Detecting Gait Phases from RGB-D Images Based on Hidden Markov Model

Area of Science:

• Biomedical Engineering
• Computer Vision
• Human Motion Analysis

Background:

• Gait analysis provides critical insights into human physiological status and health.
• Accurate monitoring of patient gait is essential for various medical applications.
• Identifying distinct gait phases aids in detailed biomechanical analysis.

Purpose of the Study:

• To develop an image processing method for extracting human gait phases from RGB-Depth images.
• To utilize a hidden Markov model (HMM) for gait phase recognition.
• To enable accurate and automated gait analysis for medical applications.

Main Methods:

• Processing front-view RGB-Depth image sequences to extract lower body depth profiles and distance features.
• Employing a hidden Markov model where extracted features serve as observation vectors and gait phases as hidden states.
• Training the HMM using randomly selected image samples for gait phase estimation.

Main Results:

• The developed method successfully estimates gait phases from image data.
• The study confirms a 60-40% distribution for two major gait phases.
• The mid-stance phase of gait was recognized with a high precision of 85%.

Conclusions:

• The proposed image processing and HMM-based method is effective for gait phase extraction.
• This technique offers a precise approach for analyzing gait patterns, particularly the mid-stance phase.
• The findings support the use of RGB-Depth imaging and HMMs for advanced gait analysis in healthcare settings.