scholarly journals Predicting continuous ground reaction forces from accelerometers during uphill and downhill running: a recurrent neural network solution

PeerJ ◽  
2022 ◽  
Vol 10 ◽  
pp. e12752
Author(s):  
Ryan S. Alcantara ◽  
W. Brent Edwards ◽  
Guillaume Y. Millet ◽  
Alena M. Grabowski
Keyword(s):  
Neural Network ◽  
Neural Networks ◽  
Recurrent Neural Network ◽  
Human Movement ◽  
The Body ◽  
Ground Reaction Forces ◽  
Accelerometer Data ◽  
Laboratory Environment ◽  
Reaction Forces ◽  
External Forces

Background Ground reaction forces (GRFs) are important for understanding human movement, but their measurement is generally limited to a laboratory environment. Previous studies have used neural networks to predict GRF waveforms during running from wearable device data, but these predictions are limited to the stance phase of level-ground running. A method of predicting the normal (perpendicular to running surface) GRF waveform using wearable devices across a range of running speeds and slopes could allow researchers and clinicians to predict kinetic and kinematic variables outside the laboratory environment. Purpose We sought to develop a recurrent neural network capable of predicting continuous normal (perpendicular to surface) GRFs across a range of running speeds and slopes from accelerometer data. Methods Nineteen subjects ran on a force-measuring treadmill at five slopes (0°, ±5°, ±10°) and three speeds (2.5, 3.33, 4.17 m/s) per slope with sacral- and shoe-mounted accelerometers. We then trained a recurrent neural network to predict normal GRF waveforms frame-by-frame. The predicted versus measured GRF waveforms had an average ± SD RMSE of 0.16 ± 0.04 BW and relative RMSE of 6.4 ± 1.5% across all conditions and subjects. Results The recurrent neural network predicted continuous normal GRF waveforms across a range of running speeds and slopes with greater accuracy than neural networks implemented in previous studies. This approach may facilitate predictions of biomechanical variables outside the laboratory in near real-time and improves the accuracy of quantifying and monitoring external forces experienced by the body when running.

scholarly journals Predicting continuous ground reaction forces from accelerometers during uphill and downhill running: A recurrent neural network solution

2021 ◽  
Author(s):  
Ryan S. Alcantara ◽  
W. Brent Edwards ◽  
Guillaume Y. Millet ◽  
Alena M. Grabowski
Keyword(s):  
Neural Network ◽  
Recurrent Neural Network ◽  
Stance Phase ◽  
Accelerometer Data ◽  
Step Frequency ◽  
Laboratory Environment ◽  
Downhill Running ◽  
The Neural Network ◽  
Reaction Forces ◽  
Active Peak

AbstractBackgroundGround reaction forces (GRFs) are important for understanding the biomechanics of human movement but the measurement of GRFs is generally limited to a laboratory environment. Wearable devices like accelerometers have been used to measure biomechanical variables outside the laboratory environment, but they cannot directly measure GRFs. Previous studies have used neural networks to predict the entire GRF waveform during the stance phase from wearable device data, but these networks require normalization of GRFs to the duration of a step or stance phase, resulting in a loss of the GRF waveform’s temporal component. Additionally, previous studies have predicted GRF waveforms during level-ground, but not uphill or downhill running. A method of predicting the normal (perpendicular to running surface) GRF waveform using wearable devices across a range of running speeds and slopes, while maintaining the GRF waveform’s temporal component, could allow researchers and clinicians to predict kinetic and kinematic variables outside the laboratory environment.PurposeWe sought to develop a recurrent neural network capable of predicting normal GRF waveforms across a range of running speeds and slopes using data from accelerometers located on the sacrum and shoe.Methods19 subjects completed 30-s running trials on a force-measuring treadmill at five slopes (0°, ±5°, ±10°) and three speeds (2.5, 3.33, and 4.17 m/s) per slope. One biaxial accelerometer was adhered to the sacrum and two uniaxial accelerometers were adhered to the right shoe during all trials. Accelerometers on the shoe were used to classify foot strike patterns as rearfoot, midfoot, or forefoot, and sacral acceleration data were divided into overlapping 12-ms windows, allowing the neural network to iteratively predict the experimentally-measured normal GRF waveform frame-by-frame. The mean, SD, and range of accelerometer data for each 12-ms window were included as neural network input features, along with the subject’s body mass, height, running speed, slope, and percentage of a trial’s steps classified as a rearfoot, midfoot, or forefoot strike. We assessed the accuracy and generalizability of the neural network using leave-one-subject-out cross validation, which provided an ensemble of Root Mean Square Error (RMSE) and relative RMSE (rRMSE) values comparing the normal GRF waveform predicted by the neural network to the normal GRF waveform measured by the force-measuring treadmill. Additionally, we calculated the mean absolute percent error (MAPE) of step frequency, contact time, normal impulse, normal GRF active peak, and loading rate between the predicted and measured GRF waveforms.ResultsThe average ± SD RMSE was 0.16 ± 0.04 BW and rRMSE was 6.4 ± 1.5% for neural network predictions of each subject’s normal GRF waveform compared to measured GRF waveforms across all conditions. RMSE values were lower during slow uphill running (2.5 m/s, +10°; 0.13 ± 0.07 BW) compared to fast downhill running (4.17 m/s, −10°; 0.20 ± 0.05 BW). The MAPE ± SD for step frequency was 0.1 ± 0.1%, contact time was 4.9 ± 4.0%, normal impulse was 6.4 ± 6.9%, normal GRF active peak was 8.5 ± 8.2%, and loading rate was 27.6 ± 36.1%.ConclusionsWe developed a recurrent neural network that uses accelerometer data to predict the continuous normal GRF waveform across a range of running speeds and slopes. The neural network does not require preliminary identification of the stance phase, maintains the temporal component of the GRF waveform, can be applied to up- and downhill running, and facilitates the prediction of kinetic and kinematic variables outside the laboratory environment. This represents a substantial step towards accurately quantifying and monitoring the external loads experienced by the body when running outdoors.

