ECSS Paris 2023: CP-AP07
INTRODUCTION: Sport climbing is a sport in which players climb various artificial walls with climbing holds. There is still no uniform performance indicator for evaluating climbing skills, since climbers climb walls of various angles in complex postures. Therefore, researches have been conducted on motion measurement and analysis of sport climbing movements. As in other sports, optical motion capture is mainly used for motion measurement in sport climbing [1], and it provides highly accurate posture measurement. However, there are environmental constraints, such as the need to install multiple cameras for measurement. In contrast, several pose estimation methods using RGBD images have been proposed [2]. However, the accuracy of pose estimation is highly dependent on the climbers posture. We propose a new method for highly accurate human pose estimation using the ICP algorithm with RGBD images. This method applies model fitting based on a three-dimensional ICP algorithm to the two-dimensional pose estimation results obtained from OpenPose [3]. METHODS: The proposed method improves the accuracy of 2D pose estimation results using OpenPose by model fitting with ICP to achieve pose estimation for various postures. The proposed method uses model fitting based on the ICP. The system estimates the climber’s posture by fitting a predefined human body model to depth images. First, the 2D position of each joint obtained by OpenPose from the RGB image is projected onto the depth image and converted to a 3D position. These positions are used as the initial positions in each frame of the human body shape model in ICP. Next, the human shape model is fitted to the depth image by ICP for 3D pose estimation. The body model, divided into parts, is prepared in advance. Then, robust 3D pose estimation can be achieved by referring to the model positions in the previous frame when the 2D pose estimation by OpenPose fails. RESULTS: To evaluate the accuracy of the proposed method, an experiment was conducted in an indoor climbing gym. The subjects were two advanced climbers and one intermediate climber. The posture measurement results from the optical motion capture system were used as the true values for accuracy evaluation. Compared with the conventional methods, the accuracy of the proposed method was improved. From the experiment, it was qualitatively confirmed that the ICP works effectively. It was also shown that the proposed method estimates 3D position of each joint accurately even when OpenPose failed to estimate 2D position of each joint. CONCLUSION: In this study, we proposed the method for highly accurate human pose estimation using the ICP with RGBD images. From the experiment conducted in an actual climbing gym, it was confirmed that the proposed method can estimate climbers posture with higher accuracy than conventional methods. As a future work, we aim to generate body models from RGBD images as well. [1] Iguma,et al., SII,2020 [2] D. Pandurevic ,et al.,icSPROTS,2020 [3] Z. Cao, et al., TPAMI,2019
Read CV Akihiro KawamuraECSS Paris 2023: CP-AP07
INTRODUCTION: Previous studies have shown that in young adults balance performance is affected by arm movements, especially during challenging, posture threatening conditions (e.g., balancing at height) [1] as well as by support surface [2]. Additionally, it has been shown that virtual reality (VR) is a suitable tool to expose individuals to posture threatening situations using simulated height scenarios [3]. However, little is known about the effects of free versus restricted arm movements and using (in-)congruent support surfaces during a dynamic balance task performed in a posture threatening virtual environment (VE) provided through a head-mounted display (HMD). METHODS: Thirty healthy, non-acrophobic young adults (16 f, 14 m; age: 22.3 ± 2.2 years) were exposed to a posture threatening VE using a HMD (Oculus Quest 2, Meta Inc. USA). Participants were instructed to balance forward across a 3-m long and 0.1-m wide virtual wooden beam extending from the 80th floor of a skyscraper and then return while walking backward. Each participant performed two trials (i.e., one with and one without arm movement) on incongruent (i.e., even ground) and congruent (i.e., physical wooden beam) support surface, respectively. The position and dimensions of the physical beam matched those of the virtual beam and the order of test conditions was randomized. Balancing times were recorded and analysed using a 2 (arm condition: free, restricted) × 2 (support surface: ground, beam) repeated