Sebastian Dill, Arjang Ahmadi, Martin Grimmer, Dennis Haufe, Maurice Rohr, Yanhua Zhao, Maziar Sharbafi, Christoph Hoog Antink

Abstract

In recent years, significant research has been conducted on video-based human pose estimation (HPE). While monocular two-dimensional (2D) HPE has been shown to achieve high performance, monocular three-dimensional (3D) HPE poses a more challenging problem. However, since human motion happens in a 3D space, 3D HPE offers a more accurate representation of the human, granting increased usability for complex tasks like analysis of physical exercise. We propose a method based on MediaPipe Pose, 2D HPE on stereo cameras and a fusion algorithm without prior stereo calibration to reconstruct 3D poses, combining the advantages of high accuracy in 2D HPE with the increased usability of 3D coordinates. We evaluate this method on a self-recorded database focused on physical exercise to research what accuracy can be achieved and whether this accuracy is sufficient to recognize errors in exercise performance. We find that our method achieves significantly improved performance compared to monocular 3D HPE (median RMSE of 30.1 compared to 56.3, p-value below 10−6) and can show that the performance is sufficient for error recognition.

DOI: https://doi.org/10.3390/s24237772

Bastian Latsch, Niklas Schäfer, Stephan Schaumann, Steffen Graffe, Asghar Mahmoudi, Martin Grimmer, Alexander A. Altmann, Omar Ben Dali, Julian Seiler, Stephan Rinderknecht, Philipp Beckerle, Mario Kupnik

Abstract

In human-machine interaction, muscle activity serves as an indicator of the human motion intent and is mostly assessed using electromyography (EMG). EMG requires special skin preparation, accurate electrode placement, and is affected by sweating. In contrast, force myography (FMG) measures the mechanical muscle contraction, thereby promising reliable long-term acquisition for gait phase detection. With one individual piezoelectric sensor patch per muscle, reduced cross-talk around the limb circumference is expected compared to state-of-the-art elastic straps comprising several sensors linked together. In this work, seven physically unimpaired participants wore four 3D-printed, highly sensitive ferroelectret sensors along with four EMG electrodes on the lower limbs while walking on an instrumented treadmill at normal walking speed. Tibialis anterior (TA) and vastus medialis (VM) are best suited for heel strike and toe-off detection with the FMG sensors to discriminate between swing and stance phases. The signal variability across all strides is exceptionally low. The characteristic FMG peaks differ from the ground reaction force reference by 3.6% ± 2.0% of the stride time for TA and 1.2% ± 1.2% for VM. Using the customizable ferroelectret sensors, this FMG system provides a substitute or supplement in sensor fusion for conventional EMG.

DOI: https://doi.org/10.1109/SENSORS60989.2024.10784617

Bastian Latsch, Niklas Schäfer, Stephan Schaumann, Steffen Graffe, Asghar Mahmoudi, Martin Grimmer, Alexander A. Altmann, Omar Ben Dali, Julian Seiler, Stephan Rinderknecht, Philipp Beckerle, Mario Kupnik

Abstract:

Muscle activity can be utilized to detect a user’s movement intention in human-machine interactions. Electromyography (EMG) is the prevalent method to assess muscle activity but has critical disadvantages in its practical implementation, particularly for long-term measurements, which are required in wearable devices. Force myography (FMG) offers an alternative approach, mechanically detecting muscle contractions by assessing the tissue deformation. Highly sensitive ferroelectrets qualify for detecting subtle muscle movements using FMG. We propose to attach individual ferroelectret sensor patches superficially to the muscle bellies, thereby avoiding crosstalk from other muscles. The sensors are additively manufactured using polypropylene, which allows for application-specific customizations. In order to investigate the suitability of these sensors for FMG during walking, we conduct a study with seven unimpaired participants. Heel strike and toe-off events are determined from the data of four leg muscles, while the ground reaction force provided by an instrumented treadmill serves as a reference. Tibialis anterior (TA) and vastus medialis (VM) suit best for determining toe-off and heel strike events, with small deviations of 3.3%±1.7% for toe-off detection at the TA and -1.2%±1.2% for heel strike detection at the VM. The experimental results demonstrate the suitability of our ferroelectret sensors as a possible substitute for EMG, underscoring their potential in assistive devices such as exoskeletons and prostheses.

