Pet ID Vision Feeder for Smart Feeding

ISEF Category: Embedded Systems

Subcategory: Other  ·  Difficulty: Intermediate  ·  Setup: Home Setup  ·  Time: 1 to 2 Months

The Hook

A feeder that serves the wrong cat is not a small glitch. It can mean the wrong diet, the wrong portion, or missed medication. Your project turns that everyday problem into a computer vision test. You get to ask when a tiny camera can tell pets apart, and when it fails.

What Is It?

This project uses an ESP32-CAM, which is a small microcontroller with a camera, to spot which pet is in front of a feeder. Think of it like a tiny security guard that checks an ID card before opening the snack door. The system looks at image features, then decides whether to dispense food for one animal or another.

The science part is not just building the feeder. You also test recognition accuracy under different lighting conditions. That means you measure how well the camera and model keep working when the room gets brighter, dimmer, warmer, or shadowed. In plain terms, you are asking how much visual noise the system can handle before it starts confusing one pet for another.

This project sits between embedded systems and computer vision. Embedded systems means a small computer does a specific job with limited power and memory. Computer vision means software reads images and makes decisions from them. Your job is to make the hardware, software, and test plan work together.

Why This Is a Good Topic

This is a strong science fair topic because you can measure it clearly. You can compare recognition accuracy across lighting setups, camera angles, pet positions, and background clutter. You can also connect it to a real problem, since multi-pet homes need better feeding control, allergy management, and medication tracking. A student can realistically build a small prototype, collect image data, and analyze mistakes without needing a university lab.

Research Questions

• How does lighting intensity affect pet recognition accuracy on an ESP32-CAM feeder?
• What is the effect of background clutter on false pet identification?
• Does changing camera angle improve correct identification of individual pets?
• To what extent does image resolution change the feeder's classification accuracy?
• Which lighting condition produces the most stable recognition across repeated trials?
• How does adding simple image preprocessing affect accuracy in low-light scenes?

Basic Materials

• ESP32-CAM board with programmer
• USB cable and stable power supply
• Breadboard and jumper wires
• Small servo motor or relay module for the feeder gate
• Cardboard, foam board, or 3D-printed feeder enclosure
• Treats or dry food for safe testing
• Printed or taped pet ID markers for controlled trial setup
• Smartphone or camera for reference photos
• Desk lamp with adjustable brightness
• Colored paper, white paper, and blackout cloth for background tests
• Notebook or spreadsheet for recording trials.

Advanced Materials

• ESP32-CAM board with camera module
• MicroSD card module for local image storage
• Servo-driven or auger-based feeder prototype
• Lux meter for measuring light levels
• Calibrated color targets or checkerboard pattern for camera checks
• External regulated power supply
• 3D printer or laser cutter for enclosure parts
• Image reference dataset collected from each pet
• Computer for model training and analysis
• Optional edge ML toolchain for deployment tests.

Software & Tools

• Arduino IDE: Programs the ESP32-CAM and logs device behavior during trials.
• Python: Organizes images, labels data, and runs accuracy analysis.
• OpenCV: Helps you inspect image quality, lighting shifts, and preprocessing steps.
• ImageJ: Measures brightness and color changes in test images.
• Google Sheets: Tracks each trial and summarizes recognition results.

Experiment Steps

1. Define the exact task your feeder must solve, such as identifying one pet from another before dispensing food.
2. Choose the image features you want to compare first, such as lighting, angle, background, or pet position.
3. Plan a repeatable test setup so each trial uses the same camera distance, feeder location, and reference labels.
4. Build a baseline recognition method, then decide how you will measure accuracy, false matches, and missed detections.
5. Design control conditions that separate lighting effects from other changes, like movement or background clutter.
6. Set up a data table and analysis plan before testing so you can compare conditions with clear statistics.

Common Pitfalls

• Training or testing with only one lighting setup, which makes the system fail as soon as the room changes.
• Using too few pet images, which makes the model memorize one pose instead of recognizing the animal.
• Letting the background change between trials, which confuses the camera and hides the real lighting effect.
• Measuring only whether the feeder opened, which misses false positives, false negatives, and near-miss identifications.
• Skipping repeated trials for each pet, which makes random motion look like a real pattern.

What Makes This Competitive

A stronger version of this project goes past a simple demo. You would compare several lighting conditions, test more than one recognition method, and report both accuracy and error type. You could also show whether simple preprocessing or camera placement changes make the system more reliable. Clear controls, repeated trials, and thoughtful analysis can make the project feel like real engineering research.

Project Variations

• Test whether infrared or low-light illumination improves recognition in dim rooms.
• Compare face-based identification with tag-based identification to see which method is more reliable for a feeder.
• Measure how well the system works when two pets approach the feeder at the same time.

Learn More

• ESP32-CAM documentation from Espressif: Search the official Espressif documentation for camera setup, pinouts, and examples.
• OpenCV documentation: Search the official OpenCV docs for image preprocessing and classification support.
• MIT OpenCourseWare, Computer Vision: Search MIT OpenCourseWare for free lecture notes on image processing and computer vision.
• NIH PubMed: Search review articles on pet recognition, embedded vision, and animal behavior monitoring.
• IEEE Xplore or arXiv: Search for papers on low-power computer vision on edge devices and pet identification systems.