measures ANOVA. Additionally, the iGroup Presence Questionnaire was administered after balancing on congruent and incongruent surface, respectively in order to analyse how present individuals experienced themselves in the VE. RESULTS: The ANOVA revealed a significant main effect for arm condition (F = 13.132, p < .01, η2p = .31) as well as for support surface (F = 40.817, p < .01, η2p = .59), but there was no significant interaction effect (p = .08) between these factors. Balancing times were significantly larger when participants walked with restricted compared to free arm movements and when walking across the beam compared to walking on the ground. Further analysis revealed that perceived presence in the VE was significantly larger (t = 2.952, p < .01, Cohen’s d = .54) when participants balanced on the physical beam compared to even ground. CONCLUSION: Similar to the real environment, restricting arm movements and/or changing the support surface seem to be effective means to modulate task difficulty during balance testing and/or training with healthy young individuals in VR. Larger balancing times during the trials using the physical beam may be explained by increases in participants perceived presence. This highlights the need to align visual and somatosensory input when a posture threatening environment is used in VR. References [1] Hill MW et al., Gait & Posture, 2023. [2] Dault MC et al., Gait & Posture, 2001. [3] Cleworth TW et al., Gait & Posture, 2012. Contact Email: simon.schedler@uni-due.de
Read CV Simon SchedlerECSS Paris 2023: CP-AP07
INTRODUCTION: Rugby scrummaging represents a crucial phase of the game, characterized by high-intensity physical efforts and a significant impact on match outcomes [1]. The horizontal force generated by the entire pack is a key determinant of scrum success. However, existing measurement systems are unable to provide 3D, individual, and on-field assessments of ground reaction forces (GRF). A previous study developed a Machine Learning (ML) model to predict the 3D-GRF with instrumented insoles during scrummaging but this was conducted on recreationally active subjects without specific scrummaging experience [2]. Thus, this study aimed to investigate to what extent this model can be used to predict the 3D-GRF for elite rugby players. METHODS: Twelve elite rugby players (12 males; age: 20+/-1 ans; height: 191+/-7 cm; weight: 116+/-13 kg) performed three pushing trials of 15 seconds, against a fixed scrum machine. They wore commercial instrumented insoles (Loadsol Pro®, Novel, Germany, 200Hz) inside their shoes with each foot on a force plate (Sensix, 1000Hz) covered with artificial turf. The force plate data served as the reference for 3D-GRF measurements. For each subject, one trial was used to infer data from the pre-trained ML model, while the remaining two trials were utilized to personalize the model for each subject. The model’s performance was assessed by computing the Root Mean Square Error (RMSE) between the prediction and the reference, the correlation coefficient (r), and the percentage of RMSE compared to the mean resultant force. RESULTS: The initial model inference yielded mean RMSE values of 42±9N on the Medio-Lateral (ML) axis, 168±87N on the Antero-Posterior (AP) axis, and 180±49N on the Vertical (V) axis, with correlation coefficients r of 0.708±0.110 (ML), 0.825±0.102 (AP), and 0.571±0.123 (V). RMSE percentages relative to the mean resultant force were 4.2±0.7% (ML), 15.8±5.0% (AP), and 18.0±4.8% (V). After personalization, RMSE were 32±10N (ML), 99±41N (AP), and 135±38N (V), and correlation coefficients r were 0.838±0.062 (ML), 0.911±0.044 (AP), and 0.680±0.168 (V). The mean percentages of RMSE compared to the average resultant force were 3.1±0.6% (ML), 9.7±2.8% (AP), and 14.0±5.5% (V). CONCLUSION: The findings of this study indicate that an ML model pre-trained on data from recreationally active individuals without specific rugby scrummaging experience is not optimal for accurately estimating 3D-GRF in elite rugby players. Model personalization for each participant improved performance, suggesting that personalization is a promising approach for enhancing ML model performance when the model is trained on a non-specific dataset. However, further improvement in model performance could be achieved by pre-training the model on a dataset more closely aligned with the data used for personalization, to ensure reliable predictions. References [1] Scott et al., J. Sci. Med. Sport, 2023 [2] Pomarat et al., IEEE Xplore, 2025
Read CV Zoé PomaratECSS Paris 2023: CP-AP07