About the Point Cloud Coordinate System Issued by the RGB-D Camera

edit

@Mike Scheutzow Does this mean that I must transform the point cloud released by this coordinate system into a common coordinate system under ROS. Then I can do related work, such as filtering too high or too low point clouds. Then transform the processed point cloud inverse coordinates back to _optical frame.

Does this mean that I must transform the point cloud released by this coordinate system into a common coordinate system under ROS

Yes, but it is easy to implement if the camera is in your TF tree: use tf2_ros Buffer.transform() to convert each point into the TF frame you want to work in (typically base_link or map.)

For a robot-mounted camera, your TF usually looks something like base_link -> camera_base -> camera_optical. The camera_optical origin is at corner of the image sensor inside the camera body.

Typically you won't need to transform the data points back to camera_optical frame, but nothing stops you from doing so.