基于双线性建模及隐马尔可夫模型的步态识别算法

Human gait recognition based on bilinear modeling and HMM

针对存在两个未知隐含特性的步态识别问题提出了一种新的识别算法,将特性分别看作“内容”和“风格”,对图像序列以状态确定的连续HMM-EM估计“内容”类型,引入非对称双线性模型理论对结果建模,通过SVD和NN聚类实现对“风格”的归类判定。实验选择人体两侧轮廓到质心垂线距离作为步态特征,通过在CASIA步态库上的实验证明,该算法能有效提高判别率,对未知风格或内容类型判断有较好的适应性。同时对影响步态识别准确性的其他因素也做了讨论。

Motivated by bi-factor-invariant human gait recognition problem, a new gait recognition algorithm was proposed in which two factors, generically called ＂style＂ and ＂content＂ were analyzed and manipulated. First, image sequences were clustered into a fixed number of content with fixed dynamics HMM-EM algorithm. Then the observation data were generated according to an asymmetric bilinear model. After that, SVD and NN were used to classify new sequences characterized by a different style label. Body width between vertical llne through centroid and outer contour was used as the feature. Test on the CASIA datasets shows the proposed method＇s advantage in increasing the recognition rate and adapting to new styles or content. Some other facts affecting ID identification were also discussed.