Design of Rehabilitation System Based on Research of Motion Classification and Virtual Reality Technology
Li, Qingqing (2018-08-13)
Design of Rehabilitation System Based on Research of Motion Classification and Virtual Reality Technology
Li, Qingqing
(13.08.2018)
Tätä artikkelia/julkaisua ei ole tallennettu UTUPubiin. Julkaisun tiedoissa voi kuitenkin olla linkki toisaalle tallennettuun artikkeliin / julkaisuun.
Turun yliopisto
Tiivistelmä
Surface electromyography (sEMG) is a composite effect of superficial muscle fibers and
nerves on the outer surface of the skin. Using skin surface EMG (sEMG) to predict
gesture is one of the important subjects in related research area, it can be widely applied
in the field of prosthetic control, rehabilitation medicine and so on.
In this thesis, the EMG signal data was collected by theMYO armband, and the signal was
wavelet-transformed to reduce the noise and then divided by moving average methods to
extract the signal data of each action. Through widely-validated phonogram techniques
in phonetics, each gesture signal data is converted into a spectrogram which contains timing
frequency distribution. The convolutional neural network(CNN) use spectrogram to
learn and generate a learning Model. This thesis also tries to use the open source Tensorflow
framework to build a neural network to train the raw data and learn a model. Due
to the Tensorflow’s operation efficiency and its good adaptability in multiple platforms,
Tensorflow can generate a model with good portability. At last, a mirror therapy(MT) VR
rehabilitation system was developed based on MYO armband, HTC virtual reality helmet
and the trained Tensorflow model. The original signal is collected from the arm of the
patient through the MYO armband, and then sent signal to the Python processing program
for feature extraction. The extracted features are processed to the trained model and
the model make gesture prediction. The HTC helmet is used to display the corresponding
gesture animation of the prediction results. VR helmets offers an immersive experience by
creating a good mirroring environment which helps generate a better treatment outcome.
In this thesis, a signal segmentation method which can remove redundancy data is introduced.
After the model was generated, the thesis compares the differences of the two
models. According to the test results, the accuracy of the spectrogram-based method is
higher with higher hardware requirements and lower process speed. However, the method
based on feature vector is faster with lower accuracy.
nerves on the outer surface of the skin. Using skin surface EMG (sEMG) to predict
gesture is one of the important subjects in related research area, it can be widely applied
in the field of prosthetic control, rehabilitation medicine and so on.
In this thesis, the EMG signal data was collected by theMYO armband, and the signal was
wavelet-transformed to reduce the noise and then divided by moving average methods to
extract the signal data of each action. Through widely-validated phonogram techniques
in phonetics, each gesture signal data is converted into a spectrogram which contains timing
frequency distribution. The convolutional neural network(CNN) use spectrogram to
learn and generate a learning Model. This thesis also tries to use the open source Tensorflow
framework to build a neural network to train the raw data and learn a model. Due
to the Tensorflow’s operation efficiency and its good adaptability in multiple platforms,
Tensorflow can generate a model with good portability. At last, a mirror therapy(MT) VR
rehabilitation system was developed based on MYO armband, HTC virtual reality helmet
and the trained Tensorflow model. The original signal is collected from the arm of the
patient through the MYO armband, and then sent signal to the Python processing program
for feature extraction. The extracted features are processed to the trained model and
the model make gesture prediction. The HTC helmet is used to display the corresponding
gesture animation of the prediction results. VR helmets offers an immersive experience by
creating a good mirroring environment which helps generate a better treatment outcome.
In this thesis, a signal segmentation method which can remove redundancy data is introduced.
After the model was generated, the thesis compares the differences of the two
models. According to the test results, the accuracy of the spectrogram-based method is
higher with higher hardware requirements and lower process speed. However, the method
based on feature vector is faster with lower accuracy.