I-Corps: Embedded Machine Vision for Accurate Gait Analyses and Body Movement Measurements

这个i-Corps项目的更广泛的影响/商业潜力是通过最大限度地发挥无处不在的平板电脑和2D摄像头的能力，用基于机器视觉的低成本解决方案来彻底改变传统的康复助手和身体运动评估工具，这些设备与更昂贵的3D摄像头相比具有显著的成本和可及性优势。基于嵌入式机器视觉技术，可以开发一套平板电脑/智能手机应用程序，团队计划使用订阅模式创建相关服务。这一创新将惠及广阔的医疗市场，包括康复设施、外科/神经科门诊诊所、理疗诊所、中风中心和个人。总体而言，这项创新将使嵌入式机器视觉技术能够接触到并服务于社会上的许多群体。这个i-Corps项目是基于我们在PC平台上获得的初步结果，以接近实时地识别和跟踪人体关节点。该原型利用了新的机器学习算法和高性能处理器(图形处理单元)。我们下一阶段的研究是从2D摄像头生成深度图，识别和跟踪身体关节点，并对身体动作进行评分，所有这些都在平板电脑平台上进行。这些挑战包括利用2D图像生成深度，以及在资源受限的嵌入式平台上执行高效的模型和算法。该团队将利用嵌入式深度学习技术，针对现代平板电脑上的嵌入式处理器架构优化算法，以便概念验证能够提供令人满意的性能。该团队还计划与一名医生和一名康复设施顾问合作，审查和改进软件功能并验证早期原型。该奖项反映了NSF的法定使命，并通过使用基金会的智力优势和更广泛的影响审查标准进行评估，被认为值得支持。

The broader impact/commercial potential of this I-Corps project is to revolutionize conventional rehabilitation assistant and body movement assessment tools with low-cost machine vision based solutions by maximizing the capability of ubiquitous tablets and 2D cameras, which have significant cost and accessibility advantage over more expensive 3D cameras. A set of tablet/smartphone Apps can be developed base on the embedded machine vision technology, and the team plan to create related services using a subscription model. This innovation will benefit broad healthcare market, which includes rehab facilities, surgery/neurology outpatient clinics, physical therapy clinics, stroke centers, and individuals. Overall, the innovation will make embedded machine vision technology accessible to and serve many population groups in the society. This I-Corps project is based on preliminary results that we obtained on a PC platform to identify and track human body joint points in close to real-time. The prototype leverages novel machine learning algorithms and high performance processors (graphics processing units). Our next phase of research is to generate depth map from a 2D camera, recognize and track body joint points, and score body movements, all on a tablet platform. The challenges include depth generation with 2D images and efficient models and algorithms to execute on a resource constraint embedded platform. The team will leverage embedded deep learning techniques to optimize the algorithms towards an embedded processor architecture on modern tablets so that the proof-of-concept can deliver satisfactory performance. The team also plan to work with a medical doctor and a rehabilitation facility advisor to review and refine the software features and validate the early prototype.This award reflects NSF's statutory mission and has been deemed worthy of support through evaluation using the Foundation's intellectual merit and broader impacts review criteria.