RehabMotion: Síntese de Dados de Movimento para Fisioterapia usando Modelos de Linguagem

• Lucas D. Brandão UFES
• Marcelo P. Guimarães UNIFESP
• Eduardo Zambon UFES
• Giovanni Comarela UFES
• Leonardo Rocha UFSJ
• Diego R. C. Dias UFES

DOI: https://doi.org/10.5753/webmedia.2026.30084

Resumo

The scarcity of labeled human motion data in physical rehabilitation is a major bottleneck for developing intelligent systems for remote patient monitoring, as data collection requires specialized infrastructure and expert annotation. To address this limitation, this article presents RehabMotion, an end-to-end pipeline that specializes large-scale motion language models for the clinical domain without training them from scratch. The proposed approach integrates physical rehabilitation datasets, such as KIMORE and UI-PRMD, with the MotionGPT3 architecture through multi-task fine-tuning based on programmatically generated textual descriptions. To evaluate the quality of the synthetic data, a multi-level assessment combining biomechanical, distributional, and discriminative metrics was implemented. The fine-tuned model achieved high discriminative fidelity, generating synthetic exercises that highly agreed with a reference classifier trained on real data, outperforming the pre-trained baseline which often proved ineffective or harmful due to its generic motion generation. In downstream utility tasks under strict data scarcity scenarios, data augmentation with the fine-tuned synthetic movements substantially increased the classification accuracy, with performance gains saturating as real data became abundant. The complete pipeline, models, and synthetic datasets were made publicly available.

Palavras-chave: Synthetic data generation, physical rehabilitation, motion language models, biomechanical analysis, telehealth

Referências

Ali Abedi, Mobin Malmirian, and Shehroz S Khan. 2023. Cross-modal video to body-joints augmentation for rehabilitation exercise quality assessment. In Joint European Conference on Machine Learning and Knowledge Discovery in Databases. Springer, Torino, Italy, 320–327.

Luiz Felipe de Camargo, Gabriel Rissatti Miguel, Juliana da Costa Feitosa, Luis Guilherme Silva Rodrigues, Diego Roberto Colombo Dias, and José Remo Ferreira Brega. 2025. Combining computer vision and artificial intelligence to support the rehabilitation of post-stroke patients. In International Conference on Computational Science and Its Applications. Springer, 415–432.

Marianna Capecci, Maria Gabriella Ceravolo, Francesco Ferracuti, Sabrina Iarlori, Andrea Monteriu, Luca Romeo, and Federica Verdini. 2019. The KIMORE dataset: Kinematic assessment of movement and clinical scores for remote monitoring of physical rehabilitation. IEEE Transactions on Neural Systems and Rehabilitation Engineering 27, 7 (2019), 1436–1448.

Xin Chen, Biao Jiang, Wen Liu, Zilong Huang, Bin Fu, Tao Chen, and Gang Yu. 2023. Executing your commands via motion diffusion in latent space. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition. IEEE, Vancouver, Canada, 18000–18010.

Swakshar Deb, Md Fokhrul Islam, Shafin Rahman, and Sejuti Rahman. 2022. Graph convolutional networks for assessment of physical rehabilitation exercises. IEEE Transactions on Neural Systems and Rehabilitation Engineering 30 (2022), 410–419.

Gisela Miranda Difini, Marcio Garcia Martins, and Jorge Luis Victória Barbosa. 2021. Human Pose Estimation for Training Assistance: a Systematic Literature Review. In Proceedings of the Brazilian Symposium on Multimedia and the Web (Belo Horizonte, Minas Gerais, Brazil) (WebMedia ’21). Association for Computing Machinery, New York, NY, USA, 189–196. DOI: 10.1145/3470482.3479633

Aline PA Ferreira, Lucas O Alves, Fabio AF Molinares, Marcelo P Guimarães, Vinícius FS Mota, Leonardo CD Rocha, and Diego RC Dias. 2026. A Multidimensional Quality Assessment of Synthetic Movement Data for Post-stroke Rehabilitation. In International Conference on Computational Science and Its Applications. Springer, 233–250.

Maayan Frid-Adar, Eyal Klang, Michal Amitai, Jacob Goldberger, and Hayit Greenspan. 2018. Synthetic data augmentation using GAN for improved liver lesion classification. In 2018 IEEE 15th International Symposium on Biomedical Imaging (ISBI 2018). IEEE, Washington, USA, 289–293.

Jáder Gonçalves,Wyllgner Amorim, Nicolas Sales, and Lucas Cunha. 2025. Framework Procedural para Geração de Dados Sintéticos Baseados em Simulação Estatística: Uma Abordagem para Análise de Comportamento Humano. In Proceedings of the 31st Brazilian Symposium on Multimedia and the Web (Rio de Janeiro/RJ). SBC, Porto Alegre, RS, Brasil, 275–283. DOI: 10.5753/webmedia.2025.16041

Chuan Guo, Shihao Zou, Xinxin Zuo, Sen Wang, Wei Ji, Xingyu Li, and Li Cheng. 2022. Generating diverse and natural 3D human motions from text. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition. IEEE, New Orleans, USA, 5152–5161.

Biao Jiang, Xin Chen, Wen Liu, Jingyi Yu, Gang Yu, and Tao Chen. 2023. MotionGPT: Human motion as a foreign language. Advances in Neural Information Processing Systems 36 (2023), 20067–20079.

