Andreas Spilz
Current Organizations / Institutes
The organizations/institutes from the scholar's latest article
Andreas Spilz has a total of 4 co-authors. Here are the top co-authors by number of shared publications:
- Michael Munz
AI for Sensor Data Analytics Research Group, Ulm University of Applied Sciences, 89081 Ulm, Germany
8 shared publications - Heiko Oppel
AI for Sensor Data Analytics Research Group, Ulm University of Applied Sciences, 89081 Ulm, Germany
3 shared publications - Marius Karge1 shared publication
Thomas Engleder
University of Applied Sciences Ulm, Ulm , Germany
1 shared publication
Latest Publications
- Open Access
Boosting Automatic Exercise Evaluation Through Musculoskeletal Simulation-Based Augmentation of IMU-Derived Orientation Data
Automated evaluation of movement quality can enhance physiotherapeutic treatment and sports training by providing objective, real-time feedback. However, deep learning models that assess movements captured by inertial measurement units (IMUs) are often limited by data scarcity, class imbalance, and label ambiguity. We present a data augmentation method for IMU-derived orientation data that generates additional examples by systematically modifying movement trajectories and passing them through a musculoskeletal simulation. The approach enforces the joint-range limits of a musculoskeletal model and enables automatic labeling by combining inverse kinematic parameters with a knowledge-based evaluation strategy. Across four datasets of varying complexity, augmented variants closely resemble real-world data and contribute to gains in classification accuracy, generalization to unseen subjects, and patient-specific fine-tuning from few examples. The magnitude of these gains varies with dataset properties, in particular class balance and label ambiguity. These findings indicate that musculoskeletal simulation-based augmentation can address common challenges faced by deep learning applications in physiotherapeutic exercise evaluation.
Open AccessBoosting Automatic Exercise Evaluation Through Musculoskeletal Simulation-Based Augmentation of IMU-Derived Orientation Data
Automated evaluation of movement quality can enhance physiotherapeutic treatment and sports training by providing objective, real-time feedback. However, deep learning models that assess movements captured by inertial measurement units (IMUs) are often limited by data scarcity, class imbalance, and label ambiguity. We present a data augmentation method for IMU-derived orientation data that generates additional examples by systematically modifying movement trajectories and passing them through a musculoskeletal simulation. The approach enforces the joint-range limits of a musculoskeletal model and enables automatic labeling by combining inverse kinematic parameters with a knowledge-based evaluation strategy. Across four datasets of varying complexity, augmented variants closely resemble real-world data and contribute to gains in classification accuracy, generalization to unseen subjects, and patient-specific fine-tuning from few examples. The magnitude of these gains varies with dataset properties, in particular class balance and label ambiguity. These findings indicate that musculoskeletal simulation-based augmentation can address common challenges faced by deep learning applications in physiotherapeutic exercise evaluation.
- Open Access
Representing and Detecting Label Ambiguity in IMU-Based Exercise Evaluation
Home-based physiotherapy is performed without supervision, which leads to incorrect execution and motivates systems that assess movement automatically from inertial measurement units (IMUs). Such systems assign each repetition to a category, yet a relevant share of repetitions fall near a class boundary, where even trained raters disagree. Classifiers trained with one-hot labels collapse these borderline repetitions onto a single class and discard this ambiguity. To address this, we build on label distribution learning, which represents each repetition as a distribution over classes instead of a single label. We introduce a way to construct such distributions without a large rater pool by perturbing the thresholds of a rule-based evaluation procedure to simulate rater disagreement. We train a network to reproduce these distributions with a Kullback–Leibler objective, which we call the ambiguity approach, and compare it against a one-hot cross-entropy baseline on four IMU exercise datasets. From the predicted distribution we then determine whether a repetition is ambiguous and which classes are relevant to it. The ambiguity approach matched or exceeded the baseline classification on all four datasets and detected ambiguity and the relevant classes more reliably. Representing the label distribution in the training target therefore adds information about ambiguity at no cost to classification.
