High Accuracy Non-Optical Motion Capture

测量物体在3D维度上的运动，包括人体的运动，在许多应用中是一项重要的任务。目前几乎所有的方法，包括申请人的NSERC资助的研究，都使用光学方法。这些依赖于相机来测量运动，因为高质量的相机无处不在，易于使用，并且相对便宜。然而，它们存在视线遮挡的问题，并且在测量人体运动的典型条件下具有较低的精度（误差>> 1 mm）。 为了验证基于相机的研究系统的准确性，并使用机器学习技术对其进行训练，我们需要比我们的系统更准确的地面实况数据，并且不会受到同样的限制。我们将获得两个互补的仪器，具有高精度（精度<<1 mm），不会受到视觉遮挡。 一种仪器使用磁场跟踪3D运动，另一种仪器使用连接到跟踪对象的机械电枢。没有一个测量系统是完美的-通过获得两个互补的系统，我们可以证明我们的研究系统在各种条件下的准确性和鲁棒性。 这些仪器将被申请人和计算机图形学、计算机视觉、人机界面和生物力学领域的其他研究人员用于开发更好的运动捕捉系统，包括使用相机和其他新型传感模式的系统。具体而言，它们将用于严格测试我们的研究系统，并定量评估这些系统的准确性。这些仪器还将用于创建地面真实运动数据库，该数据库可用于校准基于摄像头的运动捕捉系统，并生成用于训练深度神经网络的数据，以跟踪视频中的运动。 这些系统使加拿大受益，因为它们为一系列应用提供了低成本、高质量的动作捕捉系统，包括病人监护系统、自然计算机界面、虚拟现实和增强现实耳机、计算机游戏和视觉效果。

Measuring the motion of objects in 3D dimensions, including the motion of the human body, is an essential task in many applications. Almost all current methods, including the NSERC funded research of the applicants, use optical methods. These rely on cameras for measuring motion, since high quality cameras are ubiquitous, easy to use, and relatively inexpensive. However, they suffer from problems with occlusion of line of sight and have lower accuracy under the typical conditions of measuring human motion (errors >> 1mm). To validate the accuracy of camera-based research systems, and to train them using machine learning techniques, we need ground truth data that are more accurate than our systems and do not suffer from the same limitations. We will acquire two complementary instruments that have high accuracy (accuracy << 1mm) and do not suffer from visual occlusion. One instrument tracks 3D motion using magnetic fields, and another using a mechanical armature connected to the tracked object. No measurement system is perfect - by acquiring two complementary systems we can demonstrate the accuracy and robustness of our research systems under a wide range of conditions. The instruments will be used by the applicants and other researchers in the fields of computer graphics, computer vision, human-computer interfaces, and biomechanics, to develop better motion capture systems, including those using cameras and other novel sensing modalities. Specifically, they will be used to rigorously test our research systems and to quantitatively assess the accuracy of these systems. The instruments will also be used to create a database of ground truth motion that could be used for calibrating camera-based motion capture systems and to generate data for training deep neural networks to track motion from video. Such systems benefit Canada by making low cost, high quality motion capture systems for a range of applications including patient monitoring systems, natural computer interfaces, virtual reality and augmented reality headsets, computer games, and visual effects.