Identifying Small-Displacement Stereotypical Behavior of Captive Slow Lorises Based on AnimalYOLO-Bytetrack Network

Wild animals rescued and kept in captivity often display abnormal behaviors, notably in the form of stereotypic behaviors. These behaviors, characterized by purposeless and repetitive actions, can significantly affect an animal's quality of life and socialization processes. Existing methods for identifying stereotypic behaviors predominantly focus on individual animals or on behaviors involving large displacements with extensive movements and distinct trajectories. However, research addressing scenarios involving multiple animals or small-displacement stereotypic behaviors remains limited. This paper aims to address this gap by analyzing small-displacement stereotypic behaviors in multiple animals, specifically focusing on captive slow lorises. As a preliminary exploration of this research area, we propose a novel method for analyzing small-displacement stereotypic behavior utilizing automatic tracking and periodic changes in movement amplitude. We employ the AnimalYOLO-Bytetrack multi-target tracking network to monitor and quantify the movement amplitude of the slow loris, segment the time-series data to identify the specific periods of stereotypic behaviors, and analyze periodicity using autocorrelation functions and Fourier transforms. The method developed in this study demonstrates significant efficacy, achieving a detection precision of 96.7%, a recall rate of 80.3%, and a mean Average Precision (mAP) value of 88.9%. These results surpass the performance metrics of most existing detection networks. Furthermore, the method exhibits an average prediction error of 0.33 for cycle time and 0.76 s for the duration of stereotypic behavior, highlighting its effectiveness in predicting the stereotypical behavior of Lorisidae primates.