基于周期序列宽度图的步态识别方法

A Novel Gait Recognition Method Using Periodic Sequence Width Image

为了快速地提取步态,提出了一种基于周期序列宽度图的步态识别方法。该方法先按周期将侧影轮廓序列转换为宽度向量序列,然后再将宽度向量序列转换为用灰度值表示的周期序列宽度图。周期序列宽度图中的灰度值及其变化能清晰地反映步态运动,是一种以图的形式直观准确表征步态时空变化的方法。这种周期序列宽度图不仅保留了单帧图像中侧影的外观结构信息,而且很好地体现了步态随时间的变化。另外,还运用DCT对提取的步态特征进行降维,并采用RBF神经网络进行步态分类。在常用步态数据库上的测试结果表明,该方法简单而有效。

A novel gait recognition method based on periodic sequence width images is proposed in order to gain gait quickly and correctly. This method transforms the 2D silhouette contours sequences to width vector sequences according to the gait cycle. The vector sequences are turned into the periodic sequence width images, presented by grey values. These grey values can exactly depict the gait motion. The periodic sequence width images contain both the static and dynamic gait characteristics, which not only keep the shape structure information of each frame, but also represent the variant movement information of gait sequence excellently. Furthermore, the new method greatly reduces the image dimension by discrete cosine transforms and adopts the radial basis function neural networks to identify the gait. Experiments prove this method is simple and effective in theory and application.