You would want to check out the software that the film animators like Pixar use to do motion tracking on people wearing special suits with tracking targets.  That's the basis for doing fluid character animation.  I think the video game folks do the same thing for greater realism in sports video games.

I can see a couple of problems trying to do optical motion analysis of clouds.  Ideally, you want to track the particle position with time, so you could also derive velocities.  But all cloud particles look the same, and you can only view the surface, basically.  Internal measurements would require radar, which will also give velociities (speed and vector) directly, given enough transmitters and receivers.  You wouldn't be tracking particles so much as the aggregate property in a "small" volume.

The other problem with optical tracking is what I call the "point of view" problem.  You can write the equations of motion against a camera viewing a solid body in a perspective view.  That's what I did many years ago for analyzing rocket launches of robotic aircraft.  But the solution requires a solid body with multiple targets (such as nose, tail, wingtips) with a known initial position.  Multiple cameras (high speed movie cameras) provide some accuracy.  I derived angular rates and velocity from that data using multiple cameras.  Maybe cutting out angular rates from the solution would reduce the equations, but I still wonder about getting range on the trajectory!

I worked with a post-grad PhD on a project with similar elements when I was in grad school.  The guy was using a pulsed laser to make double-exposure holograms of test particles in a wind tunnel.  The idea was to take a microscope-like device, look into the hologram, and measure the 3-D distances between exposures on a particle (if you look close, they look different...).  Neat stuff at the time.