Wildlife Detection and Tracking

We’ve taken the challenge of detecting and tracking wildlife animals, specifically: wild boars, using thermal vison, unmanned aerial vehicles (UAVs) and machine learning.

Vision

We seek for composition of methods solving the following tasks:

• Detection – wild animals should be detected
• Classification – animals should be recognized and classified, e.g. is this a boar, or a deer?
• Tracking – animals should be trackable, i.e. system operators should be able to track paths of animals of interest
• Counting – the system should be able to count animals it encounters

Dataset

It is possible to get enough data without costly UAV flights. We generate synthetic data using Arma 3 tactical simulator, imitating very closely real conditions in the field. Such approaches were successfully applied to autonomous driving problems. You can see the results of our synthesis right now – it’s the animated background.

Methods

We use:

• Modern 3D graphics rendering for dataset generation
• Machine learning methods for solving detection, classification, tracking and animal counting problems

Authors

• Jakub Cwynar
• Kacper Kania
• Michał Kosturek
• Jakub Michałowski