Gait Analysis for Muscular Forces Evaluation in Human Movement: Integration Protocol of Typical Measurement Methods

2018 ◽  
Author(s):  
Luca Fontanili ◽  
Massimo Milani ◽  
Luca Montorsi ◽  
Giordano Valente
Keyword(s):  
Gait Analysis ◽  
Athletic Training ◽  
Motion Capture ◽  
Vertical Component ◽  
Human Movement ◽  
The Body ◽  
Body Motion ◽  
Ground Reaction Forces ◽  
Muscle Forces ◽  
Reaction Forces

The paper focuses on the gait analysis for the investigation of the typical events occurring in human movements and validate its use as a method for musculoskeletal disease evaluation and for the improvement of athletic training. In the present research the motion capture system is combined with an in-house developed prototype of uniaxial force plates for the measurement of the vertical component of ground reaction forces during movement. While similar techniques are implemented for gait, this equipment can be employed to investigate running, thus, covering a larger number of possible applications and providing a deeper insight either of the athlete performance or the disease analysis. For the prevention and the treatment of those events occurring during running, a thorough understanding of its mechanisms is critical; therefore, a method for evaluating both the kinematic behavior of the human body and the ground reaction forces combined to a model for determining the muscle forces is proposed. An infrared motion capture technique is adopted for measuring accurately the body motion and a multiple force-plate system is used to calculate the force exerted by the ground and sub-divided in the three components by an ad-hoc developed routine. Moreover, the data are used as input parameters for the OpenSim software to derive muscles forces. Finally, the potential of the proposed protocol is determined by an experimental campaign on healthy subjects and a significant database of muscle forces is constructed for different running speeds.

Deadlift Recognition and Application based on Multiple Modalities using Recurrent Neural Network

2020 ◽  
Vol 2020 (17) ◽  
pp. 2-1-2-6
Author(s):  
Shih-Wei Sun ◽  
Ting-Chen Mou ◽  
Pao-Chi Chang
Keyword(s):  
Neural Network ◽  
Recurrent Neural Network ◽  
Network Structure ◽  
Inertial Sensors ◽  
Body Movement ◽  
The Body ◽  
Body Parts ◽  
Multimodal Sensing ◽  
Multiple Devices ◽  
Neural Network Structure

To improve the workout efficiency and to provide the body movement suggestions to users in a “smart gym” environment, we propose to use a depth camera for capturing a user’s body parts and mount multiple inertial sensors on the body parts of a user to generate deadlift behavior models generated by a recurrent neural network structure. The contribution of this paper is trifold: 1) The multimodal sensing signals obtained from multiple devices are fused for generating the deadlift behavior classifiers, 2) the recurrent neural network structure can analyze the information from the synchronized skeletal and inertial sensing data, and 3) a Vaplab dataset is generated for evaluating the deadlift behaviors recognizing capability in the proposed method.

