Periodic Reporting for period 1 - WildDrone (Autonomous Drones for Nature Conservation Missions)
Reporting period: 2023-01-01 to 2024-12-31
Our planet is experiencing a rapid loss of biodiversity, and more species are threatened with extinction now than at any other point in human history. Habitat loss due to agricultural expansion, urbanization, land degradation, overexploitation, and climate change are among the primary factors accelerating populations decline and species extinction. There is an urgent need for more effective and feasible conservation practices to monitor wildlife populations, track their movements, and manage human-wildlife conflicts. Recent advances in technology have paved new avenues for more accurate and larger-scale interventions that can help combat the rapid decline of wildlife populations we witness globally.
WildDrone works to revolutionize wildlife conservation practices by integrating aerial robotics, computer vision, and wildlife ecology, using autonomous drone technology as a unifying platform. We will develop new autonomous systems, expand current software capabilities, and combine them to create practical tools for visual inspection and monitoring of wildlife populations, movement, behaviour, and habitats in complex field settings.
The specific objectives concern scientific Theme 1 Conservation Ecology (O1) Apply drones as a tool for predicting and preventing conflict between wildlife and humans, and (O2) Integrate drone systems with bio-logging technologies to link animal behaviour to physical and ecological contexts; Theme 2 Drone Operations (O3) Develop technologies for responsive, adaptive, and flexible use of drones to observe animals in complex environments, and (O4) Develop techniques for the safe and unobtrusive use of drones to observe animals undetected; Theme 3 Computer Vision (O5) Develop methodologies for vision-based drone control, enabling real time tracking, pose estimation and assessment of animal biometrics, and (O6) Develop principles and techniques for large-scale censuses of animal groups in their environmental context; and finally Training and Communication (O7) Form an interdisciplinary network of young researchers across European and African countries, and (O8) Raise awareness about the nature conservation impact of an interdisciplinary approach using drones, computer vision, and ecology.
WildDrone works to revolutionize wildlife conservation practices by integrating aerial robotics, computer vision, and wildlife ecology, using autonomous drone technology as a unifying platform. We will develop new autonomous systems, expand current software capabilities, and combine them to create practical tools for visual inspection and monitoring of wildlife populations, movement, behaviour, and habitats in complex field settings.
The specific objectives concern scientific Theme 1 Conservation Ecology (O1) Apply drones as a tool for predicting and preventing conflict between wildlife and humans, and (O2) Integrate drone systems with bio-logging technologies to link animal behaviour to physical and ecological contexts; Theme 2 Drone Operations (O3) Develop technologies for responsive, adaptive, and flexible use of drones to observe animals in complex environments, and (O4) Develop techniques for the safe and unobtrusive use of drones to observe animals undetected; Theme 3 Computer Vision (O5) Develop methodologies for vision-based drone control, enabling real time tracking, pose estimation and assessment of animal biometrics, and (O6) Develop principles and techniques for large-scale censuses of animal groups in their environmental context; and finally Training and Communication (O7) Form an interdisciplinary network of young researchers across European and African countries, and (O8) Raise awareness about the nature conservation impact of an interdisciplinary approach using drones, computer vision, and ecology.
Theme 1: DC1 conducted experiments to determine the behavioral tolerance of lions and other predator species to drones, evaluated drones as a tool for monitoring lion populations, and conducted initial trials of nighttime operations using thermal drones. DC2 underwent a second round of recruitment in spring 2024, with the DC beginning his position in September 2024, and focused on developing the research plan and preparing to conduct fieldwork. DC3 developed a plan for his experiment and field data collection, including designing a remotely triggered apparatus to expose visual stimuli in the field, developed skills in training object detection models and image-based tracking, and supported DC5 on a literature review. DC4 used drone-based RGB and thermal imagery to detect and track migrating humpback whales using their “flukeprints” and planned for her first field season in Kenya to take place early 2025. She also supported DC6 on two publications. DC8 executed fieldwork in Denmark, where she focused on drone-based tracking and individual identification of harbor porpoises and collected data on the body condition of captive pinnipeds to develop photogrammetric methods for assessing body condition in their wild populations.
