目标位姿测量中的三维视觉方法

要测量出一组特征点分别在两个空间坐标系下的坐标，就可以求解两个空间目标间的位姿关系，实现上述目标位姿测量方法的前提条件是要保证该组特征点在不同坐标系下，其位置关系相同，但计算误差的存在却破坏了这种固定的位置关系，为此，提出了两种基于模型的三维视觉方法－基于模型的单目视觉和基于模型的双目视觉，前者从视觉计算的物理意义入手，通过简单的约束迭代求解实现模型约束，后者则将简单的约束最小二乘法和基于模型的单目视觉方法融合在一起来实现模型约束，引入模型约束后，单目视觉方法可以达到很高的测量精度，而基于模型的双目视觉较传统的无模型立体视觉方法位移精度提高有限，但姿态精度提高很多。

3D Visual Methods for Object Pose Measurement

The related pose between two space objects can be determined if the coordinates of a group marks at two reference frames are measured. The precondition of the above method is that the site relationship among the marks must be changeless at the two reference frames, but this relationship will be destroyed by the computational error. So two model based visual methods include model based monocular vision method and model based binocular stereo method are introduced. The model based monocular vision method starts with the physical meaning of vision, and get the 3D space coordinates of marks by simple restrictive iterative. The model based binocular stereo method inoculates the model based monocular vision and advanced least square method to realize the model restriction. The compare analysis between the two methods and the experiment result will be given at last. Using these method, the model based monocular vision can reach very high measurement precise, and the model based binocular stereo has a little higher measure than the model based monocular vision. For any method, the size of quantization error will effect the measurement precise of visual method directly. These method has been used in mobile robot visual guided experiment at our laboratory successfully, the speed of image grabbing is 0 3 second per frame.