AI Wildlife Camera Traps for Low-Power Networks

ISEF Category: Embedded Systems

Subcategory: Internet of Things  ·  Difficulty: Advanced  ·  Setup: University Lab  ·  Time: Full Year

The Hook

Most wildlife camera traps waste power on empty frames. That means dead batteries, lost data, and more field trips to swap gear. Your project asks a smart question, can a tiny AI camera decide what is worth sending before it ever leaves the forest?

What Is It?

This project combines three ideas, a camera trap, a tiny AI model, and a low-power wireless network. A camera trap is a remote camera that wakes up when something moves. YOLO-nano is a small object-detection model that can recognize animals on the device itself, so the camera does not need to send every empty photo.

Think of it like a bouncer at the door. The camera checks each frame, then only sends the ones that matter. LoRa mesh networking helps those useful frames travel long distances with little energy. Adaptive duty-cycling means the device changes how often it wakes up based on predicted animal activity. If solar radiation and temperature suggest animals are more active, the system can sample more often. If conditions look quiet, it can slow down and save battery.

Why This Is a Good Topic

This is a strong science fair topic because you can test clear tradeoffs. You can measure battery life, transmission count, false alarms, and missed detections. The real-world problem is easy to understand, wildlife researchers need better monitoring tools that last longer in remote places. You can also learn embedded systems design, edge AI, sensor data, and wireless communication without needing a full research lab to understand the system.

Research Questions

• How does on-device YOLO-nano filtering change the number of images sent over LoRa compared with sending every frame?
• What is the effect of adaptive duty-cycling on battery life under different temperature and solar-radiation patterns?
• Does adding animal-activity prediction reduce missed detections compared with a fixed wake schedule?
• To what extent does mesh routing improve data delivery in wooded or uneven terrain compared with a single-hop link?
• Which confidence threshold gives the best balance between false positives and empty-frame filtering?
• How does day versus night lighting affect YOLO-nano detection accuracy on a wildlife camera trap?
• What is the effect of different compression settings on transmission success and image usefulness?

Basic Materials

• Microcontroller board with camera support, such as ESP32-CAM or similar.
• LoRa transceiver modules matched for your region.
• Rechargeable battery pack with power meter or inline current sensor.
• Small solar panel and charge controller for field-style testing.
• Temperature sensor.
• Light sensor or solar-radiation proxy sensor.
• SD card module or onboard storage.
• Laptop for data logging and analysis.
• Test images or short video clips with animals and empty frames.
• Enclosure for outdoor-style testing.
• Basic wiring kit and breadboard or perfboard.

Advanced Materials

• Single-board computer or microcontroller capable of running tiny inference.
• Camera module with adjustable exposure settings.
• Multiple LoRa nodes for mesh testing.
• Power analyzer for fine current profiling.
• Environmental sensor suite with temperature, light, and humidity.
• GPS module if you want location-aware testing.
• Solar charging hardware for field endurance tests.
• Dataset of local wildlife images for model evaluation.
• Development board for edge-AI optimization.
• Waterproof enclosure and cable glands for outdoor deployment.

Software & Tools

• Python: Organizes sensor data, evaluates model performance, and plots power use.
• ImageJ: Helps compare image quality, brightness, and detection visibility across test conditions.
• Edge Impulse: Supports tinyML model training and deployment to small devices.
• QGIS: Maps node placement and field-test coverage.
• Arduino IDE: Uploads firmware and logs embedded device behavior.

Experiment Steps

1. Define the exact performance target, such as fewer empty transmissions, longer battery life, or higher animal detection rate.
2. Choose one sensing workflow to compare against a baseline that sends every frame.
3. Decide how you will score each frame, each transmission, and each battery test so the results become measurable.
4. Build a validation set that includes animal images, empty scenes, and hard cases like shadows, branches, and partial animals.
5. Plan controls for network range, lighting, and temperature so changes in performance come from your design, not the environment.
6. Set up a comparison between fixed sampling and adaptive duty-cycling, then analyze which design gives the best tradeoff.

Common Pitfalls

• Training YOLO-nano on too few wildlife images, which makes the model fail on new animals or new angles.
• Testing the camera only in ideal daylight, which hides how badly detection drops at dawn, dusk, or night.
• Measuring battery life without separating sensor use, inference use, and LoRa transmission use, which makes the power results hard to interpret.
• Using a prediction rule for duty-cycling that follows temperature alone, which can miss the fact that animal activity also depends on habitat and season.
• Ignoring mesh routing failures in cluttered terrain, which can make the system look better in the lab than it does in a real field path.

What Makes This Competitive

A competitive version of this project does more than prove that AI can save power. It compares multiple system designs, such as fixed duty-cycling, rule-based duty-cycling, and prediction-based duty-cycling. It also reports more than one metric, like detection accuracy, empty-frame reduction, latency, and battery drain. Strong students add a careful error analysis that explains when the system fails and why.

Project Variations

• Test the same system on bird feeders, trail crossings, or backyard mammals instead of a forest setting.
• Compare YOLO-nano with a simpler motion-trigger-only baseline to see how much edge AI actually helps.
• Swap solar-radiation and temperature predictors for time-of-day or humidity-based activity models and compare which one schedules captures best.

Learn More

• NASA Earthdata: Search for free satellite and environmental data that can help you think about solar and temperature patterns.
• NOAA Climate Data Online: Find weather and climate records for matching field conditions to animal activity.
• USGS Science Explorer: Read public science articles on wildlife monitoring, sensors, and field data collection.
• NIH PubMed: Search review articles on edge AI, tiny object detection, and low-power embedded sensing.
• MIT OpenCourseWare: Look for free courses on embedded systems, machine learning, and wireless communication.
• IEEE Xplore: Read abstracts and, when available through a school library, find papers on LoRa mesh, camera traps, and energy-aware sensing.