Behavioral analysis of animals is an essential part of neuroscience research. However, most previous methods are based on manual recognition or use image recognition technology to analyze animal trajectories. The former is time-consuming and subjective, while the latter ignores the large amount of body posture information of animals in three-dimensional space. Based on the existing three-dimensional behavioral analysis framework inspired by the natural structure of animal behaviors, this study designs and builds a three-dimensional spatial posture acquisition device applicable to multiple scenarios, such as visual innate fear behavior, and optimizes some parts of the data processing and analysis process. The results show that the system is able to accurately identify the five major types of mouse movements, and can restore the dynamic process of defensive behaviors of mice under the visually life threaten with high accuracy. This study provides an efficient, objective, and quantifiable solution for the comprehensive analysis of visuallyevoked innate fear behavior. Furthermore, combined with other cutting-edge methods, this system may provide to more finely and deeply data for the neural circuit me chanism of behaviors.