scholarly journals Recurrent Neural Network for Human Activity Recognition in Embedded Systems Using PPG and Accelerometer Data

Electronics ◽  
2021 ◽  
Vol 10 (14) ◽  
pp. 1715
Author(s):  
Michele Alessandrini ◽  
Giorgio Biagetti ◽  
Paolo Crippa ◽  
Laura Falaschetti ◽  
Claudio Turchetti
Keyword(s):  
Neural Network ◽  
Embedded System ◽  
Activity Recognition ◽  
Human Activity ◽  
Recurrent Neural Network ◽  
Low Cost ◽  
Low Complexity ◽  
Human Activity Recognition ◽  
Motion Artifacts ◽  
Accelerometer Data

Photoplethysmography (PPG) is a common and practical technique to detect human activity and other physiological parameters and is commonly implemented in wearable devices. However, the PPG signal is often severely corrupted by motion artifacts. The aim of this paper is to address the human activity recognition (HAR) task directly on the device, implementing a recurrent neural network (RNN) in a low cost, low power microcontroller, ensuring the required performance in terms of accuracy and low complexity. To reach this goal, (i) we first develop an RNN, which integrates PPG and tri-axial accelerometer data, where these data can be used to compensate motion artifacts in PPG in order to accurately detect human activity; (ii) then, we port the RNN to an embedded device, Cloud-JAM L4, based on an STM32 microcontroller, optimizing it to maintain an accuracy of over 95% while requiring modest computational power and memory resources. The experimental results show that such a system can be effectively implemented on a constrained-resource system, allowing the design of a fully autonomous wearable embedded system for human activity recognition and logging.

EMOTIONS RECOGNITION IN HUMAN SPEECH USING DEEP NEURAL NETWORKS

2021 ◽  
pp. 44-51
Author(s):  
E. Yu. Shchetinin
Keyword(s):  
Neural Network ◽  
Machine Learning ◽  
Neural Networks ◽  
Convolutional Neural Network ◽  
Recurrent Neural Network ◽  
Deep Neural Networks ◽  
Learning Algorithms ◽  
Machine Learning Algorithms ◽  
Audio Recordings ◽  
Computer Studies

The recognition of human emotions is one of the most relevant and dynamically developing areas of modern speech technologies, and the recognition of emotions in speech (RER) is the most demanded part of them. In this paper, we propose a computer model of emotion recognition based on an ensemble of bidirectional recurrent neural network with LSTM memory cell and deep convolutional neural network ResNet18. In this paper, computer studies of the RAVDESS database containing emotional speech of a person are carried out. RAVDESS-a data set containing 7356 files. Entries contain the following emotions: 0 – neutral, 1 – calm, 2 – happiness, 3 – sadness, 4 – anger, 5 – fear, 6 – disgust, 7 – surprise. In total, the database contains 16 classes (8 emotions divided into male and female) for a total of 1440 samples (speech only). To train machine learning algorithms and deep neural networks to recognize emotions, existing audio recordings must be pre-processed in such a way as to extract the main characteristic features of certain emotions. This was done using Mel-frequency cepstral coefficients, chroma coefficients, as well as the characteristics of the frequency spectrum of audio recordings. In this paper, computer studies of various models of neural networks for emotion recognition are carried out on the example of the data described above. In addition, machine learning algorithms were used for comparative analysis. Thus, the following models were trained during the experiments: logistic regression (LR), classifier based on the support vector machine (SVM), decision tree (DT), random forest (RF), gradient boosting over trees – XGBoost, convolutional neural network CNN, recurrent neural network RNN (ResNet18), as well as an ensemble of convolutional and recurrent networks Stacked CNN-RNN. The results show that neural networks showed much higher accuracy in recognizing and classifying emotions than the machine learning algorithms used. Of the three neural network models presented, the CNN + BLSTM ensemble showed higher accuracy.

scholarly journals Prospects for recurrent neural network models to learn RNA biophysics from high-throughput data

2017 ◽  
Author(s):  
Michelle J Wu ◽  
Johan OL Andreasson ◽  
Wipapat Kladwang ◽  
William J Greenleaf ◽  
Rhiju Das ◽  
...  
Keyword(s):  
Neural Network ◽  
Neural Networks ◽  
Free Energy ◽  
Recurrent Neural Network ◽  
Neural Network Model ◽  
Data Augmentation ◽  
Rna Folding ◽  
Large Datasets ◽  
Turner Model ◽  
Rna Complexes