Theme 2: DC5 conducted a literature review on the impact of drone noise on wildlife, defined and tested methods for conducting empirical experiments with drone-induced disturbance of wildlife, and modeled drone propeller noise using computational fluid dynamics. DC6 developed a method to streamline drone software development, experimented with multi-perspective monitoring for wildlife census, collaborated with DC13 to enable DJI platforms for swarm missions, and integrated this with ROS2. DC7 focused on multiple aspects of BVLOS drone operations, ranging from regulatory approvals to fieldwork and software development, including technical work towards securing the necessary permits for operating in Kenya, developing a software tool that automatically generates custom procedures based on mission parameters, developing a web-based UTM system for managing drone operations, and integrating solar panels onto a glider intended for wildlife monitoring. DC13 developed a real-time animal geo-localization pipeline, which can be deployed on off-the-shelf drones.
Theme 3: DC9 developed a pipeline for extraction of scaled 3D shape models along with 3D pose information of animals from aerial data. DC10 developed a prototype for single tree reconstruction, including a data processing pipeline to extract relevant information about plant structure and connectivity. DC11 developed a prototype for point-based animal detection (POLO) and tested this against state-of-the art detectors, proving that the detector is of comparable accuracy, while needing only a fraction of the annotations than traditional, bounding box-based detectors. DC12 developed a prototype of an ultra-fast tracking method, including deployment to onboard hardware for a custom quadcopter drone, allowing the system to achieve near real-time tracking.
Theme 2: DC5 conducted a literature review on the impact of drone noise on wildlife, defined and tested methods for conducting empirical experiments with drone-induced disturbance of wildlife, and modeled drone propeller noise using computational fluid dynamics. DC6 developed a method to streamline drone software development, experimented with multi-perspective monitoring for wildlife census, collaborated with DC13 to enable DJI platforms for swarm missions, and integrated this with ROS2. DC7 focused on multiple aspects of BVLOS drone operations, ranging from regulatory approvals to fieldwork and software development, including technical work towards securing the necessary permits for operating in Kenya, developing a software tool that automatically generates custom procedures based on mission parameters, developing a web-based UTM system for managing drone operations, and integrating solar panels onto a glider intended for wildlife monitoring. DC13 developed a real-time animal geo-localization pipeline, which can be deployed on off-the-shelf drones.
Theme 3: DC9 developed a pipeline for extraction of scaled 3D shape models along with 3D pose information of animals from aerial data. DC10 developed a prototype for single tree reconstruction, including a data processing pipeline to extract relevant information about plant structure and connectivity. DC11 developed a prototype for point-based animal detection (POLO) and tested this against state-of-the art detectors, proving that the detector is of comparable accuracy, while needing only a fraction of the annotations than traditional, bounding box-based detectors. DC12 developed a prototype of an ultra-fast tracking method, including deployment to onboard hardware for a custom quadcopter drone, allowing the system to achieve near real-time tracking.
Theme 1: the DCs focused on planning and conducting fieldwork for ecological data collection and on developing technical skills necessary for subsequent data analysis, and apart from the joint work with DC5 no results beyond the state of the art have been published yet.
Theme 2: results include a review of the impact of drone noise on wildlife (O4, published, DC5), an analysis of and initial simulation-based experiments with multi-perspective monitoring for wildlife census (O3, published, DC6), a review of airspace situational awareness technologies and their respective advantages and limitations (O4, published, DC7), and preliminary experiments with real-time animal geo-localization (O3, published, DC13).
Theme 3: results include a pipeline for extraction of 3D pose estimation of animals (O5, published, DC9), a data processing pipeline for plant structure and connectivity (O5, published, DC10), and development and test of POLO, a prototype point-based animal detector made available on a public repository (O6, published, DC11).
Theme 2: results include a review of the impact of drone noise on wildlife (O4, published, DC5), an analysis of and initial simulation-based experiments with multi-perspective monitoring for wildlife census (O3, published, DC6), a review of airspace situational awareness technologies and their respective advantages and limitations (O4, published, DC7), and preliminary experiments with real-time animal geo-localization (O3, published, DC13).
Theme 3: results include a pipeline for extraction of 3D pose estimation of animals (O5, published, DC9), a data processing pipeline for plant structure and connectivity (O5, published, DC10), and development and test of POLO, a prototype point-based animal detector made available on a public repository (O6, published, DC11).