Open AccessRepresenting and Detecting Label Ambiguity in IMU-Based Exercise Evaluation
Home-based physiotherapy is performed without supervision, which leads to incorrect execution and motivates systems that assess movement automatically from inertial measurement units (IMUs). Such systems assign each repetition to a category, yet a relevant share of repetitions fall near a class boundary, where even trained raters disagree. Classifiers trained with one-hot labels collapse these borderline repetitions onto a single class and discard this ambiguity. To address this, we build on label distribution learning, which represents each repetition as a distribution over classes instead of a single label. We introduce a way to construct such distributions without a large rater pool by perturbing the thresholds of a rule-based evaluation procedure to simulate rater disagreement. We train a network to reproduce these distributions with a Kullback–Leibler objective, which we call the ambiguity approach, and compare it against a one-hot cross-entropy baseline on four IMU exercise datasets. From the predicted distribution we then determine whether a repetition is ambiguous and which classes are relevant to it. The ambiguity approach matched or exceeded the baseline classification on all four datasets and detected ambiguity and the relevant classes more reliably. Representing the label distribution in the training target therefore adds information about ambiguity at no cost to classification.
- Open Access
Correction: Spilz, A.; Munz, M. Automatic Assessment of Functional Movement Screening Exercises with Deep Learning Architectures. Sensors 2023, 23, 5
There was an error in the original publication
Open AccessCorrection: Spilz, A.; Munz, M. Automatic Assessment of Functional Movement Screening Exercises with Deep Learning Architectures. Sensors 2023, 23, 5
There was an error in the original publication
- Open Access
Time Series Similarity Score Functions to Monitor and Interact with the Training and Denoising Process of a Time Series Diffusion Model applied to a Human Activity Recognition Dataset based on IMUs
Denoising diffusion probabilistic models are able to generate synthetic sensor signals. The training process of such a model is controlled by a loss function which measures the difference between the noise that was added in the forward process and the noise that was predicted by the diffusion model. This enables the generation of realistic data. However, the randomness within the process and the loss function itself makes it difficult to estimate the quality of the data. Therefore, we examine multiple similarity metrics and adapt an existing metric to overcome this issue by monitoring the training and synthetisation process using those metrics. The adapted metric can even be fine-tuned on the input data to comply with the requirements of an underlying classification task. We were able to significantly reduce the amount of training epochs without a performance reduction in the classification task. An optimized training process not only saves resources, but also reduces the time for training generative models.
Open AccessTime Series Similarity Score Functions to Monitor and Interact with the Training and Denoising Process of a Time Series Diffusion Model applied to a Human Activity Recognition Dataset based on IMUs
Denoising diffusion probabilistic models are able to generate synthetic sensor signals. The training process of such a model is controlled by a loss function which measures the difference between the noise that was added in the forward process and the noise that was predicted by the diffusion model. This enables the generation of realistic data. However, the randomness within the process and the loss function itself makes it difficult to estimate the quality of the data. Therefore, we examine multiple similarity metrics and adapt an existing metric to overcome this issue by monitoring the training and synthetisation process using those metrics. The adapted metric can even be fine-tuned on the input data to comply with the requirements of an underlying classification task. We were able to significantly reduce the amount of training epochs without a performance reduction in the classification task. An optimized training process not only saves resources, but also reduces the time for training generative models.
- 23 January 2023
- Published by Walter de Gruyter GmbH in Biomedical Engineering / Biomedizinische Technik
- Vol. 68 (3) , 263-273
Synchronisation of wearable inertial measurement units based on magnetometer data
Objectives Synchronisation of wireless inertial measurement units in human movement analysis is often achieved using event-based synchronisation techniques. However, these techniques lack precise event generation and accuracy. An inaccurate synchronisation could lead to large errors in motion estimation and reconstruction and therefore wrong analysis outputs. Methods We propose a novel event-based synchronisation technique based on a magnetic field, which allows sub-sample accuracy. A setup featuring Shimmer3 inertial measurement units is designed to test the approach. Results The proposed technique shows to be able to synchronise with a maximum offset of below 2.6 ms with sensors measuring at 100 Hz. The investigated parameters suggest a required synchronisation time of 8 s. Conclusions The results indicate a reliable event generation and detection for synchronisation of wireless inertial measurement units. Further research should investigate the temperature changes that the sensors are exposed to during human motion analysis and their influence on the internal time measurement of the sensors. In addition, the approach should be tested using inertial measurement units from different manufacturers to investigate an identified constant offset in the accuracy measurements.
