| Summary: | 碩士 === 國立中興大學 === 資訊科學與工程學系所 === 107 === In recent years, more and more biologists and natural scientists have set up camera traps in the wild to collect wild animal images in large volume. In the way of using camera-trap to collect wildlife images, people do not need to wait for a long time in a fixed place until animals appear. However, in order to analyze these images and videos collected by camera-trap, and if this work is processed manually, it may require a lot of time and money. In addition, the speed of collecting wildlife images and videos is so fast that the biologists can’t effectively use the collected data.
With the development of deep learning in computer vision in recent years, in this paper we propose a deep learning and image processing techniques to detect and identify animals in the wild, aiming to establish an automatic detection system for wild animals. In the wild, animals are often covered by some grass or branches, which makes it impossible to identify with the deep learning method. Therefore, in the wildlife recognition system, we combined the traditional image processing method. We use the hierarchical median filter to build the background model, and background subtraction technique to obtain animals which are covered by grass or branches. After getting the animals, we will use the VGG16 and darknet53 classifiers to classify them.
The images and videos used in this study were provided by the professor Chen of the Department of Forest, National Chung Hsing University. Experimental results show that the animals covered by some grass or branches can still be identified successfully. It demonstrates the effectiveness of the proposed method.
|
|---|
