基于步态序列的跨视角步态识别

Cross-view gait recognition based on gait sequence

为了提高跨视角步态识别的准确率,充分提取步态中的时间信息,提出了一种基于步态序列的跨视角步态识别模型,该模型利用编码器,并引入三元组损失函数,以此提取步态序列的特征,通过使用生成器与判别器,以及连续帧判别损失对编码器进行修正,确保提取具有时间信息的有效步态特征.针对CASIA-B数据集和OU-MVLP数据集,对提出的方法进行了实验研究,且与卷积神经网络和步态能量图方法进行了实验比较,验证了提出方法的有效性.

In order to improve the accuracy of cross-view gait recognition and fully extract the temporal information in gait, we proposed a cross-view gait recognition model based on gait image sequence. The model combines encoder and triplet-loss to extract the features from gait sequences. Meanwhile, using a generator and discriminators, molel optimizes the encoder according the discriminant loss so that extracts the temporal information in gait. In order to validate the effectiveness of the proposed method, we conducted experimental studies on CASIA-B and OU-MVLP datasets. Moreover, we also compare with other state-of-arts methods which are based on convolutional neural network and gait energy image.