View recognition based on Procrustes Shape Analysis for gait identification

Gait recognition is an important problem in video surveillance. Most of the existing approaches have provided good performance, but they are often affected by views. In this regard, we present the piecewise approach method for view estimation before recognizing gait. Given a gait sequence, the silhouettes in a gait cycle are obtained, and the stable head and shoulder models are extracted. The Procrustes Shape Analysis (PSA) is applied to analysis the head and shoulder models to get the mean shape. The mean shape is converted to distance series. At last, the distance series is converted to piecewise approximation (PA) series. The PA series represents a view. The method is evaluated on CASIA B database. The results show that the method can recognize view correctly.