Wild Animal Recognition Method for Videos Using a Combination of Deep Learning and Motion Detection

Abstract

This paper proposes a new wildlife image recognition method that combines object recognition based on deep learning and motion detection by inter-frame difference. This method makes it possible that such animal images as difficult to detect by object recognition can be detected by motion detection, and vice versa. The experimental results show the detection ratio and hit ratio are 100% and 84%, respectively, for videos including animals. Our achieved results can provide more efficient solutions for the time consuming and costly mitigation approaches to reduce human-wildlife conflicts.