Human action recognition using the image contour

A template match-based method for human action recognition is presented by using the moving human silhouette contour.First,the human silhouette contour extraction is obtained by background subtraction and shadow elimination.A new contour descriptor called as border-radius is defined,which is employed to transform the 2D contour into a 1D distance vector.The hierarchical clustering method is used to extract key postures of human action using the cost functions as eigenvectors.At last,a human behavior sequence can be presented by encoding key postures.The edit distance is proposed to measure the similarity between the templates and the tested sequences.The experimental results show that this method can achieve the correct recognition rate above 84.3 % for 6 common actions.