<?xml version='1.0' encoding='utf-8'?>
<oai_dc:dc xmlns:dc="http://purl.org/dc/elements/1.1/" xmlns:oai_dc="http://www.openarchives.org/OAI/2.0/oai_dc/" xmlns:xsi="http://www.w3.org/2001/XMLSchema-instance" xsi:schemaLocation="http://www.openarchives.org/OAI/2.0/oai_dc/ http://www.openarchives.org/OAI/2.0/oai_dc.xsd">
  <dc:contributor>Richard J. Camp</dc:contributor>
  <dc:contributor>Matthew D. Burt</dc:contributor>
  <dc:contributor>Scott Vogt</dc:contributor>
  <dc:creator>Trevor Bak</dc:creator>
  <dc:date>2025</dc:date>
  <dc:description>&lt;p&gt;Context&lt;/p&gt;&lt;p&gt;&lt;span&gt;Population monitoring is an essential need for tracking biodiversity and judging efficacy of conservation management actions, both globally and in the Pacific. However, population monitoring efforts are often temporally inconsistent and limited to small scales. Motion-activated cameras (‘camera traps’)&amp;nbsp;offer a way to cost-effectively monitor populations, but they also generate large amounts of data that are time intensive to process.&lt;/span&gt;&lt;/p&gt;&lt;p&gt;&lt;span&gt;Aims&lt;/span&gt;&lt;/p&gt;&lt;p&gt;&lt;span&gt;To develop an automated pipeline for processing videos of ungulates (Philippine deer,&amp;nbsp;&lt;i&gt;Rusa marianna&lt;/i&gt;;&amp;nbsp;and pigs,&amp;nbsp;&lt;i&gt;Sus scrofa&lt;/i&gt;) on Andersen Air Force Base in Guam.&lt;/span&gt;&lt;/p&gt;&lt;p&gt;&lt;span&gt;Methods&lt;/span&gt;&lt;/p&gt;&lt;p&gt;&lt;span&gt;We processed camera videos with a machine learning model for object detection and classification. To estimate density using distance sampling methods, we used a separate machine learning model to estimate the distance of target animals from the camera. We compared density estimates generated using manual versus automated methods and assessed accuracy and processing time saved.&lt;/span&gt;&lt;/p&gt;&lt;p&gt;&lt;span&gt;Key results&lt;/span&gt;&lt;/p&gt;&lt;p&gt;&lt;span&gt;The object detection and classification model achieved an overall accuracy &amp;gt;80% and F1 score ≥0.9 and saved 36.9&amp;nbsp;h of processing time. The automated distance estimation was fairly accurate, with a 1.1&amp;nbsp;m (±1.4&amp;nbsp;m) difference from manual distance estimates, and saved 16.8&amp;nbsp;h of processing time. Density estimates did not differ substantially between manual and automated distance estimation.&lt;/span&gt;&lt;/p&gt;&lt;p&gt;&lt;span&gt;Conclusions&lt;/span&gt;&lt;/p&gt;&lt;p&gt;&lt;span&gt;Machine learning models accurately processed camera videos, allowing efficient estimates of density from camera data.&lt;/span&gt;&lt;/p&gt;&lt;p&gt;&lt;span&gt;Implications&lt;/span&gt;&lt;/p&gt;&lt;p&gt;&lt;span&gt;Further adoption of motion-activated cameras coupled with automated processing could lead to continuous, large-scale monitoring of populations, helping to understand and address changes in biodiversity.&lt;/span&gt;&lt;/p&gt;</dc:description>
  <dc:format>application/pdf</dc:format>
  <dc:identifier>10.1071/PC25008</dc:identifier>
  <dc:language>en</dc:language>
  <dc:publisher>CSIRO Publishing</dc:publisher>
  <dc:title>Automated methods for processing camera trap video data for distance sampling</dc:title>
  <dc:type>article</dc:type>
</oai_dc:dc>