基於深度學習的異質性感測器上肢建模用於中風復健

Upper Extremity Modeling For Stroke Rehabilitation With Heterogeneous Sensors Based On Machine Learning Algorithms

徐浩翔(Hao-Hsiang Hsu)

指導教授 ： 周承復

國立臺灣大學/電機資訊學院/資訊工程學研究所/碩士(2020年)

https://doi.org/10.6342/NTU202003573

摘要

在中風復健系統和應用中，應同時考慮可靠性、準確性和遮擋造成的影響。然而，大多數現有的方法主要關注於解決可靠性及準確性的問題。由於遮擋也是影響醫療判斷的重要因素，為了同時解決這三個重要問題，我們提出了一種異質感測器融合框架由RGB-D相機和可穿戴設備組成，用於遮擋時為復健病患提供可靠的關節位置。為了在補償遮擋時融合多個傳感器的測量結果，我們應用了異質感測同時定位,跟蹤和建模以估算關節和傳感器，並用機器學習的方式構建上肢模型 。基於此模型的虛擬測量用於估計遮擋期間關節的位置。使用模擬生成 與收集來自十位受測者資料的應用結果顯示出，提出的演算法可以得到有遮蔽時誤差 3.9 公分的成果，並且隨著遮擋時間的增加還是能使誤差降至5.9公分。由此本篇論文成功展示了在室內環境下，使用機器學習解決遮蔽造成的問題與影響的能力。

關鍵字

機器學習 ； 神經網路

並列摘要

In stroke rehabilitation systems and applications,The impact of reliability, accuracy and occlusion should be considered at the same time. However, most existing methods mainly focus on solving reliability and accuracy problems. Since occlusion is also an important factor affecting medical judgment, in order to solve these three important problems at the same time, we propose a heterogeneous sensor fusion framework composed of RGB-D camera and wearable device, it is used to provide reliable joint position for rehabilitation patients when covering. In order to fuse the measurement results of multiple sensors when compensating for occlusion, we applied heterogeneous sensing to simultaneously locate, track and model to estimate joints and sensors, and use machine learning algorithms to build upper limb models.Virtual measurement based on this model used to estimate the position of joints during occlusion. The application results of using simulation to generate and collect data from ten subjects show that the proposed algorithm can obtain an error of 3.9 cm with occlusion, and with the increase of occlusion time, the error can still be reduced to 5.9 cm. As a result, this paper successfully demonstrated the ability to use machine learning to solve the problems and effects of shading in an indoor environment.

並列關鍵字

Machine Learning ； Neural Network

參考文獻

Y. Bengio. Learning deep architectures for AI. Now Publishers Inc, 2009.

D. Carroll. A quantitative test of upper extremity function. Journal of chronic diseases, 18(5):479–491, 1965.

C. Chang, S. Wang, and C. Wang. Exploiting moving objects: Multi-robot simultaneous localization and tracking. IEEE Transactions on Automation Science andEngineering, 13(2):810–827, 2016.

J. Chen, C. Wang, E. H. Wu, and C. Chou. Simultaneous heterogeneous sensor localization, joint tracking, and upper extremity modeling for stroke rehabilitation.IEEE Systems Journal, pages 1–12, 2020.

J. Chung, C. Gulcehre, K. Cho, and Y. Bengio. Empirical evaluation of gated recurrent neural networks on sequence modeling. arXiv preprint arXiv:1412.3555, 2014.