Impact of input channel grouping on CNN-based kinematic prediction from inertial gait data

Abstract

Accurate gait analysis is fundamental to assessing mobility, diagnosing neurological disorders and monitoring rehabilitation progress. Optical motion capture remains the gold standard for kinematic parameters, but its high cost, need for specialised laboratories and limited capture volume restrict clinical use. Wearable inertial measurement units offer a low-cost alternative for continuous monitoring in natural environments, yet deriving clinically meaningful kinematic variables from raw accelerometer and gyroscope signals remains a signal processing challenge. Deep learning, particularly convolutional neural networks (CNNs), can learn hierarchical features directly from raw sensor data, potentially eliminating hand-crafted feature engineering. This study investigates the impact of input channel grouping on the prediction performance of a bespoke CNN architecture for kinematic signal regression from inertial data, specifically for predicting the stride length normalised to body height. We used data from the Machine Learning and Data Analytics Laboratory at Friedrich-Alexander-Universität Erlangen-Nürnberg, comprising 101 gait trials with synchronised triaxial accelerometer and gyroscope data from both ankles and reference stride length from a GAITRite walkway. The CNN took a 400×12 input matrix: 12 sensor signals (6 per ankle), 400 time steps. Two input representations were compared: (i) standard stacking in the original channel order, and (ii) a novel grouped representation with all X-axis (primary gait direction) signals placed contiguously to facilitate cross-channel feature learning by the convolutional filters. Models were trained on 91 trials and tested on 10, using root-mean-square error (RMSE) as primary metric, comparing mini-batch size 30 against full-batch. The grouped representation achieved a substantially lower RMSE (0.17) than the standard one (0.21), a 19% relative improvement. Mini-batch size affected learning-curve smoothness but not final RMSE. These results show that a suitably designed CNN can directly regress kinematic variables from raw inertial data, and that input channel organisation strongly influences performance, underscoring the importance of input structuring even in end-to-end deep learning, particularly for biomechanical time series with strongly co-varying signals. Task-driven channel reorganisation offers practical guidance for deep learning in wearable sensor-based movement analysis.

Keywords

• Convolutional neural network
• gait analysis
• inertial measurement unit
• input representation
• kinematic prediction

Thanks

The authors thank Professor Björn Eskofier and his research group at the Machine Learning and Data Analytics Laboratory, Friedrich ‑ Alexander ‑ Universität Erlangen ‑ Nürnberg, for sharing the inertial and kinematic gait data.

References

1. Ardestani, M. M., Zhang, X., Wang, L., Lian, Q., Liu, Y., He, J., Li, D., & Jin, Z. (2014). Human lower extremity joint moment prediction: A wavelet neural network approach. Expert Systems with Applications, 41(9), 4422‑4433. https://doi.org/10.1016/j.eswa.2013.11.003
2. Carcreff, L., Gerber, C. N., Paraschiv-Ionescu, A., De Coulon, G., Newman, C. J., Aminian, K., & Armand, S. (2020). Comparison of gait characteristics between clinical and daily life settings in children with cerebral palsy. Scientific Reports, 10(1), 2091. https://doi.org/10.1038/s41598-020-59002-6
3. Dorschky, E., Nitschke, M., Martindale, C. F., van den Bogert, A. J., Koelewijn, A. D., & Eskofier, B. M. (2020). CNN‑based estimation of sagittal plane walking and running biomechanics from measured and simulated inertial sensor data. Frontiers in Bioengineering and Biotechnology, 8, 604. https://doi.org/10.3389/fbioe.2020.00604
4. Gong, J., Wei, D., Gao, T., & Liu, Z. (2024). Multi-sensor and deep learning based ankle joint motion intention prediction under diverse gait speed conditions. In 2024 IEEE 20th International Conference on Automation Science and Engineering (CASE) (pp. 2705-2710). IEEE.
5. Hannink, J., Kautz, T., Pasluosta, C. F., Gassmann, K.-G., Klucken, J., & Eskofier, B. M. (2017). Sensor-based gait parameter extraction with deep convolutional neural networks. IEEE Journal of Biomedical and Health Informatics, 21(1), 85-93.
6. Ma, S., Cao, Y., Robertson, I. D., Shi, C., Liu, J., & Zhang, Z. Q. (2025). Knowledge-based deep learning for time-efficient inverse dynamics. IEEE Transactions on Neural Systems and Rehabilitation Engineering, 33, 522-531. https://doi.org/10.1109/TNSRE.2025.3530992
7. Mundt, M., Koeppe, A., David, S., Bamer, F., Potthast, W., & Markert, B. (2020). Prediction of ground reaction force and joint moments based on optical motion capture data during gait. Medical Engineering & Physics, 86, 29‑34. https://doi.org/10.1016/j.medengphy.2020.10.001
8. Ng, G., Gouda, A., & Andrysek, J. (2023). Convolutional neural network for estimating spatiotemporal and kinematic gait parameters using a single inertial sensor. Sensors, 23(24), 9803. Ohno, S., & Nagamune, K. (2024, November). Development of Gait Classification System Using Six-Axis Inertial Sensors by CNN-LSTM. In 2024 Joint 13th International Conference on Soft Computing and Intelligent Systems and 25th International Symposium on Advanced Intelligent Systems (SCIS&ISIS) (pp. 1-5). IEEE. https://doi.org/10.1109/SCISISIS61014.2024.10760026

9. Prisco, G., Pirozzi, M. A., Santone, A., Esposito, F., Cesarelli, M., Amato, F., & Donisi, L. (2024). Validity of wearable inertial sensors for gait analysis: A systematic review. Diagnostics, 15(1), 36. https://doi.org/10.3390/diagnostics15010036
10. Rampp, A., Barth, J., Schuelein, S., Gassmann, K.‑G., Klucken, J., & Eskofier, B. M. (2015). Inertial sensor‑based stride parameter calculation from gait sequences in geriatric patients. IEEE Transactions on Biomedical Engineering, 62(4), 1089‑1097. https://doi.org/10.1109/TBME.2014.2368211
11. Renggli, D., Graf, C., Tachatos, N., Singh, N., Meboldt, M., Taylor, W. R., Steglitz, L., & Schmid Daners, M. (2020). Wearable inertial measurement units for assessing gait in real-world environments. Frontiers in Physiology, 11, 90. https://doi.org/10.3389/fphys.2020.00090
12. Ripic, Z., Kuenze, C., Andersen, M. S., Theodorakos, I., Signorile, J., & Eltoukhy, M. (2022). Ground reaction force and joint moment estimation during gait using an Azure Kinect‑driven musculoskeletal modeling approach. Gait & Posture, 95, 49‑55. https://doi.org/10.1016/j.gaitpost.2022.04.005
13. Shah, V. R., & Dixon, P. C. (2025). Gait speed and task specificity in predicting lower-limb kinematics: A deep learning approach using inertial sensors. Mayo Clinic Proceedings: Digital Health, 3(1), 100183. https://doi.org/10.1016/j.mcpdig.2024.11.004
14. Zhu, K., Li, D., Li, J., & Shull, P. B. (2024). Real-time IMU-based kinematics in the presence of wireless data drop. IEEE Journal of Biomedical and Health Informatics, 28(11), 6512-6524. https://doi.org/10.1109/JBHI.2024.3432668.