Effective Morphological Features for Human Posture Recognition in camera-based systems

The effective morphological features for four main postures recognition are proposed in this paper. The morphological features are extracted from the relationship between length and width of posture envelop, which covers the silhouette in each frame. The achievement in recognition, 96.26%, implementing on SOM classifier and testing with complex posture set, implied that the proposed morphological features are suitable for four main postures classification. For the previous features comparison, Length-Width ratio and DFT coefficients, our features increased the accuracy by 2.82%. The recognition with low-cost camera-based system makes the proposed system ordinary, flexible, and nonspecific, which can be applied to a falling detection application in an available surveillance system in the house.