AI Trail Cameras for Wildlife Detection

This thesis explores the use of AI-enhanced trail cameras, focusing on low-power, small-formfactor devices such as the ESP32-CAM and Raspberry Pi Zero 2W, to assess their effectiveness for wildlife detection. The study evaluates the performance, power consumption, and reliability of these devices when paired with TinyML models like MobileNet, comparing them with traditional non-AI camera setups. The goal is to identify popular, easily accessible, low-power, low-cost devices and to implement a model that is easy to modify and iterate, enabling use by researchers, ecologists, and conservation officers. This study focuses on evaluating usability of devices in real-world environment away from internet and power connectivity. We review prior work, then develop a prototype and a simple benchmark for power and accuracy evaluation. These prototypes are then tested in a controlled environment to collect data on their ability to detect and classify wildlife accurately.