Human Gait Recognition for Arbitrary View Angles

The methods of human gait recognition can be categorized into image-based and model-based methods. Image-based methods are easily affected by the inaccurate foreground detection. Model-based methods can extract the gait features (joint angles) robustly and avoid the noise interference problem. However, the gait model fitting process is time consuming. To overcome these shortcomings, a dynamic gait model fitting algorithm is proposed to fit the gait model to the silhouette image sequence of a walking people efficiently. Then a new model-based projection feature and model-based gait features are generated to improve the recognition accuracy. Furthermore, view transformation for the gait images is applied to recognized human identities for arbitrarily view angles. Finally, the accuracy of human gait recognition using various kinds of gait features is analyzed.