【摘要】 提出了一种满足一定约束条件的与视角无关的步态识别算法.首先给出了与视角无关步态特征的定义及约束条件,进而探讨了在单目平行线约束下空间点的坐标重建方法,利用相应的坐标转换因子可以从拍摄到的二维图像恢复出关键点的空间三维坐标.然后将人体建模成一个相互连接的三棍模型,利用这种坐标重建方法可以恢复出模型的参数,并定义了由模型参数表示的步态特征向量,即与视角无关的步态特征.理论推导和实验结果表明,这种方法在理想情况下能克服视角因素的影响.虽然得到的正确识别率不高,但它提供了多种视角交叉进行识别的可能性.更多还原

【Abstract】 This paper proposes a novel gait recognition algorithm that is independent of viewpoints under certain constraints.First,we described the definition of the proposed gait feature and its constraints.Then we discussed a coordinate reconstruction method for spatial points under the constraints of monocular parallel lines.With this we could employ the relevant coordinate conversion factor to recover 3D coordinates of some key points from 2D monocular camera images.The human body was modeled and simplified as three connected sticks,and the parameters of that model were estimated using the proposed reconstruction method.Thus,we obtained viewpoint-independent gait features represented by those parameters.Both theoretical calculations and experimental results revealed that the proposed gait features partially avoid the influence of viewpoint under ideal circumstances.Though correct classification rates are not high enough,it provides a useful tool for the identification of human gaits at arbitrary viewing angles.更多还原