Gait Phase Estimation by Using LSTM in IMU-Based Gait Analysis—Proof of Concept
Abstract
:1. Introduction
2. Materials and Methods
2.1. Data Analysis and Setup
2.2. Data Collection
2.2.1. Train Dataset
- Task 1: walk straight forward at a self-selected convenient pace for about 10 m (without running or jumping) and then stop, turn and return at the same speed. The task would be repeated for two more round-trips, one at a faster and one at a slower speed compared to the first round-trip. This would be considered as one trial. Each subject performed two trials.
- Task 2: turn 360 degrees while taking small steps without changing the place. This task was performed at a self-selected convenient pace and then repeated for two more tries, one at a faster and one at a slower speed compared to the first try. This would be considered as one trial. Each subject completed two trials.
- Task 3: the first task was repeated; however, the subject was asked to change the speed while walking. Each subject performed one trial.
- Task 4: stepping with both legs. This task was performed first at a self-selected convenient pace (two steps with each leg) without moving or changing the place. After waiting for two seconds the task would be repeated for two more tries, one at a faster and one at a slower pace compared to the first try. This would be considered as one trial. Each subject completed one trial.
- Task 5: the 4th task was repeated; however, the subject was asked to slightly step forwards and backwards while performing the task without changing the place and turning. Each subject performed one trial.
- Task 6: walk forward while turning. The subject was asked to stand at one side of the walking track just before the end of the track and then start turning 180 degrees while walking to the other side and then stop. After two seconds they returned at the same speed. After that, the task was repeated for two more round-trips, one at a faster and one at a slower pace. This would be considered as one trial. Only one subject completed this task for two trials.
2.2.2. Validation Dataset
- Task 1 and 2: each subject completed the same task for only one trial.
- Task 3: each subject performed this task for only one trial; however, the subjects were asked to complete only one round-trip.
- Task 4, 5 and 6: each subject completed the same task for only one trial.
2.2.3. Test Dataset
- Test_1_H and Test_1_G, including gait data from healthy individuals and glaucoma patients, respectively. Subjects walked forward at their self-selected convenient pace for about 13 m and then turned while walking to the left and walked back to the start position.
- Test_2_H, Test_2_G, including gait data from healthy individuals and glaucoma patients, respectively. Subjects walked first about 10 m straight forward at their self-selected convenient pace, then stopped and turned 180 degrees and returned at their self-selected slow pace, then stopped and turned 180 degrees and walked again 10 m forward at their self-selected fast pace and finally stopped and turned 180 degrees.
- Test_3_cLBP, including gait data from cLBP patients. Subjects walked straight forward over-ground at their self-selected convenient pace. The walking track was 15 m long and subjects were asked to walk about 30 s without stopping and continue walking while turning at each end of the track.
2.3. Data Processing
2.3.1. Data Cleaning and Signal Filtering
2.3.2. Feature Extraction
2.3.3. Feature Labeling
2.4. Neural Network Architecture
2.4.1. Background
2.4.2. Design
3. Results
3.1. Datasets
3.2. The Performance of Each Model
4. Discussion
Author Contributions
Funding
Institutional Review Board Statement
Informed Consent Statement
Conflicts of Interest
References
