A Novel Method to Improve Individual Animal Identification Based on Camera-Trapping Data

摘要

We present a novel method to improve individual identification of animals based on camera-trapping data. The method combines computer tools and human visual recognition to help multiple users to reach identification agreement. Application of this method to a bobcat (Lynx rufus) picture database from the Jasper Ridge Biological Preserve resulted in a progressive increase in identification agreement between 2 users, as measured by the adjusted Rand index (ARI). An initial ARI value of 0.28 increased to a final value of 0.84 (1 maximum agreement). In contrast, comparisons involving random picture groupings consistently rendered low ARI values (<= 0.05). The numbers of individuals named by the 2 users decreased from initial values of 46 and 43 to final values of 25 and 29, respectively. The tool presented here will help researchers and wildlife managers to identify individual mammals and monitor populations