wildfire smoke, open-burn smoke, and steam to see which aerosols confuse the classifier most.

Learn More

• NOAA Air Resources Laboratory: Search for materials on smoke, visibility, and atmospheric optics to understand why haze changes image contrast.
• NASA Earthdata: Look for tutorials and datasets on aerosol and smoke imaging to study remote sensing methods.
• PubMed: Search for review articles on smoke opacity, particulate matter, and visual exposure assessment.
• US EPA Air Quality Resources: Find background on open burning, smoke regulations, and air pollution measurement standards.
• OpenCV Documentation: Read the official tutorials for image capture, preprocessing, and object detection basics.