Continuous Monitoring of Gopher Tortoise Burrows Using Time-Lapse Imaging and Machine Learning

Camera trap time-lapse recordings can collect vast amounts of data on wildlife in their natural habitat. Transforming these data into information useful to ecologists is a major challenge. Machine learning techniques show promise for becoming important tools to meet this challenge in a cost-effective way. Over the past year, we recorded 5-second interval time-lapse video of twelve active gopher tortoise burrows at Boyd Hill Nature Preserve in St. Petersburg, Florida, generating more than 100 Terabytes of data in the process. Herein, we describe a suite of open-source software tools we developed to manage the collection and analysis of these data, and present preliminary results on tortoise activity levels at this study site. The tools incorporate a convolutional neural network trained to detect gopher tortoises and to generate a draft video segmentation marking when tortoises are present. These tools allow a single human grader to review and refine the draft segmentations for a week’s worth of time-lapse recordings (11.5 hours of video if played back at standard speed) in under 3 hours.  This research demonstrates that the tools developed can facilitate future studies across research groups to assess key population features as well as to remotely monitor wildlife populations efficiently.