| 研究課題/領域番号 |
21K18105
|
|---|
| 研究機関 | 九州大学 |
|---|
研究代表者 |
|
|---|
| 研究期間 (年度) |
2021-04-01 – 2023-03-31
|
| キーワード | Motion Intention / HD EMG |
|---|
| 研究実績の概要 |
In the first stage of the research we have classified the human motion intention for gesture recognition of human hand as a preliminary study with high density EMG signals. In this work we used an already existing data set with data from 192 channels of surface EMG, measured from the human forearm. The data was preprocessed with band pass filtering and, the filtered data was used to derive power of the HDEMG signals. Resulting data was used to train a long short terms memory (LSTM) network to predict different gestures of human hand based on the HD EMG recorded from human forearm. The results showed HDEMG signals can be used successfully to predict human motion intention with deep learning techniques based approaches. It was also observed that not all the used channels for the prediction did not contain motion related information and hence highlighted the requirement of using a channel selection method in future steps. The results were presented in the 39th annual conference of the Robotic Society of Japan (RSJ2021).
|
| 現在までの達成度 (区分) |
現在までの達成度 (区分)
2: おおむね順調に進展している
理由
In order to collect data for analysis, we have purchased a 64CH HDEMG system from OT bioelecttronica, Italy. It allows different configuration of HD EMG measurement with different electrode grids. At the moment we are trying to understand the muscle activation of human forearm when performing different types of finger motions, especially different grasping patterns in activities of daily living. We are collecting HD EMG data from the different muscle locations using the purchased HD EMG system. After preprocecssing, blind source separation techniques are used to identify different locations of EMG signal generation related to different muscle activties for finger motions.
|
| 今後の研究の推進方策 |
As a result of the above blind source separation techniques, it will allow us to understand the prominent locations of EMG signals generation. In the next stage of the study, these results will be used to create a customized layout of electrodes. The customized electrode layout will be focused to meassure muscle activation only from the most relevant areas of the human muscles.These measured muscle signals from customized ellectrode grid will be used with different signals processing and machine learning techniques to effectively predict the human motion intention. If the time permits, we plan to implement these techniques to control a robotic hand to generate similar motion of the human motion intention in real-time.
|
| 次年度使用額が生じた理由 |
There were plans to present the result in internatinal conferences. However, becuase of the ongoing pandemic these plans were postponed to the next fiscal year. Therefore the remaining amount of the grant will be used to present the resuts of the current study in an international conference.
|