- Altun, K.; Barshan, B. Human Activity Recognition Using Inertial/Magnetic Sensor Units. In Human Behavior Understanding; Salah, A.A., Gevers, T., Sebe, N., Vinciarelli, A., Eds.; Springer: Berlin/Heidelberg, Germany, 2010; pp. 38–51. [Google Scholar]
- Jarchi, D.; Pope, J.; Lee, T.K.M.; Tamjidi, L.; Mirzaei, A.; Sanei, S. A Review on Accelerometry-Based Gait Analysis and Emerging Clinical Applications. IEEE Rev. Biomed. Eng. 2018, 11, 177–194. [Google Scholar] [CrossRef] [PubMed]
- Tao, W.; Liu, T.; Zheng, R.; Feng, H. Gait Analysis Using Wearable Sensors. Sensors 2012, 12, 2255–2283. [Google Scholar] [CrossRef] [PubMed]
- Gazit, E.; Buchman, A.S.; Dawe, R.; Curran, T.A.; Mirelman, A.; Giladi, N.; Hausdorff, J.M. What Happens before the First Step? A New Approach to Quantifying Gait Initiation Using a Wearable Sensor. Gait Posture 2020, 76, 128–135. [Google Scholar] [CrossRef] [PubMed]
- Edwards, W.B.; Derrick, T.R.; Hamill, J. Time Series Analysis in Biomechanics. In Handbook of Human Motion; Springer International Publishing: Cham, Switzerland, 2018; pp. 349–371. [Google Scholar] [CrossRef]
- Oudre, L.; Barrois-Müller, R.; Moreau, T.; Truong, C.; Vienne-Jumeau, A.; Ricard, D.; Vayatis, N.; Vidal, P.-P. Template-Based Step Detection with Inertial Measurement Units. Sensors 2018, 18, 4033. [Google Scholar] [CrossRef] [PubMed] [Green Version]
- Caldas, R.; Mundt, M.; Potthast, W.; Buarque de Lima Neto, F.; Markert, B. A Systematic Review of Gait Analysis Methods Based on Inertial Sensors and Adaptive Algorithms. Gait Posture 2017, 57, 204–210. [Google Scholar] [CrossRef] [PubMed]
- Xing, H.; Li, J.; Hou, B.; Zhang, Y.; Guo, M. Pedestrian Stride Length Estimation from IMU Measurements and ANN Based Algorithm. J. Sens. 2017, 2017, 1–10. [Google Scholar] [CrossRef] [Green Version]
- Aich, S.; Pradhan, P.M.; Chakraborty, S.; Kim, H.-C.; Kim, H.-T.; Lee, H.-G.; Kim, I.H.; Joo, M.; Jong-Seong, S.; Park, J. Design of a Machine Learning-Assisted Wearable Accelerometer-Based Automated System for Studying the Effect of Dopaminergic Medicine on Gait Characteristics of Parkinson’s Patients. J. Healthc. Eng. 2020, 2020, 1823268. [Google Scholar] [CrossRef] [PubMed]
- Giorgi, G.; Martinelli, F.; Saracino, A.; Sheikhalishahi, M. Walking Through the Deep: Gait Analysis for User Authentication through Deep Learning. In ICT Systems Security and Privacy Protection; Janczewski, L.J., Kutyłowski, M., Eds.; IFIP Advances in Information and Communication Technology; Springer International Publishing: Cham, Switzerland, 2018; Volume 529, pp. 62–76. [Google Scholar] [CrossRef]
- Khokhlova, M.; Migniot, C.; Morozov, A.; Sushkova, O.; Dipanda, A. Normal and Pathological Gait Classification LSTM Model. Artif. Intell. Med. 2019, 94, 54–66. [Google Scholar] [CrossRef] [PubMed]
- Santhiranayagam, B.; Lai, D.; Begg, R. Support Vector Machines for Young and Older Gait Classification Using Inertial Sensor Kinematics at Minimum Toe Clearance. In Proceedings of the 10th EAI International Conference on Body Area Networks, Sydney, Australia, 28–30 September 2015; ICST: Brussels, Belgium, 2015. [Google Scholar] [CrossRef] [Green Version]
- Paulich, M.; Schepers, M.; Rudigkeit, N.; Bellusci, G. Xsens MTw Awinda: Miniature Wireless Inertial-Magnetic Motion Tracker for Highly Accurate 3D Kinematic Applications; Xsens: Enschede, The Netherlands, 2018. [Google Scholar]
- Dot, T.; Quijoux, F.; Oudre, L.; Vienne-Jumeau, A.; Moreau, A.; Vidal, P.-P.; Ricard, D. Non-Linear Template-Based Approach for the Study of Locomotion. Sensors 2020, 20, 1939. [Google Scholar] [CrossRef] [PubMed]
- Aminian, K.; Najafi, B.; Büla, C.; Leyvraz, P.-F.; Robert, P. Spatio-Temporal Parameters of Gait Measured by an Ambulatory System Using Miniature Gyroscopes. J. Biomech. 2002, 35, 689–699. [Google Scholar] [CrossRef]
