Evaluating the accuracy of a Vision-Based Algorithm for Groundline Estimation in Trotting Horses Using Multiple Camera Angles

This dataset supports the study titled “Evaluating the Accuracy of a Vision-Based Algorithm for Groundline Estimation in Trotting Horses Using Multiple Camera Angles”. The data were collected as part of an experimental comparative study evaluating a deep learning-based computer vision algorithm for equine gait analysis. Recordings were made of eight Standardbred trotter mares trotting on a high-speed treadmill using seven iPhones placed at various fixed and handheld positions. Each video was processed using a trained neural network that detected 2D anatomical keypoints frame-by-frame. The dataset includes: Vertical Displacement Signal (VDS) values for eye, withers, and croup Reference groundline data (fixed and algorithm-estimated) Stride-level Maxdiff and Mindiff values This dataset enables the comparison of groundline estimation and stride symmetry metrics across different camera angles, including handheld usage. It may serve as a reference for future development and validation of markerless, portable gait analysis systems in veterinary medicine.本数据集可支撑题为《采用多相机视角评估快步马地面线估计算法准确率》的研究工作。本次数据采集自一项实验对比研究，该研究旨在评估一款基于深度学习的计算机视觉算法在马步态分析中的应用效果。 研究人员使用7台部署于不同固定机位与手持机位的iPhone，对8匹在高速跑台上快步奔跑的标准快步赛马（Standardbred trotter）母马的步态进行录制。每段视频均通过经过训练的神经网络逐帧检测二维解剖关键点。 本数据集涵盖以下内容： 1. 眼部、马肩隆（withers）与臀部（croup）的垂直位移信号（Vertical Displacement Signal, VDS）数值； 2. 参考地面线数据（包含固定参考地面线与算法估计地面线）； 3. 步幅级最大差值与最小差值数值。 本数据集可用于对比不同相机视角（含手持机位）下的地面线估计效果与步幅对称性指标，同时可作为未来兽医领域无标记便携式步态分析系统的开发与验证工作的参考依据。