Processing citizen science- and machine-annotated time-lapse imagery for biologically meaningful metrics

Area of Science:

• Ecology
• Computer Vision
• Remote Sensing

Background:

• Time-lapse cameras offer high-resolution wildlife monitoring but generate large datasets requiring processing.
• Automated analysis of ecological imagery is crucial for efficient data extraction and interpretation.

Purpose of the Study:

• To publish processing pipelines for raw time-lapse imagery of animal and plant communities.
• To generate count data and nearest neighbor distance measurements for ecological monitoring.
• To compare citizen science, machine learning, and expert-derived counts.

Main Methods:

• Developed image processing pipelines for time-lapse data.
• Utilized the Penguin Watch citizen science project for image annotation.
• Implemented a computer vision algorithm (Pengbot) for automated image analysis.
• Calculated nearest neighbor distances to assess colony spatial structure and movement.

Main Results:

• Generated count data (number of penguins per image) and nearest neighbor distance metrics.
• Provided example files from 14 Penguin Watch cameras, processed from 63,070 images.
• Compared the accuracy of citizen science, machine learning, and expert counts.
• Demonstrated the utility of nearest neighbor distances for monitoring phenological stage and detecting movement.

Conclusions:

• Published open-source processing methodologies and datasets for ecological studies.
• Encouraged the use of these resources for advancing ecological research and machine learning development.
• Highlighted the value of automated image analysis for remote wildlife monitoring.