Tf frames of a mobile robot

edit

Please read the support page. Do not create answers for updates, discussion or subsequent questions. You can always edit your question. And no, the tf between two frames is not necessarily a constant matrix. It's the (maybe changing) relation between the frames.

You could have edited your initial question. Transforms are not static (but can be). You typically use static transforms for your base -> sensor displacements. The odom -> base_link transform is dynamic and accumulates all movements of the robot. (starting point -> current position)

So this means in order to convert a current point cloud to Starting frame i should first apply static /tf between kinect and base.And then dynamic /odom tf to the resulting point cloud in previous step to convert it to starting point frame map.And then keep adding points to first cloud to get a map?

You can get and apply the complete transform in one step. No need to manually apply multiple transforms. If you odometry would be perfect, you could do just what you descibed. But since IMUs are very noisy and experience a lot of drift you need a more complex slam approach like for example rgbdslam.

Thanks i think i get it now... so i should use IMU transform for a certain distance till the noise is not much and fuse this with rgbdslam's estimate with the package robot_pose_ekf... i guess that should work..!