基于WM-CoSaMP重构算法的压缩感知在步态识别中的应用研究

Compressed sensing for gait recognition with WM-CoSaMP

针对步态识别中步态特征提取高维处理的复杂性,在研究压缩感知理论的基础上,提出将压缩感知理论应用于步态识别中的步态特征提取方面。在充分利用步态图像稀疏性的前提下,利用观测矩阵对步态图像进行投影观测,得到的观测值作为步态特征用于步态识别中,实现了特征提取的降维处理,大大降低了计算的复杂性。在步态图像的重构方面,在压缩采样匹配追踪(CoSaMP)的基础上,提出了基于小波树模型的压缩采样匹配(wavelet model-CoSaMP,WM-CoSaMP)的重构算法,进一步提高了重构精度。通过对比实验,验证了WM-CoSaMP重构算法的优越性,以及压缩感知在步态特征提取方面的优越性。

There are some problems in the gait recognition, especially for the complexity of the high dimension features. Based on the study of compressed sensing theory, this paper proposed that compressed sensing was used as a new paradigm for gait feature extraction. While the gait image had sparse representation in some orthonormal basis, projections of the images, which were taken as the gait features could be gotten by the projection matrix. In the reconstruction of the gait image, this pa- per proposed that the WM-CoSaMP based on the wavelet tree model could be used as the recovery algorithm, which could fur- ther improve the precision of the reconstruction. Experiments show that the WM-CoSaMP outperforms the OMP and the Co- SaMP. And other experiments demonstrate that the application of compressed sensing in the gait feature extraction performs better than the PCA and MPCA.