Research On Robot Stereo Vision Detection Based On Geometric Invariant

In the control of multi-degree of freedom (MDOF) parallel robot, compared with tradition measurement methods, computer vision based measurement system has the advantage of intelligence, non-touch and more fast. However, in three-dimensional space, the same objects in the image have different characteristics for different viewpoint of camera, or while the camera keep stationary, the objects in the image have different characteristics after the changes through the rotation and translation, and that is a great challenge for the identification of the end of the manipulator in three-dimensional space, so the algorithm which was immune for the rotation and the translation have be need to solve this problem.In view of this visual recognition demand, this article imports the concept of geometric invariants. By the geometric structure analysis of the end-effector of manipulator, inferred geometric invariants for the recognition of space structure. On the derivation of geometric invariants, also import conformal geometric algebra, because it is free from the coordinate model in this framework, only requires geometrical relationship of geometrical objects to process operation and exchanges. The main task of this article is as follows:1) Detailed analysis of the three kinds of geometric transformation group and the basic invariants, and the illustration for the relationship of three kinds of camera projection model and the three kinds of geometric transformation group.2) Doing research for conformal geometric algebra, and gives the interpretation of Geometric algebra's basic operations, the basic definition of conformal geometric algebra and the representation and operation of geometry.3) Describes in simple for the background and application of geometric invariants, analyzes in detail for several geometry invariants of space plane, and relevant method of derivation and application direction.4) The derivation of six spatial points geometric invariant and five spatial lines geometric invariant for Six-DOF parallel manipulator. Using spatial five-lines restraining structure, extract invariant value from the CAD image of the end-effector of manipulator on different viewpoint, and use invariant values as feature to recognize the end-effector of manipulator. In the process of matching also joined the target structure constraint for rid of wrong matching and reduce the calculation.Experiments show that the algorithm still has good effects as there is a larger change with perspective, translation, rotation, brightness and scale in image, prove the feasibility and feasibility ofall algorithms in this thesis.