Search

Research on Efficient Collaborative Technology of Kalman Filter Fused with LSTM Algorithm

Author(s)

Jiayi Shen, Luyan Shen, Sitong Ruan, Zhihui Xu, Jingyi Xu*

Affiliation(s)

Artificial Intelligence College, Zhejiang Dongfang Polytechnic, Wenzhou, Zhejiang, China *Corresponding Author

Abstract

Aiming at the industrial technical pain points of traditional intelligent rehabilitation training systems, such as poor time-series synchronization of equipment data, low adaptation accuracy of XR devices, high positioning and tracking delay, as well as the long development cycle and high R&D cost of rehabilitation digitalization, this paper proposes an efficient collaborative technical scheme integrating Kalman filter with Long Short-Term Memory (LSTM) network. The proposed technology gives full play to the advantages of Kalman filter in real-time noise reduction, accurate dynamic state estimation and low-latency iterative updating, and combines the powerful capabilities of LSTM network in mining nonlinear features and modeling long-term dependencies of time-series data. It realizes complementary advantages and in-depth collaboration of the two algorithms, and constructs a low-latency, high-precision and high-robustness dynamic positioning and tracking system, which effectively solves the problem of data asynchronization and adaptation disconnection between sensing data of traditional rehabilitation training equipment and XR virtual reality devices. Meanwhile, relying on procedural modeling and dynamic compilation component technology, a lightweight and reusable rehabilitation system development framework is established, breaking through the limitations of traditional customized development modes, greatly shortening the development cycle of digital rehabilitation systems and reducing R&D and iteration costs. Experimental results show that the fused technology can significantly improve the tracking accuracy of rehabilitation motion data and the efficiency of equipment synchronous adaptation, reduce system operation delay, and effectively optimize the project development process, providing core technical support for the efficient development, accurate operation and large-scale implementation of XR intelligent rehabilitation systems.

Keywords

Kalman Filter; LSTM; Algorithm Fusion; XR Rehabilitation Equipment; Time-Series Synchronization; Low-Latency Positioning; Procedural Modeling

References

[1] Hao G J, Lin S C, Lian Y L. Electromagnetic Array Positioning Method Based on LSTM-EKF Fusion. Navigation Positioning and Timing, 2025, 12(06):78-88. [2] Xu B Y, Pang Y L, Li J F. Electromagnetic Radiation Source Positioning Method Based on Mobile Direction-Finding LSTM-KF. Signal Processing, 2026, 42(02):249-256. [3] Zhang M, Wang H Y. Kalman-LSTM Time-Series Data Fusion Positioning Algorithm Based on Attention Mechanism. Computer Engineering and Applications, 2025, 61(12):156-163. [4] Liu C L, Jin Y R. Research on Intelligent Motion Posture Tracking Technology Based on Multi-Modal Sensor Fusion. Optics and Precision Engineering, 2024, 32(09):210-218. [5] Chen X, Li Y T. Real-Time Positioning Technology Collaborated by Lightweight LSTM and Kalman Filter. Application Research of Computers, 2025, 42(05):1482-1487. [6] Wang M X, Wu K. Application of Procedural Modeling and Component-Based Development in XR Rehabilitation System. Information Technology and Informatization, 2026(01):90-93. [7] Meta Reality Labs. “Varifocal Display Latency Optimization.” SID Symposium Digest, 2023, 54(1):112-115. [8] DeepMind. “Reinforcement Learning for Adaptive Fire Scenarios.” Nature Machine Intelligence, 2022, 6(7):732-744. [9] Yang Dai, Zhiyuan Luo. “Review of Unsupervised Person Re-Identification.” Journal of New Media, 2021, 3(4):129-136. [10] Meskat Jahan, Manajir Hassan, Sahadat Hossin, et al. “Unsupervised person Re-identification: A review of recent works,” Neurocomputing, 2024, 572(1):127193–127196.