This paper proposes a marker-based motion capture system for rat motion detection and gait analysis. Motion capture in small animals such as rodents is more challenging than in humans because of their small bodies and rapid motion. Existing algorithms have poor applicability in rat motion capture in environments outside the studio. Moreover, gait analysis targeting on the rat is not performed by existing motion capture software. Our method consists of four procedures. First, Region of Interest (ROI) is extracted from the background using the inter-frame difference method and a depth filter. Second, a double-threshold marker detection method is used to detect markers in ROI and a marker shape filter is used to classify the markers. Third, a marker corrector is designed to modify missing and incorrect markers. Finally, a deep learning network is used to analyse the gait trajectory to classify rat as healthy, injured, or rehabilitated. The experimental result shows that marker recognition accuracy is 99.33%, higher than that of most existing software. The validation accuracy of the network is 100% and the loss is 0.0001. This method is conductive to the development of motion capture systems for small animals and research into the gait kinematics of rodents.

本文提出了一种基于标记的运动捕捉系统，用于大鼠运动检测和步态分析。啮齿动物等小动物的动作捕捉比人类更具挑战性，因为它们身体小，动作快。现有算法在工作室外环境中的老鼠动作捕捉中的适用性较差。此外，现有的动作捕捉软件不执行针对大鼠的步态分析。我们的方法由四个程序组成。首先，使用帧间差异方法和深度过滤器从背景中提取感兴趣区域（ROI）。其次，使用双阈值标记检测方法检测ROI中的标记，并使用标记形状滤波器对标记进行分类。第三，标记校正器旨在修改丢失和不正确的标记。最后，深度学习网络用于分析步态轨迹以将大鼠分类为健康、受伤或康复。实验结果表明，标记识别准确率为99.33%，高于大多数现有软件。网络的验证准确率为 100%，损失为 0.0001。该方法有利于小动物运动捕捉系统的开发和啮齿动物步态运动学的研究。