Automated Detection and Counting of Wild Boar in Camera Trap Images

Area of Science:

• Ecology
• Computer Science
• Wildlife Management

Background:

• Camera traps are essential for wildlife monitoring but manual image analysis is inefficient.
• Automated analysis is needed to process large datasets from camera traps effectively.

Purpose of the Study:

• To evaluate the efficacy of computer vision techniques for automated wildlife image analysis.
• To develop a model for automatic species detection and activity quantification from camera trap data.

Main Methods:

• A computer vision model was trained on 1600 images of animals near wild boar carcasses.
• The model was used to automatically detect and classify five animal species: wild boar, fox, raccoon dog, deer, and bird.
• Image sequences were analyzed to determine wild boar visit frequency and group sizes.

Main Results:

• The automated system achieved a mean average precision of 98.11% for species detection.
• The model successfully identified 'wild boar', 'fox', 'raccoon dog', 'deer', and 'bird' in natural environments.
• Automated analysis quantified wild boar visits and group sizes from image sequences.

Conclusions:

• Computer vision offers a highly accurate and efficient solution for analyzing wildlife camera trap data.
• This automated approach can significantly accelerate wildlife monitoring and disease surveillance, particularly for African swine fever.
• The technology has the potential to improve wildlife management strategies through rapid data extraction.