Smart and automated web-services to analyze wildlife image data

ICOET 2021
Authors
David Waetjen, Road Ecology Center, UC Davis
Fraser Shilling, Road Ecology Center, UC Davis
Abstract

Wildlife camera traps are essential equipment when monitoring animal movement and occupancy in a region or near infrastructure-crossing structures. Large arrays of cameras (dozens to hundreds) result in large numbers (hundreds of thousands) of images, especially when any vegetation or traffic are in the camera view. Rapidly and accurately processing images through most workflows can involve a lot of staff time and potentially result in transcription and other errors. For this investigative method to be efficient and accurate at large scales, it is necessary to automate certain steps to reduce labor and increase accuracy and consistency. The Road Ecology Center at UC Davis has developed a set of tools and platforms for managing camera trap projects and photographs to solve some of these workflow needs: (1) Cam-WON provides the project management and photo tagging features, allowing an operator to map their locations and compile the image data associated with each camera position; (2) EventID measures the rate of successful crossing or repulsion by using image data from both sides of a structure, comparing the times and species for these important events; (3) BehaviorID provides the ability to annotate an animal’s behavior within captured video (and series of photographs), generating graphical summaries of an animal’s activities and proportion of time engaging in that activity, such as grooming, feeding, or walking; and (4) ImageID, which uses cutting edge artificial intelligence to classify images as containing animals, or as false positives (no animals), which can be a substantial time saver. We and others have used Cam-WON to manage >40 camera trap projects, providing a consistent platform for image management. We and others also use a standard workflow involving these web-services, beginning with ImageID to process raw imagery from SD cards and identify animal-containing images, then uploading and animal identification in Cam-WON and finally using EventID and BehaviorID to classify activity and behavior. These are novel tools for camera array users to improve wildlife monitoring within their organization and we are integrating this work into a more streamlined workflow, requiring fewer steps by camera trap operators to utilize these tools.