- 23 January 2023
- Published by Walter de Gruyter GmbH in Biomedical Engineering / Biomedizinische Technik
- Vol. 68 (3) , 263-273
Synchronisation of wearable inertial measurement units based on magnetometer data
Objectives Synchronisation of wireless inertial measurement units in human movement analysis is often achieved using event-based synchronisation techniques. However, these techniques lack precise event generation and accuracy. An inaccurate synchronisation could lead to large errors in motion estimation and reconstruction and therefore wrong analysis outputs. Methods We propose a novel event-based synchronisation technique based on a magnetic field, which allows sub-sample accuracy. A setup featuring Shimmer3 inertial measurement units is designed to test the approach. Results The proposed technique shows to be able to synchronise with a maximum offset of below 2.6 ms with sensors measuring at 100 Hz. The investigated parameters suggest a required synchronisation time of 8 s. Conclusions The results indicate a reliable event generation and detection for synchronisation of wireless inertial measurement units. Further research should investigate the temperature changes that the sensors are exposed to during human motion analysis and their influence on the internal time measurement of the sensors. In addition, the approach should be tested using inertial measurement units from different manufacturers to investigate an identified constant offset in the accuracy measurements.
- Open Access
Automatic Assessment of Functional Movement Screening Exercises with Deep Learning Architectures
(1) Background: The success of physiotherapy depends on the regular and correct unsupervised performance of movement exercises. A system that automatically evaluates these exercises could increase effectiveness and reduce risk of injury in home based therapy. Previous approaches in this area rarely rely on deep learning methods and do not yet fully use their potential. (2) Methods: Using a measurement system consisting of 17 inertial measurement units, a dataset of four Functional Movement Screening exercises is recorded. Exercise execution is evaluated by physiotherapists using the Functional Movement Screening criteria. This dataset is used to train a neural network that assigns the correct Functional Movement Screening score to an exercise repetition. We use an architecture consisting of convolutional, long-short-term memory and dense layers. Based on this framework, we apply various methods to optimize the performance of the network. For the optimization, we perform an extensive hyperparameter optimization. In addition, we are comparing different convolutional neural network structures that have been specifically adapted for use with inertial measurement data. To test the developed approach, it is trained on the data from different Functional Movement Screening exercises and the performance is compared on unknown data from known and unknown subjects. (3) Results: The evaluation shows that the presented approach is able to classify unknown repetitions correctly. However, the trained network is yet unable to achieve consistent performance on the data of previously unknown subjects. Additionally, it can be seen that the performance of the network differs depending on the exercise it is trained for. (4) Conclusions: The present work shows that the presented deep learning approach is capable of performing complex motion analytic tasks based on inertial measurement unit data. The observed performance degradation on the data of unknown subjects is comparable to publications of other research groups that relied on classical machine learning methods. However, the presented approach can rely on transfer learning methods, which allow to retrain the classifier by means of a few repetitions of an unknown subject. Transfer learning methods could also be used to compensate for performance differences between exercises.
Open AccessAutomatic Assessment of Functional Movement Screening Exercises with Deep Learning Architectures
(1) Background: The success of physiotherapy depends on the regular and correct unsupervised performance of movement exercises. A system that automatically evaluates these exercises could increase effectiveness and reduce risk of injury in home based therapy. Previous approaches in this area rarely rely on deep learning methods and do not yet fully use their potential. (2) Methods: Using a measurement system consisting of 17 inertial measurement units, a dataset of four Functional Movement Screening exercises is recorded. Exercise execution is evaluated by physiotherapists using the Functional Movement Screening criteria. This dataset is used to train a neural network that assigns the correct Functional Movement Screening score to an exercise repetition. We use an architecture consisting of convolutional, long-short-term memory and dense layers. Based on this framework, we apply various methods to optimize the performance of the network. For the optimization, we perform an extensive hyperparameter optimization. In addition, we are comparing different convolutional neural network structures that have been specifically adapted for use with inertial measurement data. To test the developed approach, it is trained on the data from different Functional Movement Screening exercises and the performance is compared on unknown data from known and unknown subjects. (3) Results: The evaluation shows that the presented approach is able to classify unknown repetitions correctly. However, the trained network is yet unable to achieve consistent performance on the data of previously unknown subjects. Additionally, it can be seen that the performance of the network differs depending on the exercise it is trained for. (4) Conclusions: The present work shows that the presented deep learning approach is capable of performing complex motion analytic tasks based on inertial measurement unit data. The observed performance degradation on the data of unknown subjects is comparable to publications of other research groups that relied on classical machine learning methods. However, the presented approach can rely on transfer learning methods, which allow to retrain the classifier by means of a few repetitions of an unknown subject. Transfer learning methods could also be used to compensate for performance differences between exercises.
