Unsupervised flexible speed gait real-time classification

The paper deals with human gait classification based on the notion thatgait types can be analyzed into a series of consecutive postures types.Pedestrian's silhouettes are extracted using the Background Subtraction methodand then features are represented by moment. In the learning stage, a methodusing recursion for establishing the standard gait base matrix is proposed. Thesimilarity sequences between the incoming silhouette and the silhouette indatabase using the Zernike Velocity Moments lead to a matrix. In the automaticclassification stage, the weighted average of the maximal mean value and theminimal standard deviation algorithm for vector of each motion type is proposedwhich is adopted to deduce behavior classification of walker at flexible speedin the outdoor environment. Finally we test the algorithm on CASIA dataset,which includes walking, running, bending, etc. eight actions. The experimentresults indicate the method can be a choice for solving the flexible speed gaitreal-time classification problem. ICIC International © 2009.