Effective method to reduce Covariate issues in gait-based human identification

Current research on gait-based human identification is increasingly interested in improving the performance of various existing gait recognition algorithms using various gait templates. However, the effectiveness of the existing gait template repre-sentation methods to tackle the influence of covariates condition need to be enhanced. An alternative method for reducing the effect of the covariates factor is proposed in the paper. The paper proposes integrating GEI (Gait energy image), a commonly used gait representation method, with GGMI (Gait gradient magnitude image), a Gaussian edge preserving technique for images. Subsequently, HoG(Histogram of Oriented Gradient) feature descriptor is used to extract features, followed by LDA (Linear Discriminant Analysis) is used for identifying different subjects. The experimental study is performed using CASIA-B dataset, the largest gait dataset with three variations, namely normal walk, clothing, and carrying conditions in this paper. We examine the effect of normal walking, clothing, and carrying conditions using the proposed method, using some existing CCR(Correct Classification Rate) evaluation methods. Experimental results show that the proposed method is effective against the factors that reduce the gait recognition performance with an average CCR of 73.33% for all the covariates.