I have a human skeletal data for 25 different joints from a depth sensor camera (Kinect) while performing a lifting work. But the data has some noise and inferred data due to occlusion. I am trying to use Tobit-Kalman Filter for data processing but could not figure out how to get the co-variance of process and observation noise for my data? I read multiple papers regarding the use of Tobit-Kalman Filter for processing Kinect data but could not find an in-depth explanation of the process involved. Can anyone help me out with my problem?

Thank you.