基于步态高斯图及稀疏表示的步态识别

为了突出步态特征的差异性,解决单一视角及跨视角下识别率低的问题,分析了步态高斯图及稀疏表示的原理,提出在步态高斯图的基础上,应用稀疏表示的方法来进行分类。提取的步态高斯图增加了不同人步态的差异性进而提高了步态识别率。该方法首先获取步态高斯图,其次获取测试阶段所需的过完备字典,最后通过测试样本的重构误差进而分类。实验结果表明:该方法显著提高了跨视角下的步态识别率,对穿大衣、携带背包下的步态识别具有很强的鲁棒性。

Gait Recognition based on Sparse Representation and Gait Gaussian Image

In order to highlight the difference in gait characteristics and solve the problem of low recognition rate in single and cross perspective, this paper analyses the principles of Gait Gaussian Image(GGI) and sparse representation,and proposes the method that applies sparse representation to recognition based on the Gait Gaussian Image.The extracted Gait Gaussian Image increases the gait difference of different people and then improves the gait recognition rate.Firstly,this paper obtains the Gait Gaussian Image,and then gets the over-complete dictionary needed in the test period.Finally,classify the object according to the reconstruction error of test samples.Experimental results show that this method significantly improves the gait recognition rate of cross perspective,and has strong robustness to the condition of carrying a coat or a bag.