Gait trajectory prediction via extreme learning machine
In this thesis, an algorithm to estimate the gait trajectory based upon the electromyography (EMG) signal is proposed. The algorithm is developed using Extreme Learning Machine (ELM). Experiments were conducted to acquire the gait parameters from 20 healthy human subjects. EMG signals from Tibialis...
| Main Author: | |
|---|---|
| Format: | Thesis (University of Nottingham only) |
| Language: | English |
| Published: |
2021
|
| Subjects: | |
| Online Access: | https://eprints.nottingham.ac.uk/63929/ |
| _version_ | 1848800076443418624 |
|---|---|
| author | Lim, Han Leong |
| author_facet | Lim, Han Leong |
| author_sort | Lim, Han Leong |
| building | Nottingham Research Data Repository |
| collection | Online Access |
| description | In this thesis, an algorithm to estimate the gait trajectory based upon the electromyography (EMG) signal is proposed. The algorithm is developed using Extreme Learning Machine (ELM). Experiments were conducted to acquire the gait parameters from 20 healthy human subjects. EMG signals from Tibialis Anterior (TA) and Gastrocnemius Lateral (GL) muscles were obtained during the gait cycle. Three stages were completed to ensure the viability of EMG signals and ELM used to predict the gait trajectory. The first stage comprises of using EMG signals to predict temporal gait parameters and benchmarked with a basic same architecture Artificial Neural Network (ANN). The target temporal gait parameters are gait speed and stance/swing phase which were measured using an inertia sensor and camera system. The ELM algorithm was developed using a single hidden layer feedforward network architecture where the weights from the input layer to the hidden layer were randomized and not updated during the run. Results obtained from ELM were compared with an ANN model with the same architecture as the ELM algorithm. In ELM, the mean estimation errors of gait speed, stance percentage, and swing percentage were 11.86%, 7.62%, and 6.07% respectively. This was compared to the errors of 12.92%, 11.75%, and 9.56% using ANN. Besides that, ELM achieved shorter training and testing time. The robustness of the ELM algorithm demonstrated the capability of real-time computation due to superior computing performance compared to conventional ANN models. The second stage superseded the first with a more advanced variant of ELM to classify the endpoints of EMG for each gait cycle during the whole gait duration. This was done with a Weighted ELM, and for 20 participants, it was able to achieve an average testing classification rate of 99.38%. In the last part, a boosted Ensemble ELM was proposed to predict the x trajectory of the gait using features of EMG signals. The best configuration of 8 ELMs boosted is shown to produce an average RMSE of 0.1055m when compared to an un-boosted version of 0.1344m. |
| first_indexed | 2025-11-14T20:45:49Z |
| format | Thesis (University of Nottingham only) |
| id | nottingham-63929 |
| institution | University of Nottingham Malaysia Campus |
| institution_category | Local University |
| language | English |
| last_indexed | 2025-11-14T20:45:49Z |
| publishDate | 2021 |
| recordtype | eprints |
| repository_type | Digital Repository |
| spelling | nottingham-639292025-02-28T15:08:15Z https://eprints.nottingham.ac.uk/63929/ Gait trajectory prediction via extreme learning machine Lim, Han Leong In this thesis, an algorithm to estimate the gait trajectory based upon the electromyography (EMG) signal is proposed. The algorithm is developed using Extreme Learning Machine (ELM). Experiments were conducted to acquire the gait parameters from 20 healthy human subjects. EMG signals from Tibialis Anterior (TA) and Gastrocnemius Lateral (GL) muscles were obtained during the gait cycle. Three stages were completed to ensure the viability of EMG signals and ELM used to predict the gait trajectory. The first stage comprises of using EMG signals to predict temporal gait parameters and benchmarked with a basic same architecture Artificial Neural Network (ANN). The target temporal gait parameters are gait speed and stance/swing phase which were measured using an inertia sensor and camera system. The ELM algorithm was developed using a single hidden layer feedforward network architecture where the weights from the input layer to the hidden layer were randomized and not updated during the run. Results obtained from ELM were compared with an ANN model with the same architecture as the ELM algorithm. In ELM, the mean estimation errors of gait speed, stance percentage, and swing percentage were 11.86%, 7.62%, and 6.07% respectively. This was compared to the errors of 12.92%, 11.75%, and 9.56% using ANN. Besides that, ELM achieved shorter training and testing time. The robustness of the ELM algorithm demonstrated the capability of real-time computation due to superior computing performance compared to conventional ANN models. The second stage superseded the first with a more advanced variant of ELM to classify the endpoints of EMG for each gait cycle during the whole gait duration. This was done with a Weighted ELM, and for 20 participants, it was able to achieve an average testing classification rate of 99.38%. In the last part, a boosted Ensemble ELM was proposed to predict the x trajectory of the gait using features of EMG signals. The best configuration of 8 ELMs boosted is shown to produce an average RMSE of 0.1055m when compared to an un-boosted version of 0.1344m. 2021-02-24 Thesis (University of Nottingham only) NonPeerReviewed application/pdf en arr https://eprints.nottingham.ac.uk/63929/1/LIM%20HAN%20LEONG%20-%20THESIS.pdf Lim, Han Leong (2021) Gait trajectory prediction via extreme learning machine. MPhil thesis, University of Nottingham. gait trajectory electromyography signal extreme learning machine architecture artificial neural network machine learning |
| spellingShingle | gait trajectory electromyography signal extreme learning machine architecture artificial neural network machine learning Lim, Han Leong Gait trajectory prediction via extreme learning machine |
| title | Gait trajectory prediction via extreme learning machine |
| title_full | Gait trajectory prediction via extreme learning machine |
| title_fullStr | Gait trajectory prediction via extreme learning machine |
| title_full_unstemmed | Gait trajectory prediction via extreme learning machine |
| title_short | Gait trajectory prediction via extreme learning machine |
| title_sort | gait trajectory prediction via extreme learning machine |
| topic | gait trajectory electromyography signal extreme learning machine architecture artificial neural network machine learning |
| url | https://eprints.nottingham.ac.uk/63929/ |