AUTHOR=Marsh Peter K. , Kurfess Franz J. TITLE=A software pipeline for automated wildlife population sampling JOURNAL=Frontiers in Conservation Science VOLUME=4 YEAR=2023 URL=https://www.frontiersin.org/journals/conservation-science/articles/10.3389/fcosc.2023.1219482 DOI=10.3389/fcosc.2023.1219482 ISSN=2673-611X ABSTRACT=

Ecologists today face significant challenges in accurately modeling wildlife populations. Population surveys provide an essential understanding of an ecosystem; however, they currently require an extensive amount of labor and resources to carry out which limits the frequency at which they are conducted. Lack of population data presents a significant barrier to ecologists in their ability to understand and model interactions between species and their surroundings. Preliminary work has been done in employing consumer drones and object detection software to automate data collection and processing on large mammal species. Such work suggests these technologies can significantly ease the process of data collection while maintaining an accuracy comparable to manual surveying techniques. While previous studies indicate the use of drone and object detection technology can aid in the collection of population data, there remain significant barriers in applying such methods to aid in ecological research on a broader scale. In particular, using object detection to identify target individuals involves combining many software tools, each of which comes with its own challenges and complexities. This paper presents a flexible software framework for automated population sampling that is accessible to researchers in the field of wildlife research. To achieve this we combine orthomosaic stitching, object detection, label post-processing, and visualization solutions into a single software pipeline. We then show how such a pipeline can be run in the cloud and provide documentation for others to replicate this process. Finally, we use a consumer drone and free navigation software to demonstrate the proposed workflow on a herd of cattle and assess its viability in providing useful population data.