Work

Calibration-Free Lower Limb Activity Recognition via Streaming Feature Calibration

Year: 2026

Type: conference-paper

Abstract: Surface electromyography (sEMG) signals pattern recognition is essential for lower-limb exoskeleton control, yet its long-term reliability is severely compromised by dynamic domain shifts such as muscle fatigue and inter-subject variability. Compounding this issue, conventional static models often report inflated accuracies due to data leakage in randomized cross-validation, failing critically under strict chronological and crossdomain testing. This paper introduces Streaming Feature Calibration via Active State tracking (SFC-AS) for lower limb activity recognition. Instead of updating classifier weights, SFC-AS employs exponential moving averages to track the continuous manifold evolution of sEMG features in real time, dynamically calibrating inputs to mitigate domain discrepancies. We validated this framework using a chronological evaluation protocol on multi-day, multi-subject lower-limb data. Results reveal that SFC-AS eliminates the performance illusions of static models and demonstrates strong resilience to temporal drift.

Source: International Conference on Computer, Control and Robotics (ICCCR)

DOI: 10.1109/icccr69988.2026.11642722

Authors: Qiuping Bi, Cheng Shen, Zhongyi Li, Hao Zheng

Institutions: Liaoning Communication University, Shenyang Aerospace University, Chinese Academy of Sciences, Ningbo Institute of Industrial Technology, Beihang University

Language: English

Topic: Muscle activation and electromyography studies

Subfield: Biomedical Engineering

Field: Engineering

Domain: Physical Sciences

SDG: Good health and well-being

Open Access status: hybrid

OpenAlex ID: W7203748388

FWCI:

Cites: 12

Cited by:

Related to: 10