DOI: https://doi.org/10.1109/UFFC-JS60046.2024.10794064

Niklas Schäfer, Bastian Latsch, Stephan Schaumann, Steffen Graffe, Omid Mohseni, Julian Seiler, Alexander A. Altmann, Omar Ben Dali, Martin Grimmer, André Seyfarth, Mario Kupnik, Philipp Beckerle

Abstract:

Wearable sensors for measuring muscle activity and the resulting limb motion are essential for various applications, including motion analysis, human-machine interaction, and control of assistance systems such as exoskeletons. We examine wearable ferroelectret sensors for force myography (FMG), in particular comparing the FMG data to muscle activity measurements obtained by electromyography (EMG). Muscle activity data were acquired from a participant during walking using pairs of FMG and EMG sensors placed on the vastus medialis and biceps femoris muscles. We analyze activation timing, signal variability, and the effect of different walking speeds on signal amplitude. The FMG sensors prove to be effective in capturing activity patterns similar to those obtained by EMG, with mean absolute errors of 6.1% for the vastus medialis and 18.9% for the biceps femoris. Furthermore, the FMG signals are more consistent across strides, exhibiting less than half the stride-to-stride variability. In conclusion, our ferroelectret sensors for FMG have the potential to increase robustness of muscle activity measurements in applications such as motion analysis and control of wearable robots.

DOI: https://doi.org/10.1109/UFFC-JS60046.2024.10793709

Asghar Mahmoudi, Morteza Khosrotabar, Jannis Kuhmann, Martin Grimmer, Stephan Rinderknecht, Maziar A. Sharbafi

Abstract:

This study explored the feasibility of incorpo-rating a passive Brain-Computer Interface (BCI) into the individualization process of wearable assistive devices, with a specific focus on lower-limb exoskeletons. These devices can greatly improve mobility and quality of life for those with lower limb impairments. However, their optimal performance relies on personalized adjustments. This study offered a unique approach to user feedback without intentional control. A one-degree-of-freedom knee exoskeleton was utilized, introducing different levels of impedance to create distinct assistance levels. The analysis identified specific brain clusters, notably the right premotor and supplementary motor cortex, exhibiting significant differences in activity between the easiest and the most challenging conditions. This critical proof-of-concept step demonstrated that through decoding the user's brain activity, passive BCI could support the individualization of assistive technology. This approach indicated the potential to enhance the performance and user experience of wearable assistive devices.

DOI: https://doi.org/10.1109/BioRob60516.2024.10719932

Bastian Latsch, Niklas Schäfer, Martin Grimmer, Omar Ben Dali, Omid Mohseni, Niklas Bleichner, Alexander A. Altmann, Stephan Schaumann, Sebastian Wolf, André Seyfarth, Philipp Beckerle, Mario Kupnik

Description

In the experiment, one test person wearing a ferroelectret insole in the right shoe walks on an instrumented treadmill. The dataset contains data from four ferroelectret sensors in the insole and vertical ground reaction forces (GRF) from an instrumented treadmill across five different walking speeds. The ferroelectret insole data are filtered and given in Volt. The GRF treadmill data from the right side are filtered and given in Newton. The data are segmented into steps from one heel strike to the same foot's following heel strike and normalized into 1000 data points per step. Each row contains one step with 1000 columns. The amount of steps/rows depends on the walking speed: 74 for slowest v050, 124 for fastest v150. Please find the full description in the corresponding publication, full instructions for usage in the README.md file. Copyright of thumbnail image 2024, IEEE.

Link: https://tudatalib.ulb.tu-darmstadt.de/handle/tudatalib/4152