Empirical Mode Decomposition and Hilbert Spectrum for Abnormality Detection in Normal and Abnormal Walking...

Area of Science:

• Biomedical Engineering
• Computer Vision
• Signal Processing

Background:

• Sensor-based Human Activity Recognition (HAR) offers remote monitoring capabilities, including gait analysis.
• Wearable sensors for HAR can be complex and inconvenient.
• Vision-based approaches, like PoseNET, provide an alternative for capturing human pose data.

Purpose of the Study:

• To develop a method for detecting gait abnormalities using vision-based pose detection.
• To process raw skeleton and joint data from PoseNET into angular displacement signals.
• To analyze gait patterns for transitions from normal to abnormal movement.

Main Methods:

• Utilized PoseNET for skeleton and joint detection from video data.
• Transformed key-joint and skeleton data into angular displacement signals representing gait patterns.
• Applied Empirical Mode Decomposition and the Hilbert Huang Transform (HHT) to analyze time-frequency characteristics of gait signals.
• Calculated energy in the time-frequency domain to differentiate between normal and abnormal gait transitions.

Main Results:

• Extracted joint change information using HHT to study subject behavior, particularly during turning.
• Demonstrated that the energy of the gait signal is higher during transition periods compared to normal walking periods.
• Successfully identified gait abnormalities through analysis of time-frequency energy.

Conclusions:

• The proposed method effectively detects gait abnormalities using vision-based pose estimation and HHT.
• This approach offers a promising non-invasive technique for remote gait monitoring and analysis.
• The findings highlight the utility of time-frequency energy analysis for identifying abnormal gait transitions.