[Paper Review] Clinical gait data analysis based on Spatio-Temporal features

Rohit Katiyar, Vinay Kumar Pathak|arXiv (Cornell University)|Mar 7, 2010

Gait Recognition and Analysis41 references3 citations

TL;DR

This paper proposes a spatio-temporal gait analysis method using wavelet-transformed joint trajectories and vector quantization with a self-organizing map to classify individuals with and without locomotion impairments. The approach successfully extracts discriminative features that enable accurate differentiation between pathological and normal gait patterns.

ABSTRACT

Analysing human gait has found considerable interest in recent computer vision research. So far, however, contributions to this topic exclusively dealt with the tasks of person identification or activity recognition. In this paper, we consider a different application for gait analysis and examine its use as a means of deducing the physical well-being of people. The proposed method is based on transforming the joint motion trajectories using wavelets to extract spatio-temporal features which are then fed as input to a vector quantiser; a self-organising map for classification of walking patterns of individuals with and without pathology. We show that our proposed algorithm is successful in extracting features that successfully discriminate between individuals with and without locomotion impairment.

Motivation & Objective

• To develop a gait analysis framework that supports clinical assessment of physical well-being rather than just person identification or activity recognition.
• To address the gap in using gait analysis for diagnosing locomotion-related pathologies in clinical settings.
• To extract meaningful spatio-temporal features from gait data that reflect underlying motor impairments.
• To classify gait patterns as pathological or normal using unsupervised learning techniques.
• To validate the method’s ability to distinguish between individuals with and without gait abnormalities using real-world data.

Proposed method

• Joint motion trajectories from gait sequences are extracted using motion capture or video-based tracking.
• The trajectories are transformed using discrete wavelet transform (DWT) to extract multi-resolution spatio-temporal features.
• The wavelet coefficients are used as input to a vector quantizer to reduce dimensionality and extract representative feature vectors.
• A self-organizing map (SOM) is employed for clustering and classifying the quantized features into pathological and non-pathological gait patterns.
• The method leverages the topological preservation of SOMs to visualize and analyze gait pattern distributions.
• Feature selection focuses on temporal and spatial dynamics of joint movements to emphasize gait deviations linked to pathology.

Experimental results

Research questions

• RQ1 Can spatio-temporal features derived from gait trajectories effectively distinguish between individuals with and without locomotion impairments?
• RQ2 How well does wavelet-based feature extraction preserve discriminative information for clinical gait classification?
• RQ3 To what extent can a self-organizing map accurately cluster and classify gait patterns based on pathological status?
• RQ4 Are the extracted features robust enough to detect subtle gait deviations associated with motor dysfunction?
• RQ5 Can this method be applied effectively in a clinical setting for non-invasive gait-based health assessment?

Key findings

• The wavelet-based transformation successfully captures both temporal and spatial variations in gait, enhancing feature discriminability.
• The combination of vector quantization and self-organizing maps enables effective clustering of gait patterns into pathological and normal groups.
• The method demonstrates clear separation between pathological and non-pathological gait clusters in the SOM visualization.
• The approach achieves high accuracy in distinguishing individuals with locomotion impairments from healthy subjects based on gait features.
• The results confirm that spatio-temporal features derived from wavelet-transformed trajectories are effective for clinical gait assessment.
• The framework provides a viable, data-driven method for automated gait-based health screening without requiring expert annotation.