Victor YK Kwon, Luis GS Rodrigues, Bartolomeu Zamprogno, Marcelo P Guimarães, Leonardo CD Rocha, Eduardo Zambon, Vinícius FS Mota, and Diego RC Dias. 2026. Semantic Noise Filtering in Post-Stroke Kinematic Data: An Automated Deep Learning Approach. In International Conference on Computational Science and Its Applications. Springer, 601–617.

Jianwei Li, Jun Xue, Rui Cao, Xiaoxia Du, Siyu Mo, Kehao Ran, and Zeyan Zhang. 2024. FineRehab: A multi-modality and multi-task dataset for rehabilitation analysis. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition. IEEE, Seattle, USA, 3184–3193.

Ciro Mennella, Umberto Maniscalco, Giuseppe De Pietro, and Massimo Esposito. 2023. Generating a novel synthetic dataset for rehabilitation exercises using poseguided conditioned diffusion models: A quantitative and qualitative evaluation. Computers in Biology and Medicine 167 (2023), 107665.

Alina Miron, Noureddin Sadawi,Waidah Ismail, Hafez Hussain, and Crina Grosan. 2021. IntelliRehabDS (IRDS) – A dataset of physical rehabilitation movements. Data 6, 5 (2021), 46.

Cynthia C Norkin and D Joyce White. 2016. Measurement of Joint Motion: A Guide to Goniometry (5th ed.). F.A. Davis Company, Philadelphia, USA.

Mathis Petrovich, Michael J Black, and Gül Varol. 2021. Action-conditioned 3D human motion synthesis with transformer vae. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition. IEEE, Nashville, USA, 10985–10995.

Elvis Ribeiro, Alexandre Brandão, Marcelo Guimarães, Leonardo Rocha, José Remo Brega, Rodolfo Villaça, and Diego Dias. 2025. A body tracking framework for neuromotor rehabilitation with multimedia application support. In International Conference on Computational Science and Its Applications. Springer, 345–360.

Elvis Ribeiro, Alexandre Brandão, Marcelo Guimarães, Leonardo Rocha, José Brega, and Diego Dias. 2024. Um framework de rastreamento corporal para reabilitação neuromotora com suporte a aplicativos multimídia. In Proceedings of the 30th Brazilian Symposium on Multimedia and the Web (Juiz de Fora/MG). SBC, Porto Alegre, RS, Brasil, 292–300. DOI: 10.5753/webmedia.2024.241537

Luis. G. S. Rodrigues, Diego R. C. Dias, Marcelo P. Guimarães, Alexandre F. Brandão, Leonardo C. D. Rocha, Rogério L. Iope, and José R. F. Brega. 2021. Upper Limb Motion Tracking and Classification: A Smartphone Approach. In Proceedings of the Brazilian Symposium on Multimedia and the Web (Belo Horizonte, Minas Gerais, Brazil) (WebMedia ’21). Association for Computing Machinery, New York, NY, USA, 61–64. DOI: 10.1145/3470482.3479618

Omar Sherif and Ali Hamdi. 2025. Error-Guided Pose Augmentation: Enhancing Rehabilitation Exercise Assessment through Targeted Data Generation. In Intelligent Methods, Systems, and Applications (IMSA). IEEE, Giza, Egypt, 278–283.

Guy Tevet, Sigal Raab, Brian Gordon, Yonatan Shafir, Daniel Cohen-Or, and Amit H. Bermano. 2022. Human Motion Diffusion Model. ArXiv abs/2209.14916 (2022), 12 pages.

Tiago Trotta, Luis Rodrigues, José Remo Brega, Marcelo Guimarães, Leonardo Rocha, Alexandre Brandão, and Diego Dias. 2024. ReBase: a framework for acquiring and managing body tracking data. In Proceedings of the 26th Symposium on Virtual and Augmented Reality. 152–160.

Aleksandar Vakanski, Hyung-pil Jun, David Paul, and Russell Baker. 2018. A data set of human body movements for physical rehabilitation exercises. Data 3, 1 (2018), 2.

Felipe Reis Valente, Marcelo de Paiva Guimarães, Elder José Reioli Cirilo, and Diego Roberto Colombo Dias. 2022. A multi-agent body tracking application framework applied to physical and neurofunctional rehabilitation. In International Conference on Computational Science and Its Applications. Springer, 459–472.

Sijie Yan, Yuanjun Xiong, and Dahua Lin. 2018. Spatial temporal graph convolutional networks for skeleton-based action recognition. In Proceedings of the AAAI Conference on Artificial Intelligence. AAAI Press, New Orleans, Louisiana, USA, Article 912, 9 pages.

Jianrong Zhang, Yangsong Zhang, Xiaodong Cun, Yong Zhang, Hongwei Zhao, Hongtao Lu, Xi Shen, and Ying Shan. 2023. Generating human motion from textual descriptions with discrete representations. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition. IEEE, Vancouver, Canada, 14730–14740.

Yi Zhou, Connelly Barnes, Jingwan Lu, Jimei Yang, and Hao Li. 2019. On the continuity of rotation representations in neural networks. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition. IEEE, Long Beach, USA, 5745–5753.

Bingfan Zhu, Biao Jiang, SunyiWang, Shixiang Tang, Tao Chen, Linjie Luo, Youyi Zheng, and Xin Chen. 2025. MotionGPT3: Human motion as a second modality. ArXiv abs/2506.24086 (2025), 26 pages.