Real-Time Tiger Detection using YOLOv3

Based on the current population of tigers around the world and high tiger killing rate in forest adjacent area, there is a major need of automated visual surveillance to safeguard the tourists and civilians in forest adjacent area as well as decrease tiger killing rate due to the presence of tigers in residential areas as predators. The objective of this paper is to detect tigers in real-time, visually. The proposed method is using YOLOv3 algorithm and comparing the success rate with the template matching approach. A dataset of 1644 tiger images was collected with all possible angles and applied to train the YOLOv3 model. Then test dataset of 10 images was used to validate the results of YOLOv3. The detector performed exceptionally well to detect tiger in different images with different rotations, providing 80% accuracy. A real time environment can be ideal for using it at the full capacity.