Analysis and classification of dog behaviours using deep learning

Observation and recognition of animal faces, body postures, behaviors, and physical movements has recently become one of the most fundamental tasks of multidiscipline. The outstanding technological advances in areas such as computer vision and digital image processing can be used in the analysis of human behaviors on videos, as well as helping to detect animal behavior. In animals, it can contribute to the prediction of subsequent behaviors that may occur with computerized vision, and to domesticate animals. In the proposed thesis work, a system based on deep learning has been proposed to analyze and classify the behavior of dogs. A database was created by collecting videos showing the behavior of two different species dogs that do not avoid contact with people by using the wearable action camera and fixed camera. In the next stage, a deep learning-based application was conducted to analyze and classify the behavior of dogs. Determined behaviors such as mouth opening, tongue out, ears up, tail wagging, sniffing, playing, after the necessary examinations were made on the obtained videos, a special set of more meaningful sections was introduced. After feature extraction and obtaining meaningful sections from these videos, analysis of the behavior was performed by R-CNN. On the experimental studies, the behavior of the dog such as mouth opening, tongue out, sniffing, ears up, tail wagging, playing was examined and, training performance was obtained 100%, 99.99%, 99.28%, 99.99%, 95.99%, 99.64% for these behaviors, respectively. Keywords: Deep Learning; Convolutional Neural Networks (CNN); Computer Vision; Dog Behaviours; Animal Behavior Detection