Joshua J. Millspaugh - US grants

Abstract

One of the greatest challenges in understanding the role of free-ranging wildlife in maintaining diversity, tracking invasive species, and the spread of emerging diseases is obtaining unobtrusive visual information required in studying the behaviors and interactions of wildlife species in their environment. The state-of-the-art wildlife monitoring technologies, including radio-tracking and wireless sensor networks, do not provide sampling of animal resource selection, behavior patterns, and the environmental context of the animal's behavior. The central objective of this interdisciplinary award is to bring the video monitoring capability to wireless sensor networks so as to collect important visual information for wildlife behavior analysis and interaction modeling. The overall goal of the research is to develop a long-lived and unobtrusive wildlife video monitoring system capable of real-time video streaming with remote control capability. The captured video in real time will be transmitted over wireless sensor networks to a remote monitoring center for real-time viewing and camera control. Because real-time transmission requirements are particularly challenging to wireless sensor network design, the research will address important issues on energy minimization and performance optimization in video sensing over mobile wireless sensor networks. The PIs will engineer a portable, low-energy, rugged, wireless network video/GPS/motion sensor coupled with an optimized transmission protocols and routing schemes for transferring the video images through sensor node design, access control, and robust routing protocol. Deer-net is a highly innovative interdisciplinary engineering, computer science, and wildlife science project in the field of cognitive ecology. The video monitoring has the potential to significantly advance this area of science and engineering and has broader application to other fields including surveillance, security, process monitoring, and other industrial applications. The broader impacts will arise from multi-disciplinary collaboration among computer scientists, engineers, and wildlife biologists; participation by students spanning undergraduate to doctoral levels; integration into ongoing courses; recruitment of under-represented groups; and educational opportunities for K-12 students to observe wildlife behavior.

The University of Missouri and the University of Illinois at Urbana-Champaign are awarded collaborative grants to develop advanced computational and informatics tools that will support wildlife data collection, analysis, and management at large scales. Project objectives include investigation of 1) advanced computer vision methods for detecting and tracking animals in dynamic and cluttered environments; 2) adaptive classification, machine learning, and information fusion methods for recognizing animal species and individual ID; and 3) data summarization and database management schemes to support collaborative wildlife research. The performance of these computational and informatics tools will be evaluated using existing camera trap datasets and field studies in terms of their potential to support collaborative wildlife research.

This project will broadly advance the state-of-the-art in computer vision, wildlife monitoring, ecology, and conservation research. It will provide new methods and tools for automated processing and mining of massive wildlife monitoring data at large scales. This will allow individual or coordinated networks of wildlife researchers to analyze and manage camera-trap data with minimum effort and compare and share data between research groups across different geographical regions. Collaborative wildlife monitoring and tracking at large geographical and time scales will help us understand the complex dynamics of wildlife systems, evaluate the impact of human actions and environmental changes on wildlife species, and answer many important wildlife, ecological, and conservation research questions. The database will be hosted by Smithsonian. This will provide exciting interdisciplinary opportunities for mentoring graduate students and involving K-12 and undergraduate students into professionally guided research. Software and results of this project will be available from the website http://videonet.ece.missouri.edu.

Understanding the processes determining abundance of terrestrial wildlife communities across large scales and estimating the abundance of wildlife across large areas remain a major challenge. For most species, the factors that regulate their distribution and yearly fluctuations in population size are unknown. This project takes advantage of the key innovation of using motion sensitive camera traps as a network of sensors for estimating animal abundance. This new approach can produce abundance data for any terrestrial animal >100g, typically ~60% of the terrestrial animals in the eastern USA. Furthermore, the method is amenable to citizen science programs, without the biases or data quality issues typical of other programs, opening the possibility of a sustainable dense sampling effort across large areas. Software will be developed efficiently enter and to manage camera images and associated data. Using standardize field techniques citizen?s groups will sample their local wildlife communities with cameras. The resulting data will be analyzed using new multi-scale statistical models to discover the processes regulating wildlife abundance over large areas. Mapping the local abundance of wildlife populations across broad areas will be key to understanding the mechanisms responsible for changes resulting from land-management decisions and regional climate variation. By involving citizens in data collection this project will be helping local wildlife populations. All data and images will be made freely available online, providing a tool not only for scientists, but also to give the public a new window into the animal communities of their region.

Minimizing the impact of human actions on wildlife is a priority for conservation biology. Distributed motion-sensitive cameras, or camera traps, are popular tools for monitoring wildlife populations. Recent work has shown that camera trap surveys can be expanded to large scales by crowdsourcing through citizen science, producing big data sets needed to evaluate the effect of sustainability strategies on wildlife populations. However, these large-scale surveys create millions of photographs that create new challenges for data processing and quality control.

This project seeks to develop advanced computing technologies for cloud-based large-scale data sensing, analysis, annotation, management, and preservation for wildlife and ecological sustainability to inform effective resource management, decision-making, and polices on human actions to protect wildlife and natural resources. Specifically, the project aims to: (1) explore a citizen scientist-based approach and system for large-scale sustainable data collection; (2) develop deep-learning based fine-grain animal species recognition from large data sets and automated content annotation; (3) study cloud-based computing with thin-client access and resource allocation for scalable deployment and easy access by citizen scientists; and (4) develop a comprehensive data annotation quality monitoring and control framework with tightly coupled computer annotation, crowd-sourcing, and expert review to ensure a high scientific standard of data quality. These tools will be integrated into the eMammal infrastructure to study three questions on wildlife sustainability: energy development, housing development, and wildlife harvest. These tools will also be available to other wildlife researchers using the eMammal system, enabling an improved understanding how humans can live sustainably with wildlife.

Collaborative wildlife monitoring and tracking at large geographical and time scales will contribute to the understanding of complex dynamics of wildlife systems, and provide important scientific evidence for informed decisions and effective solutions to sustainability issues in wildlife environments. This project will provide unique, exciting, and interdisciplinary opportunities for mentoring graduate students and involving K-12 and undergraduate students into professionally guided research. The citizen science approach used in this project should accommodate hundreds of students in research.