Automatic detection of stereotypical behaviors of captive wild animals based on surveillance videos of zoos and animal reserves
Year of Publication:
|Zixuan Yin, Yaqin Zhao, Zhihao Xu, Qiuping Yu
|animal tracking, animal welfare, captive animal, Motion trajectory, Siamese network, stereotypical behavior
The timely detection of the depressive and stereotypical behaviors often observed in captive wild animals and the subsequent intervention can contribute to improving their living environment in enclosures, which is crucial for safeguarding animal welfare, enhancing animal husbandry practices, regulating human–animal relationships. Several studies have analyzed factors that influence animal stereotypical behaviors and identified preventive measures via regular animal observations. An automatic detection method based on video technology can yield long-term automatic recordings of motion trajectories of animals after a professionally trained automatic detection software is integrated into the human–machine interaction operation interface of animal management. As an initial exploration of this research paradigm, we propose a novel method for automatically tracking and recognizing the stereotypical behavior of animals in surveillance videos based on the periodic analysis of motion trajectories. First, we introduced a Siamese relation network to track the motion trajectories of animals. This network accurately tracked animals and distinguished different individuals in complex environments. Second, an autocorrelation function was used to analyze the periodicity of the motion trajectory, which was divided into several periodic curves. Finally, a cross-correlation function was introduced to determine the linear correlation between the two variables of the periodic curves. This function distinguished the three types of motion trajectories. The success rate and precision of the animal-tracking method adopted in this study were 67.4% and 90.4%, respectively, which were superior to those of common Siamese tracking networks. The average prediction error of the cycle time was 0.095 s. Therefore, the proposed method can accurately track the motion trajectories of animals and identify their stereotypical behaviors. Furthermore, this study provides data to facilitate the scientific management of animals and improve animal welfare. The codes and datasets used in the study are available at https://github.com/yinyinzixuan/animal-stereotypical-behavior.git.