Revolutionizing Wildlife Monitoring: A Novel Approach to Camera Trap Image Analysis with YOLOv5

The use of camera traps to estimate the size of animals is gaining traction in ecology due to its applicability in population conservation and management. In wildlife conservation there are ongoing discussions regarding standards for AI-based monitoring tools and techniques. This paper utilizes the methods YOLOv5 and Regions-Convolutional Neural Networks (R-CNN) to improve the detection of multiple-object overlaps in wildlife and reduce image redundancy. In the case of R-CNN training and using these models can be computationally intensive, requiring significant resources. This can be a limitation in resource-constrained environments or for real-time applications. The model utilized the PASCAL VOC dataset with an expected input image size of 234×234 pixels, while the original images were 3457 × 3457 pixels. To address potential issues with changes in size smaller images from everyday real-life scenarios were chosen for training and testing. The experimental results demonstrate high-level performance on the PASCAL VOC dataset, achieving an accuracy of 99.38%, compared to surpassing other existing models such as Deep Neural Network and Regions-Convolutional Neural Network.