Angular features analysis for gait recognition

Automatic gait recognition is an emergent biometrics identification system for recognizing humans by the way they walk. Its system is non-invasive because it operates from a distance via video cameras. The videos cum image frames are manually labeled to extract angular displacements of thigh's and lower leg's rotation, and foot flexion. The angular displacements data is analyzed using standard approach of Principal Component Analysis (PCA) and Canonical Analysis (CA). A cycle extraction procedure consisting of cubic-spline interpolation in SVR (Support Vector machine for Regression) and resampling within zero crossings is performed beforehand for an invariant analysis due to difference in walking speed of subjects. Combined dataset, is proposed for analyzing features that provide the most variations in gait recognition. Results have shown that the hip accounts for most variations among the three limbs' displacements data. Also, difference in temporal information of gait's signal does affect the recognition performance.