5732

The main goal is to solve a computer vision task, which is to be able to identify native animals of Greenland on a camera image and create log entries in case of detection. The application shall be based on a YOLO real-time machine learning model or similar. The AI model shall be deployed directly on a local edge device (STM32N6570-DK Discovery kit ). The device should be able to send the log messages via LoRa communication or similar.​

The Arctic region is a unique part of Earth's ecosystems. Even in the 21st century, we still have much to learn about Arctic animals: their migration routes, how climate change affects them, population trends, is their population increasing or decreasing? 

Some of these animals, such as polar bears, can also pose a danger to humans, for example, to polar researchers working in the region. Because of this threat, researchers must organize polar bear watches during fieldwork to avoid direct encounters. 

This use case focuses on automating the task of animal monitoring. The proposed solution is a detection and alert system that helps polar researchers detect nearby animals as early as possible.

To power the setup, I used 3 lithium batteries (Samsung INR18650-30Q), which are placed within a transparent Peli M60 transparent case. This case serves as protection (with silica gel beads applied as well) against the environmental conditions. The box itself was installed to a tripod stand.

The code is based on the example program "GettingStartedObjectDetection" from ST. Added functionality to switch off LCD display on trigger from the User button.
Attempted to implement saving images on the microSD card, but I could not add the fatFS library properly. Expected finish for the project is end of 2025.