A simple but high-precision registration method in 3D vision measurement

Abstract

3D vision measurement is a popular technique to solve non-contact and non-destructive measurement task. In any 3D measurement technology, registration is a very important procedure to put two adjacent but different areas into one area. And it is a key step to ensure the final 3D measurement precision. Most algorithms used to solve the registration problem can be classified to either area-based techniques or feature-based techniques. Feature-based technique is the most commonly used method in 3D and 2D registration. Some flags may be pasted or signed on the entities in order to make the common area easy to be identified. A lot of researcher focus on feature-based technique, and present some algorithms to compute the rotation matrix and translation matrix, to solve the registration problem. But the registration effect is poor and the object usually distorts its real shape. We present a novel registration method, without complex algorithms; just use the searching merits of computer program. Rotation matrix is computed by three key angles Ψ, θ, and Φ. If these angles are found with high precision, the rotation matrix will be confirmed. Then the translation matrix can be calculated using the coordinates of three common points. From a lot of experiments, the proposed registration method is proved that it is a simple, easy programmed, easy operated, and with high precision. Furthermore, it can be widely used in any other 3D measurement system.