Quantitative movement analysis using single-camera videos

In this project we provide tools for analyzing gait using a video footage from a single camera. Our software is based on machine learning models, in particular on convolutional neural networks. To train our machine learning models we used a dataset of over 3,000 videos processed with computer vision software finding body landmarks in frames (OpenPose). We provide the source code, dataset of clinical measurements as well as trajectories of body landmarks from OpenPose.Two key objectives of this SimTK project are:- providing a toolset for derviation of gait parameters from videos- providing a dataset allowing further research on gait analysis from videos using data-driven methods.Try out an online demo at: http://gaitlab.stanford.eduDownload the source code at: https://github.com/stanfordnmbl/mobile-gaitlab <br/><br/>This project includes the following software/data packages: <br/> <ul> <li> <a href="https://simtk.org/frs?group_id=1918#pack_2190">Dataset used in the paper </a> : This dataset contains:- trajectories of body keypoints extracted using OpenPose- corresponding labels derived from optical motion capture or annotated by physicians, including speed, cadence, GDI, knee flexion at max extension, surgerical decisions </li> </ul>

在本项目中，我们提供了一整套用于分析步态的工具体验，该工具基于单摄像头拍摄的视频素材。我们的软件基于机器学习模型，特别是卷积神经网络。为了训练我们的机器学习模型，我们使用了一个包含超过3,000个视频的数据集，这些视频经过计算机视觉软件处理，以识别帧中的身体关键点（OpenPose）。我们提供了源代码、临床测量数据集以及OpenPose中的身体关键点轨迹。SimTK项目的两大关键目标包括：提供一套从视频中推导步态参数的工具；提供一套数据集，允许使用数据驱动方法进一步研究步态分析。您可以通过访问以下链接尝试在线演示：http://gaitlab.stanford.edu，或在以下链接下载源代码：https://github.com/stanfordnmbl/mobile-gaitlab。本项目包含以下软件/数据包：

<br/> <ul> <li> <a href="https://simtk.org/frs?group_id=1918#pack_2190">论文中使用的数据集 </a>：该数据集包含使用OpenPose提取的身体关键点轨迹，以及从光学动作捕捉或由医生标注的相关标签，包括速度、步频、GDI、最大伸展时的膝屈曲角度，以及手术决策。
</ul>