Video-based gait analysis of 3D silhouettes

在加拿大，肌肉骨骼疾病和紊乱是就诊的主要原因之一，在疾病、长期残疾和经济方面对社会产生重大影响。随着加拿大人口的老龄化，这些问题在未来将迅速增加。因此，需要新的生物医学技术来识别和诊断早期的显著异常，以便更有效地治疗和更快地恢复患者。因此，步态分析已经显示出许多证据，证明其在临床医学实践中的潜力。在这种情况下，该研究计划的总体长期愿景是从重建的3D身体轮廓的动态中识别相关参数，以进行活动识别。特别是，我们将在未来几年专注于诊所和家庭的3D步态分析。不同的创新多摄像机的方法和技术将被调查的实时三维重建。然后，正常主体的3D轮廓运动将使用机器学习在低维空间中用参数的统计分布来描述。对于患者，在该低维空间中识别与正常的显著差异将导致“步态质量”指数以评估行走问题。病理性步态也将以类似方式表示，用于比较和分类目的。在所有情况下，计算机生成的动画人类将被添加到训练数据集中，以代表人类的大范围人体测量和步态特性，并改善学习。一个关键的步骤是验证的三维重建和步态评估的方法。3D数据精度将通过已知尺寸的3D几何对象和人体模型进行测量。步态分析将首先用计算机生成的动画人物和通过视觉船体提取重建的多个3D行走人物的数据库进行测试。随后，将在实验室中与学生志愿者进行正常和模拟步态障碍的进一步实验。最后，将在大学医院招募患者进行真实的临床评估。还计划与高端专业设备和黄金标准功能测试进行比较，以确定我们的系统准确量化功能性流动性能的能力。除了病理步态分析之外，我们预计医疗保健中的其他应用（例如平衡评估、身体屈曲测量（弯曲测试）、跌倒和其他事故检测、患者监测、运动学/运动评估等）和其他领域（例如生物统计、活动识别、监控、游戏、3D人机界面、机器人等）将从这项研究中受益 **

Musculoskeletal diseases and disorders are one of the leading causes of physician visits in Canada and have a major impact on society in terms of illness, long-term disability and economics. As the Canadian population is aging, these problems will rapidly increase in the future. New biomedical technologies to identify and diagnose early significant abnormalities for more efficient treatment and a faster recovery of the patient are thus needed. For that matter, gait analysis has shown a lot of evidences demonstrating its potential for clinical medical practice. In this context, the general long-term vision of this program of research is to identify relevant parameters from the dynamics of reconstructed 3D body silhouettes for activity recognition. In particular, we will focus during the next years on 3D gait analysis in the clinic and at home.***Different innovative multi-camera approaches and technologies will be investigated for the real-time 3D reconstruction. Then the 3D silhouette motion of normal subjects will be described with a statistical distribution of parameters in a lower-dimensional space using machine learning. For a patient, the identification of significant differences from normality in this low-dimensional space will lead to a "gait quality" index to assess ambulatory problems. Pathological gaits will also be represented similarly for comparison and classification purposes. In all cases, computer-generated animated humans will be added to the training dataset to represent the large range of anthropometric and gait properties of humans and improve learning.***A crucial step is the validation of the approach for both the 3D reconstruction and gait assessment. 3D data precision will be measured with 3D geometric object and human mannequins of known dimensions. Gait analysis will be tested first with computer-generated animated humans and a database of multiple 3D walking humans reconstructed by visual hull extraction. Later on, further experiments will be conducted with student volunteers performing normal and simulated gait impairments in laboratory. Finally, patients will be recruited at University Hospitals for a real clinical assessment. Comparisons with high-end specialized equipment and gold standard functional tests are also planned to determine the ability of our systems to accurately quantify functional ambulatory performance.***In addition to pathological gait analysis, we expect that other applications in healthcare (e.g. balance assessment, body flexion measurements (bending tests), fall and other accident detections, patient monitoring, evaluation in kinesiology/sport etc.) and other areas (e.g. biometry, activity recognition, surveillance, gaming, 3D human-computer interface, robotics etc.) will benefit from this research.**