- Gouwanda, D.; Senanayake, S.M.N.A. Application of Hybrid Multi-Resolution Wavelet Decomposition Method in Detecting Human Walking Gait Events. In Proceedings of the 2009 International Conference of Soft Computing and Pattern Recognition, Washington, DC, USA, 4–7 December 2009; IEEE: Malacca, Malaysia, 2009; pp. 580–585. [Google Scholar] [CrossRef]
- Hochreiter, S.; Schmidhuber, J. Long short-term memory. Neural Comput. 1997, 9, 735–1780. [Google Scholar] [CrossRef] [PubMed]
- Lempereur, M.; Rousseau, F.; Rémy-Néris, O.; Pons, C.; Houx, L.; Quellec, G.; Brochard, S. A New Deep Learning-Based Method for the Detection of Gait Events in Children with Gait Disorders: Proof-of-Concept and Concurrent Validity. J. Biomech. 2020, 98, 109490. [Google Scholar] [CrossRef] [PubMed]
- Gao, J.; Gu, P.; Ren, Q.; Zhang, J.; Song, X. Abnormal Gait Recognition Algorithm Based on LSTM-CNN Fusion Network. IEEE Access 2019, 7, 163180–163190. [Google Scholar] [CrossRef]
- Kreuzer, D.; Munz, M. Deep Convolutional and LSTM Networks on Multi-Channel Time Series Data for Gait Phase Recognition. Sensors 2021, 21, 789. [Google Scholar] [CrossRef] [PubMed]
Dataset | No. of Subjects | No. of Trials | Age (Mean ± std.) | BMI 1 (Mean ± std.) | No. of Sample Frames | No. of Labeled Frames for Each Gait Phase 2 |
---|---|---|---|---|---|---|
Healthy individuals | ||||||
| 2 | 32 | 29.5 ± 4.9 | 23.8 ± 1.6 | 197,556 | 379 |
Healthy individuals | ||||||
| 2 | 24 | 22.5 ± 0.7 | 21.5 ± 1 | 76,262 | 174 |
Healthy individuals | ||||||
| 5 | 10 | 71 ± 7.2 | 27.6 ± 3.9 | 27,438 | 172 |
| 11 | 22 | 71.4 ± 4.9 | 25.2 ± 2.4 | 118,422 | 514 |
Glaucoma patients | ||||||
| 4 | 8 | 69.8 ± 4.1 | 26.7 ± 2.6 | 21,058 | 133 |
| 7 | 14 | 71.1 ± 5.3 | 25.4 ± 3.6 | 86,050 | 341 |
cLBP patients | ||||||
| 37 | 74 | 68.9 ± 8.4 | 28 ± 4.6 | 260,054 | 1690 |
Datasets | LSTM_FO Accuracy | LSTM_FO Loss | LSTM_MidS Accuracy | LSTM_MidS Loss | LSTM_FC Accuracy | LSTM_FC Loss |
Train | 0.9981 | 0.0071 | 0.9981 | 0.0080 | 0.9981 | 0.0072 |
Validation | 0.9977 | 0.0098 | 0.9977 | 0.0102 | 0.9977 | 0.0098 |
Datasets | LSTM_FO Accuracy | LSTM_FO Loss | LSTM_MidS Accuracy | LSTM_MidS Loss | LSTM_FC Accuracy | LSTM_FC Loss |
Test_1_H | 0.9937 | 0.0241 | 0.9937 | 0.0280 | 0.9937 | 0.0296 |
Test_1_G | 0.9936 | 0.0322 | 0.9936 | 0.0303 | 0.9936 | 0.0395 |
Test_2_H | 0.9957 | 0.0213 | 0.9957 | 0.0223 | 0.9957 | 0.0216 |
Test_2_G | 0.9960 | 0.0186 | 0.9960 | 0.0192 | 0.9960 | 0.0199 |
Test_3_cLBP | 0.9935 | 0.0284 | 0.9935 | 0.0309 | 0.9935 | 0.0343 |
Average | 0.9945 | 0.0249 | 0.9945 | 0.0261 | 0.9945 | 0.0290 |
Publisher’s Note: MDPI stays neutral with regard to jurisdictional claims in published maps and institutional affiliations. |
© 2021 by the authors. Licensee MDPI, Basel, Switzerland. This article is an open access article distributed under the terms and conditions of the Creative Commons Attribution (CC BY) license (https://creativecommons.org/licenses/by/4.0/).
Share and Cite
Sarshar, M.; Polturi, S.; Schega, L. Gait Phase Estimation by Using LSTM in IMU-Based Gait Analysis—Proof of Concept. Sensors 2021, 21, 5749. https://doi.org/10.3390/s21175749
Sarshar M, Polturi S, Schega L. Gait Phase Estimation by Using LSTM in IMU-Based Gait Analysis—Proof of Concept. Sensors. 2021; 21(17):5749. https://doi.org/10.3390/s21175749
Chicago/Turabian StyleSarshar, Mustafa, Sasanka Polturi, and Lutz Schega. 2021. "Gait Phase Estimation by Using LSTM in IMU-Based Gait Analysis—Proof of Concept" Sensors 21, no. 17: 5749. https://doi.org/10.3390/s21175749
APA StyleSarshar, M., Polturi, S., & Schega, L. (2021). Gait Phase Estimation by Using LSTM in IMU-Based Gait Analysis—Proof of Concept. Sensors, 21(17), 5749. https://doi.org/10.3390/s21175749