M.Goffredo -Self-Calibrating View-Invariant Gait Biometrics读后记

题目：Self-Calibrating View-Invariant Gait Biometrics

作者：Michela Goffredo, Member, IEEE, Imed Bouchrika, Member, IEEE, John N. Carter, Member, IEEE, and Mark S. Nixon, Associate Member, IEEE

期刊：IEEE TRANSACTIONS ON SYSTEMS, MAN, AND CYBERNETICS—PART B: CYBERNETICS

Abstract—We present a new method for viewpoint independent gait biometrics. The system relies on a single camera, does not require camera calibration, and works with a wide range of camera views. This is achieved by a formulation where the gait is self-calibrating. These properties make the proposed method particularly suitable for identification by gait, where the advantages of completely unobtrusiveness, remoteness, and covertness of the biometric system preclude the availability of camera information
and specific walking directions. The approach has been assessed for feature extraction and recognition capabilities on the SOTON gait database and then evaluated on a multiview database to establish recognition capability with respect to view invariance. Moreover, tests on the multiview CASIA-B database, composed of
more than 2270 video sequences with 65 different subjects walking freely along different walking directions, have been performed.
The obtained results show that human identification by gait can be achieved without any knowledge of internal or external camera parameters with a mean correct classification rate of 73.6% across all views using purely dynamic gait features. The performance of the proposed method is particularly encouraging for applicationin surveillance scenarios.

该作者前期相关文章：Human Perambulation as a Self Calibrating Biometric 

View Invariant Gait Recognition

可以说发表在这个期刊上的文章确实是很牛的。这篇文章的工作也确实值得肯定。

但读起来太不容易了，从去年看过这篇文章，看不懂，想看懂，仍然看不懂，搁置，到现在翻出来想搞懂，中间仍然步态明白。翻出作者之前的相关工作看，看懂一点，仍然不懂。

现把大体方法介绍下吧，以备忘。

首先，该文是 视角无关的步态识别，采用了Pose-based方法（相对于Pose-free），即位姿都转换到侧面视角lateral view。

第二，该文采用的model-based方法（相对于statistical apprearance based)，即提取人体行走时的腿部关节点，然后建立大腿，小腿的角度周期变化模型，利用FFT提取参数（大小腿角度钟摆模型是nixon团队前期的工作，这里应用到了视角无关的步态上）

第三，具体内容：（本文的主要工作点）

         a,首先估计关节点，可能遮挡的关节如何估计，本文使用先验（人体的比例），加能找到的腿部点，然后估计关节，拟合大小腿，提取钟摆模型；

         b,（难点）利用摄影几何，试图建立各种角度到 侧面视角的映射方法，利用了腿部在行走中，图像虽然变化，但实际总是等长的建立等式。

         

         c,分类；打分制，类似KNN, K近邻的 为1，其他为0，然后再加权，c类内，第i个近邻，那么权值逐渐变小 （平方函数）。这样越接近的权重越大。

总之，三(b)是难点，仍然没搞懂，后面如有空，继续搞。为之学习了 摄影几何，发现对该文不需要那么深的知识。只需要简单的就可以，但是仍然不懂。