Gaussian mixture model based features for stationary human identification in urban radar imagery

In this paper, we propose a Gaussian mixture model (GMM) based approach to discriminate stationary humans from their ghosts and clutter in indoor radar images. More specifically, we use a mixture of Gaussian distributions to model the image intensity histograms corresponding to target and ghost/clutter regions. The mixture parameters, namely, the means, standard deviations, and weights of the component distributions, are used as features and a K-Nearest Neighbor classifier is employed. The performance of the proposed method is evaluated using real-data measurements of multiple humans standing or sitting at different locations in a small room. Experimental results show that the nature of the targets and ghosts/clutter in the image allows successful application of the GMM feature based classifier to distinguish between target and ghost/clutter regions.