AbstractRNA is a functionally versatile molecule that plays key roles in genetic regulation and in emerging technologies to control biological processes. Computational models of RNA secondary structure are well-developed but often fall short in making quantitative predictions of the behavior of multi-RNA complexes. Recently, large datasets characterizing hundreds of thousands of individual RNA complexes have emerged as rich sources of information about RNA energetics. Meanwhile, advances in machine learning have enabled the training of complex neural networks from large datasets. Here, we assess whether a recurrent neural network model, Ribonet, can learn from high-throughput binding data, using simulation and experimental studies to test model accuracy but also determine if they learned meaningful information about the biophysics of RNA folding. We began by evaluating the model on energetic values predicted by the Turner model to assess whether the neural network could learn a representation that recovered known biophysical principles. First, we trained Ribonet to predict the simulated free energy of an RNA in complex with multiple input RNAs. Our model accurately predicts free energies of new sequences but also shows evidence of having learned base pairing information, as assessed by in silico double mutant analysis. Next, we extended this model to predict the simulated affinity between an arbitrary RNA sequence and a reporter RNA. While these more indirect measurements precluded the learning of basic principles of RNA biophysics, the resulting model achieved sub-kcal/mol accuracy and enabled design of simple RNA input responsive riboswitches with high activation ratios predicted by the Turner model from which the training data were generated. Finally, we compiled and trained on an experimental dataset comprising over 600,000 experimental affinity measurements published on the Eterna open laboratory. Though our tests revealed that the model likely did not learn a physically realistic representation of RNA interactions, it nevertheless achieved good performance of 0.76 kcal/mol on test sets with the application of transfer learning and novel sequence-specific data augmentation strategies. These results suggest that recurrent neural network architectures, despite being naïve to the physics of RNA folding, have the potential to capture complex biophysical information. However, more diverse datasets, ideally involving more direct free energy measurements, may be necessary to train de novo predictive models that are consistent with the fundamentals of RNA biophysics.Author SummaryThe precise design of RNA interactions is essential to gaining greater control over RNA-based biotechnology tools, including designer riboswitches and CRISPR-Cas9 gene editing. However, the classic model for energetics governing these interactions fails to quantitatively predict the behavior of RNA molecules. We developed a recurrent neural network model, Ribonet, to quantitatively predict these values from sequence alone. Using simulated data, we show that this model is able to learn simple base pairing rules, despite having no a priori knowledge about RNA folding encoded in the network architecture. This model also enables design of new switching RNAs that are predicted to be effective by the “ground truth” simulated model. We applied transfer learning to retrain Ribonet using hundreds of thousands of RNA-RNA affinity measurements and demonstrate simple data augmentation techniques that improve model performance. At the same time, data diversity currently available set limits on Ribonet’s accuracy. Recurrent neural networks are a promising tool for modeling nucleic acid biophysics and may enable design of complex RNAs for novel applications.

scholarly journals Ground Reaction Forces during Vertical Hops Are Correlated with the Number of Supervised Physiotherapy Visits after Achilles Tendon Surgery

2021 ◽  
Vol 10 (22) ◽  
pp. 5299
Author(s):  
Łukasz Sikorski ◽  
Andrzej Czamara
Keyword(s):  
Achilles Tendon ◽  
Study Groups ◽  
The Body ◽  
Ground Reaction Forces ◽  
Group I ◽  
Tendon Surgery ◽  
Reaction Forces ◽  
Number Of Visits ◽  
Group Ii ◽  
The Right

The objective of this study was to assess the effectiveness of, and the correlation between, an average of 42 supervised physiotherapy (SVPh) visits for the vertical ground reaction forces component (vGRF) using ankle hops during two- and one-legged vertical hops (TLH and OLH, respectively), six months after the surgical suturing of the Achilles tendon using the open method (SSATOM) via Keesler’s technique. Hypothesis: Six months of supervised physiotherapy with a higher number of visits (SPHNVs) was positively correlated with higher vGRF values during TLH and OLH. Group I comprised male patients (n = 23) after SSATOM (SVPh x = 42 visits), and Group II comprised males (n = 23) without Achilles tendon injuries. In the study groups, vGRF was measured during TLH and OLH in the landing phase using two force plates. The vGRF was normalized to the body mass. The limb symmetry index (LSI) of vGRF values was calculated. The ranges of motion of the foot and circumferences of the ankle joint and shin were measured. Then, 10 m unassisted walking, the Thompson test, and pain were assessed. A parametric test for dependent and independent samples, ANOVA and Tukey’s test for between-group comparisons, and linear Pearson’s correlation coefficient calculations were performed. Group I revealed significantly lower vGRF values during TLH and OLH for the operated limb and LSI values compared with the right and left legs in Group II (p ≤ 0.001). A larger number of visits correlates with higher vGRF values for the operated limb during TLH (r = 0.503; p = 0.014) and OLH (r = 0.505; p = 0.014). An average of 42 SVPh visits in 6 months was insufficient to obtain similar values of relative vGRF and their LSI during TLH and OLH, but the hypothesis was confirmed that SPHNVs correlate with higher relative vGRF values during TLH and OLH in the landing phase.

