提出将改进的步态光流图(LK-GFI)与视角相结合的方法来解决步态识别易受视角影响的问题。该方法采用Lacus-Kanade (LK)光流法获得连续两帧侧影图像间的光流场，并构造步态特征图像LK-GFI，利用成像原理计算人的行走方向以确定视角。首先，离线建立目标在各视角下的LK-GFI数据库；然后，提取待识别人的当前视角和LK-GFI；最后，用欧式距离度量同一视角下待识别人与目标的LK-GFI之间的相似性。分别采用CASIA数据库和实际室内获得的步态序列对该方法进行了验证。结果显示，错误拒绝率分别为7.95%和9.12%，与采用传统的步态能量图(GEI)相比分别降低了12.5%和14.45%；与采用步态光流图(GFI)相比分别降低了7.77%和6.74%。该方法识别准确性高，实时性强，对多视角有较强的鲁棒性。

A method combined the modified Lucas-Kanade Gait Flow Image (LK-GFI) with the view was proposed to solve the problem that personal identification based on a gait is sensitive to view change. The Lacus-Kanade optical method was used to compute the optical flow between two silhouettes to construct LK-GFI, and the view was obtained according to the walking direction of the person. The LK-GFI database for the target at different views was established, then the new person's view and LK-GFI were extracted. At last, the similarity between the new persons LK-GFI and the targets LK-GFI at the same view was computed by the Euclidian distance method. The performance of this method was evaluated on the data in the CASIA database and the data obtained in indoor lab environment, and the False Rejection Rate (FRR) is 7.95% and 9.12% respectively. It is reduced by 12.5% and 14.45% respectively compared with that of the Gait Energy Image (GEI), and by 7.77% and 6.74% respectively compared with that of the Gait Flow Image (GFI). The proposed method has high recognition accuracy, strong real-time and the robustness to view changes.