Michael Hilton

Camera traps are increasingly being used as a research tool in behavioral ecology to observe animals in the field. One challenge of continuous data collection by camera traps is that the amount of data produced can be overwhelming to process manually. Machine learning techniques show promise for becoming important tools to meet this challenge in a time- and cost-effective way. Over the past year, we collected images of over 950 social interactions among gopher tortoises from twelve active tortoise burrows at Boyd Hill Nature Preserve in St. Petersburg, Florida. The individual tortoises in each interaction must be identified to comprehensively analyze this data. To automate the re-identification step, we developed a machine learning algorithm to distinguish between tortoises. It takes a large amount of data to train the machine learning algorithm to recognize patterns. Therefore, we describe how our training dataset was collected and created. We also recount how we trained a Siamese neural network to identify individual tortoises based on their carapace markings in a manner similar to how facial recognition is used on humans. This research demonstrates that machine learning has the potential to be a powerful tool to aid data processing and analysis in behavioral ecology. 

A new open-source image processing pipeline for analyzing camera trap time-lapse recordings is described. This pipeline includes machine learning models to assist human-in-the-loop video segmentation and animal re-identification. We present some performance results and observations on the utility of this pipeline after using it in a year-long project studying the spatial ecology and social behavior of the gopher tortoise.