- Open Access
Automatic Assessment of Functional Movement Screening Exercises with Deep Learning Architectures
(1) Background: The success of physiotherapy depends on the regular and correct performance of movement exercises. A system that automatically evaluates these could support the therapy. Previous approaches in this area rarely rely on Deep Learning methods and do not yet fully use their potential. (2) Methods: Using a measurement system consisting of 17 IMUs, a dataset of four Functional Movement Screening (FMS) exercises is recorded. Exercise execution is evaluated by physiotherapists using the FMS criteria. This dataset is used to train a neural network that assigns the correct FMS score to an exercise repetition. We use an architecture consisting of CNN, LSTM and Dense layers. Based on this framework, we apply various methods to optimize the performance of the network. For the optimization, we perform a extensive hyperparameter optimization. In addition, we are comparing different CNN structures that have been specifically adapted for use with IMU data. Finally, the developed network is trained with the data of different FMS exercises and the performance is compared. (3) Results: The evaluation shows that the presented approach achieves a convincing performance in the classification of unknown repetitions of already known subjects. However, the trained network is yet unable to achieve consistent performance on the data of a previously unknown subjects. Additionally, it can be seen that the performance of the network differs significantly depending on the exercise it is trained for.
Open AccessAutomatic Assessment of Functional Movement Screening Exercises with Deep Learning Architectures
(1) Background: The success of physiotherapy depends on the regular and correct performance of movement exercises. A system that automatically evaluates these could support the therapy. Previous approaches in this area rarely rely on Deep Learning methods and do not yet fully use their potential. (2) Methods: Using a measurement system consisting of 17 IMUs, a dataset of four Functional Movement Screening (FMS) exercises is recorded. Exercise execution is evaluated by physiotherapists using the FMS criteria. This dataset is used to train a neural network that assigns the correct FMS score to an exercise repetition. We use an architecture consisting of CNN, LSTM and Dense layers. Based on this framework, we apply various methods to optimize the performance of the network. For the optimization, we perform a extensive hyperparameter optimization. In addition, we are comparing different CNN structures that have been specifically adapted for use with IMU data. Finally, the developed network is trained with the data of different FMS exercises and the performance is compared. (3) Results: The evaluation shows that the presented approach achieves a convincing performance in the classification of unknown repetitions of already known subjects. However, the trained network is yet unable to achieve consistent performance on the data of a previously unknown subjects. Additionally, it can be seen that the performance of the network differs significantly depending on the exercise it is trained for.
- Open Access
- 1 September 2019
- Published by Walter de Gruyter GmbH in Current Directions in Biomedical Engineering
- Vol. 5 (1) , 513-515
Development of a smart fabric force-sensing glove for physiotherapeutic Applications
In this work, a low-priced smart fabric forcesensing glove is developed which is able to measure the total amount and direction of the force a person applies on a patient in a physiotherapeutic context. A device like this would be beneficial for the education of physiotherapists, to measure the progress of a patient and to evaluate the treatment. The proposed device uses a new sensor, which is based on a piezoresistive fabric. This fabric changes its electrical resistance according to the applied stress. The characterization of this sensor revealed that the change in resistance of the sensor is dependent of the amount of force, the loaded area, the total time the sensor is loaded and hysteresis. To compensate these behaviours, an additional sensor based on the same smart fabric was developed which measures the loaded area of the first sensor. By combining these two sensors, it is possible to calculate the applied force. The results show the feasibility to build a measurement system out of smart fabric material that can measure the applied force. Furthermore, the prototype shows promising results in determining the applied force in amount and direction.
- 1 September 2019
- Published by Walter de Gruyter GmbH in Current Directions in Biomedical Engineering
- Vol. 5 (1) , 513-515
Open AccessDevelopment of a smart fabric force-sensing glove for physiotherapeutic Applications
In this work, a low-priced smart fabric forcesensing glove is developed which is able to measure the total amount and direction of the force a person applies on a patient in a physiotherapeutic context. A device like this would be beneficial for the education of physiotherapists, to measure the progress of a patient and to evaluate the treatment. The proposed device uses a new sensor, which is based on a piezoresistive fabric. This fabric changes its electrical resistance according to the applied stress. The characterization of this sensor revealed that the change in resistance of the sensor is dependent of the amount of force, the loaded area, the total time the sensor is loaded and hysteresis. To compensate these behaviours, an additional sensor based on the same smart fabric was developed which measures the loaded area of the first sensor. By combining these two sensors, it is possible to calculate the applied force. The results show the feasibility to build a measurement system out of smart fabric material that can measure the applied force. Furthermore, the prototype shows promising results in determining the applied force in amount and direction.