The essence of point cloud registration is to transform the point cloud from one coordinate system to another .

         Point cloud registration usually requires two point cloud data . The first type of point cloud data is called original point cloud , use S（source） To express . The second type of point cloud data is called target point cloud , use T（Target） To express .

         Point cloud registration is to make the original point cloud S At the target point cloud T Display on the coordinates of . We can determine the transformation relationship of coordinates by finding the point cloud with similar characteristics in the point cloud . for example , The point cloud of the same object appears in both the original point cloud and the target point cloud , And there are some point clouds with similar characteristics in the two point clouds , According to these similar point cloud information, the transformation relationship is calculated .

         Suppose that the rigid body transformation occurs from the original point cloud to the target point cloud , That is, the original point cloud can get the target point cloud through rotation and translation . The rotation and translation processes here use the rotation transformation matrix R Peaceshift transformation matrix T To express . We use it P(S) Represents the point in the original point cloud ,P(T) Represents the point of the original point cloud in the target point cloud coordinate system . Then this transformation relationship can be expressed as ：

         therefore , The main task of point cloud registration is to calculate the rotation matrix R Peaceshift matrix T.

Iterative nearest point algorithm (Iterative Closest Point, ICP)

         First step ： initialization R、T matrix , according to R、T Matrix can get P(T), That is, the coordinates of the original point cloud in the target point cloud coordinate system .

         The second step ： Look for and in the target point cloud P(T) Some recent , And the distance is less than the specified threshold , This threshold can be defined by yourself .

         The third step ： Calculate the Euclidean distance error for the matched points in the second step , And the least square method is used to optimize R、T matrix .

         Step four ： Optimize the third step R、T The matrix goes back to the first step , Re iterate , Until the iteration meets the requirements , Get the final optimized R、T matrix .

ICP Methods classification

        ICP Methods can be divided into point-to-point （PointToPoint） And point to plane （PointToPlane） Two types of .

        PointToPoint： Calculation P(t) And target point cloud T The distance is in the form of point-to-point distance .

        PointToPlane： Calculation P(t) Midpoint to target point cloud T The distance from the plane where the point of , The normal vector of the target point cloud is usually used here .

python 3D point cloud research program _Coding Leaves blog -CSDN Blog _3d Point cloud python It will be continuously updated according to the following directory …… Introduction to point cloud format 、 Point cloud visualization 、 Point cloud projection 、 Generate aerial view 、 Generate front view 、 Point cloud registration 、 Point cloud segmentation 、 Three dimensional target detection 、 Point cloud reconstruction 、 Deep learning point cloud algorithm ……https://blog.csdn.net/suiyingy/article/details/124017716

More 3D 、 Please pay attention to two-dimensional perception algorithm and financial quantitative analysis algorithm “ Lele perception school ” WeChat official account , And will continue to update .

​