Related Experiment Video
Updated: Jul 18, 2026
06:49
Evaluation of a Smartphone-based Human Activity Recognition System in a Daily Living Environment
Published on: December 11, 2015
8.9K
A Method of Detecting Human Movement Intentions in Real Environments.
Summary
This study presents a novel method for detecting human movement intentions for exoskeleton control, achieving 93.4% accuracy by fusing eye-tracking and IMU data with terrain identification for enhanced safety.
Area of Science:
- Robotics
- Human-Computer Interaction
- Biomechanics
Background:
- Exoskeleton control requires accurate movement intention detection for seamless transitions between locomotion modes.
- Real-world environmental uncertainties pose challenges to reliable intention detection, risking user safety.
Purpose of the Study:
- To develop and validate a self-correcting method for detecting human movement intentions in real environments.
- To enhance the accuracy and reliability of intention detection for exoskeleton applications.
Main Methods:
- Fusion of gaze data (eye tracker) and inertial measurement unit (IMU) signals at the feature extraction level.
- Utilized a convolutional neural network for terrain identification using scene camera images.
- Implemented a decision fusion layer for online self-correction based on predicted intentions and identified terrains.
Main Results:
- Achieved an overall accuracy of 93.4% using combined feature and decision fusion.
- Demonstrated improved prediction accuracy by fusing gaze data with IMU signals.
- Validated the method's feasibility across various locomotion tasks including level walking, ramps, and stairs.
Conclusions:
- The proposed method effectively detects human movement intentions in real-world scenarios.
- Fusion of multimodal sensor data (gaze, IMU, vision) significantly enhances detection accuracy and reliability.
- This approach holds promise for safer and more intuitive exoskeleton control.