scholarly journals Machine Learning-Based Estimation of Ground Reaction Forces and Knee Joint Kinetics from Inertial Sensors While Performing a Vertical Drop Jump

Sensors ◽  
2021 ◽  
Vol 21 (22) ◽  
pp. 7709
Author(s):  
Serena Cerfoglio ◽  
Manuela Galli ◽  
Marco Tarabini ◽  
Filippo Bertozzi ◽  
Chiarella Sforza ◽  
...  
Keyword(s):  
Neural Network ◽  
Machine Learning ◽  
Knee Joint ◽  
Field Analysis ◽  
Ground Reaction Forces ◽  
Drop Jump ◽  
Joint Moments ◽  
Data Set ◽  
Joint Kinetics ◽  
Reaction Forces

Nowadays, the use of wearable inertial-based systems together with machine learning methods opens new pathways to assess athletes’ performance. In this paper, we developed a neural network-based approach for the estimation of the Ground Reaction Forces (GRFs) and the three-dimensional knee joint moments during the first landing phase of the Vertical Drop Jump. Data were simultaneously recorded from three commercial inertial units and an optoelectronic system during the execution of 112 jumps performed by 11 healthy participants. Data were processed and sorted to obtain a time-matched dataset, and a non-linear autoregressive with external input neural network was implemented in Matlab. The network was trained through a train-test split technique, and performance was evaluated in terms of Root Mean Square Error (RMSE). The network was able to estimate the time course of GRFs and joint moments with a mean RMSE of 0.02 N/kg and 0.04 N·m/kg, respectively. Despite the comparatively restricted data set and slight boundary errors, the results supported the use of the developed method to estimate joint kinetics, opening a new perspective for the development of an in-field analysis method.

Correlation between Ground Reaction Force and Tibial Acceleration in Vertical Jumping

2007 ◽  
Vol 23 (3) ◽  
pp. 180-189 ◽  
Author(s):  
Niell G. Elvin ◽  
Alex A. Elvin ◽  
Steven P. Arnoczky
Keyword(s):  
Ground Reaction Force ◽  
Reaction Force ◽  
Force Plate ◽  
Coefficient Of Determination ◽  
Ground Reaction Forces ◽  
Accelerometer Data ◽  
Jump Height ◽  
Vertical Jumping ◽  
Average Coefficient ◽  
Reaction Forces

Modern electronics allow for the unobtrusive measurement of accelerations outside the laboratory using wireless sensor nodes. The ability to accurately measure joint accelerations under unrestricted conditions, and to correlate them with jump height and landing force, could provide important data to better understand joint mechanics subject to real-life conditions. This study investigates the correlation between peak vertical ground reaction forces, as measured by a force plate, and tibial axial accelerations during free vertical jumping. The jump heights calculated from force-plate data and accelerometer measurements are also compared. For six male subjects participating in this study, the average coefficient of determination between peak ground reaction force and peak tibial axial acceleration is found to be 0.81. The coefficient of determination between jump height calculated using force plate and accelerometer data is 0.88. Data show that the landing forces could be as high as 8 body weights of the jumper. The measured peak tibial accelerations ranged up to 42 g. Jump heights calculated from force plate and accelerometer sensors data differed by less than 2.5 cm. It is found that both impact accelerations and landing forces are only weakly correlated with jump height (the average coefficient of determination is 0.12). This study shows that unobtrusive accelerometers can be used to determine the ground reaction forces experienced in a jump landing. Whereas the device also permitted an accurate determination of jump height, there was no correlation between peak ground reaction force and jump height.

Sign in / Sign up

Export Citation Format

Share Document