Chimpanzee face recognition from videos in the wild using deep learning

Area of Science:

• Ethology
• Computer Vision
• Bioinformatics

Background:

• Manual analysis of animal behavior video data is time-consuming and resource-intensive.
• Large-scale video datasets offer significant potential for behavioral and conservation research.
• Automated tools are needed to efficiently process and analyze extensive video recordings.

Purpose of the Study:

• To develop a fully automated pipeline for face detection, tracking, and recognition of wild chimpanzees using deep convolutional neural networks (CNNs).
• To assess the accuracy of the automated system for chimpanzee identity and sex recognition.
• To demonstrate the utility of automated video analysis for studying social network dynamics.

Main Methods:

• A deep convolutional neural network (CNN) approach was implemented for automated video analysis.
• The pipeline performed face detection, tracking, and recognition on a 14-year video dataset of wild chimpanzees.
• Co-occurrence matrices were generated from identified faces to analyze social network structures.

Main Results:

• The automated system achieved 92.5% accuracy for identity recognition and 96.2% for sex recognition.
• The system processed 10 million face images from 23 individuals over 50 hours of footage.
• Analysis of social network structure changes in an aging chimpanzee population was enabled.

Conclusions:

• Deep learning offers an efficient and accurate method for analyzing large-scale animal behavior video data.
• Automated tools can unlock the potential of longitudinal video archives for behavioral ecology and conservation.
• The developed pipeline is adaptable for processing video data from various species.