• computer vision
• environmental monitoring
• quadruped robot
• robotic framework

This study presents a computer vision approach to monitor vulnerable European habitats by integrating robotic and artificial intelligence technologies into traditional monitoring methods. The unstructured nature of natural habitats presents a challenge for monitoring activities leading to expensive and time-consuming actions from the human operators. Our solution leverages a quadruped robotic platform, ANYmal C, to navigate unstable, irregular, and rough terrain and to adapt to unstructured environments conditions. This thesis in particular explores the integration of artificial intelligence (AI) with conventional monitoring techniques to enhance the efficiency and accuracy of environmental monitoring. Our multidisciplinary team embarked on a series of field missions to collect extensive data, following meticulously defined guidelines for data acquisition. These missions enabled the creation of a comprehensive dataset, which was meticulously labeled to facilitate the training of object detection and segmentation algorithms. The data collection process encompassed diverse natural habitats, ensuring the robustness and applicability of the developed solutions across various environments.
The first result of this work is a structured guideline for data acquisition in natural habitats and a dataset built across several on-field activities. The ultimate result of this thesis is a computer vision block in a proactive and scalable framework for monitoring natural habitats that utilizes agile and autonomous-legged robots and artificial intelligence algorithms to improve data acquisition efficiency and its evaluation to assist human experts.