CHS: Medium: Data Driven Biomechanically Accurate Modeling of Human Gait on Unconstrained Terrain

在人体工程学、动画、生物力学、康复、理疗、虚拟现实和娱乐等领域，对人体步态进行参数、高效、准确的建模是一个开放而富有挑战性的问题。步态是一个复杂的过程，因为人类的许多关节自由度必须同时协调和适应不同类型的地形、步态类型以及相关的运动学和动力学。因此，在不受约束的复杂地形上，没有通用的步态运动学模型，甚至可以构成开发此类模型的基础的基本运动学步态数据库和相关的地面力也不能提供准确的信息。大多数步态数据是由常用的表面标记系统采集的，这些系统存在软组织伪影，不足以完全理解步态运动学。开发一种适用于无约束地形的人体步态参数，将在虚拟现实、下一代鞋子设计、计算机动画、工作场所安全、人体工程学、运动医学、生物医学和临床研究等领域产生广泛的影响，以帮助步态异常的人。项目成果，包括算法和数据集，最终将被纳入包括计算机科学和生物医学工程在内的许多领域的课程，并将产生一个更知情的步态建模及其应用人员社区。利用最近购买的新型全地形步态跑步机、运动捕捉、摄像机、基于力板的地面反作用力以及膝盖和脚踝关节的高速超声波图像，Pi和他的团队收集了多个人的步态数据。该项目的第一个目标是通过利用新的计算机视觉和多模式优化算法来整合所有这些数据，以产生具有人体特定关节、人体形状、地面反作用力和运动学数据的解剖学正确的人体骨骼。第二个目标是开发一个高效、准确和通用的人体步态运动学和生物力学的参数化模型，能够再现数据，更重要的是预测和推广适用于新类型地形的人类特定步态。要模拟的步态变化范围从非常缓慢的步态到在各种地形条件下的正常行走，例如倾斜和倾斜的坡度、左右交叉坡度和上下楼梯。

Modeling human gait parametrically, efficiently and accurately is an open and challenging problem with many applications such as ergonomics, animation, biomechanics, rehabilitation, physical therapy, virtual reality and entertainment. Gait is a complex process, because numerous human joint degrees-of-freedom have to be simultaneously coordinated and adapted to varying types of terrain, types of gait and related kinematics and dynamics. As a consequence, no general-purpose models of gait kinematics on unconstrained complex terrain exist, and even basic kinematic gait databases and related ground forces that can form the basis for developing such models do not provide accurate information. Most gait data are collected by the commonly used surface marker systems, which suffer from soft tissue artifacts and are not sufficient for a full understanding of gait kinematics. Developing a parameterized human gait for unconstrained terrain will have broad impact on the fields of virtual reality, next generation shoe design, computer animation, workplace safety, ergonomics, sports medicine, biomedical and clinical research to aid people with gait abnormalities. Project outcomes, including algorithms and datasets, will ultimately be incorporated in the curricula of many fields including computer science and biomedical engineering, and will result in a better informed community of people working on gait modeling and its applications.Using recently purchased novel all-terrain gait treadmills, motion capture, video cameras, ground reaction forces based on force plates, and high speed ultrasonic images of knee and ankle joints, the PI and his team have collected gait data from multiple people. The first goal of this project is to integrate all of this data by exploiting novel computer vision and multimodal optimization algorithms to produce an anatomically correct human skeleton with human specific joints, human shape, ground reaction forces and kinematic data. The second goal is to develop an efficient, accurate, and general purpose parameterized model of human gait kinematics and biomechanics capable of reproducing the data, and more importantly, predicting and generalizing human specific gait for new types of terrain. The gait variations to be modeled range from very slow shuffling to normal walking on a variety of terrain conditions such as inclined and declined slope, left and right cross-slope, and up and down stairs.