多伦多康复中风姿势数据集

数据包括中风患者和健康参与者使用中风康复机器人执行一组任务的3D人体姿势估计。该项目的目标是使用机器人中风康复视频中的特征来检测可能导致练习无效的偏离姿势。此数据集仅供研究使用。内容中风通常会导致上肢运动障碍。为了适应新的限制，中风幸存者有时会改变运动模式，以使用更强壮或不受影响的关节和肌肉。然而，如果在康复训练中使用，这种补偿运动可能会导致无效的结果。一个能够自动检测补偿运动的系统将有助于指导中风幸存者使用正确的定位。为了开发这种自动化工具，我们提供了机器人康复训练期间临床相关运动的数据集。数据集由Microsoft Kinect传感器采集，包含两组参与者——10名健康者和9名中风幸存者——使用上肢康复机器人进行一系列坐姿动作。健康的参与者进行了额外的脚本动作，以模拟常见的中风后补偿动作。数据集还包括常见的临床评估分数。两位专家对健康和中风参与者的补偿运动进行了注释，并将其纳入数据集。我们还对数据集在检测选定任务的补偿运动方面的敏感性和特异性进行了初步评估。该数据集很有价值，因为它包括使用成本效益高、便携式和方便的传感器在临床环境中的临床相关运动。

This dataset contains 3D human pose estimates collected from stroke patients and healthy participants while they performed a series of tasks using a stroke rehabilitation robot. The objective of this project is to detect deviant postures that may lead to ineffective exercise by extracting features from videos of robot-assisted stroke rehabilitation training. This dataset is strictly for research use only.

Stroke often causes upper extremity motor impairments. To adapt to their new physical limitations, stroke survivors sometimes modify their movement patterns to utilize stronger or unaffected joints and muscles. However, such compensatory movements, when employed during rehabilitation training, may result in ineffective therapeutic outcomes. An automated system capable of detecting compensatory movements would help guide stroke survivors to adopt correct postures during training.

To develop such an automated tool, we provide a dataset of clinically relevant movements captured during robot-assisted rehabilitation sessions. The dataset was collected using a Microsoft Kinect sensor, and includes two groups of participants: 10 healthy individuals and 9 stroke survivors, who completed a series of seated upper extremity movements with an upper-limb rehabilitation robot. Healthy participants also performed additional scripted movements to simulate common post-stroke compensatory motions. The dataset also includes standard clinical assessment scores. Two experts annotated the compensatory movements of both healthy and stroke participants, and these annotations are included in the dataset. We also performed a preliminary evaluation of the dataset’s sensitivity and specificity for detecting compensatory movements in selected tasks.

This dataset is valuable as it includes clinically relevant movements captured using a cost-effective, portable, and user-friendly sensor in a clinical environment.

搜集汇总

背景与挑战

背景概述

多伦多康复中风姿势数据集包含中风患者和健康参与者使用康复机器人时的3D姿势数据，旨在通过Kinect传感器检测康复训练中的无效姿势。数据集还包括专家注释的补偿运动信息，支持中风康复研究。