针对关节式物体检测的复杂性, 本文提出一种新颖的视觉推理方法。该方法基于可变形的物

体模型, 同时利用图像中所包含的边缘信息特征( 不依赖于局部特征, 如肤色等) 及各子部件的空间位置关

系, 迭代地进行关节式物体检测和定位估计。实验证明, 该方法有较强的抗背景干扰能力, 视觉上能大幅

度改进关节式物体检测、定位的结果-Aiming at the complex ity of art iculated object detect ion, this paper pro poses a nov el cast􀀁

ing of v isual inference, based on a standard object model, makes full use of the edge featur e o f the imag e

( it does not rely on local features) , as w ell as the space informat ion of objects, and performs an iterat iv e

parsing process, w here w e can sequent ially handle the art iculated object detectio n and estimate the local

object components. Exper iment s show that this metho d has a st rong resistance to the backg round inter􀀁

ference, and can also great ly impro ve the results o f articulated object detect io n by human vision