Related Experiment Video
Updated: Dec 11, 2025
07:35
Selecting Multiple Biomarker Subsets with Similarly Effective Binary Classification Performances
Published on: October 11, 2018
7.9K
Artificial Neural Networks in Motion Analysis-Applications of Unsupervised and Heuristic Feature Selection Techniques
Marion Mundt1, Arnd Koeppe1, Franz Bamer1
1Institute of General Mechanics, RWTH Aachen University, 52062 Aachen, Germany.
Sensors (Basel, Switzerland)
|August 23, 2020
Summary
Machine learning with long short-term memory neural networks can estimate 3D lower limb joint angles using only three inertial sensors. This simplifies in-field motion analysis for future real-time applications.
Area of Science:
- Biomechanics
- Machine Learning
- Wearable Technology
Background:
- In-field motion analysis using inertial sensors is crucial for understanding human movement.
- Reducing the number of sensors simplifies data collection and processing for motion analysis.
- Estimating 3D lower limb joint angles accurately with minimal sensors remains a challenge.
Purpose of the Study:
- To investigate the efficacy of machine learning, specifically long short-term memory (LSTM) neural networks, in estimating 3D lower limb joint angles from a minimal number of inertial sensors.
- To determine the optimal sensor placement and quantity for accurate motion analysis.
- To explore the impact of data dimensionality reduction techniques like Principal Component Analysis (PCA) on prediction accuracy.
Main Methods:
- Training an LSTM neural network to predict 3D lower limb joint angles using data from inertial sensors.
- Evaluating different sensor configurations, focusing on a minimal set including pelvis and shanks.
- Assessing the utility of Principal Component Analysis (PCA) for simplifying data dimensions.
- Comparing prediction accuracy using longer motion sequences versus time-normalized gait cycles.
Main Results:
- Three inertial sensors, placed on the pelvis and both shanks, were found to be sufficient for estimating 3D lower limb joint angles.
- The application of PCA to data from five sensors did not yield improved prediction results.
- Utilizing longer motion sequences demonstrated an advantage in prediction accuracy compared to time-normalized gait cycles.
Conclusions:
- A minimal sensor setup (three inertial sensors) combined with LSTM networks can effectively estimate 3D lower limb joint angles for in-field motion analysis.
- Longer motion sequences improve prediction accuracy, paving the way for real-time applications.
- Further research can refine these methods for enhanced biomechanical assessments.
