Workflow for Actor Walking Toward Camera

Haven’t posted in awhile, which means, I have forgotten much of what I learned. Yay. So I have this shot (see link below). So that car behind the actor is a 3D still which I tracked to the camera wobble due to high winds. She starts walking in front of the car so obviously I had to roto her out and drop the car behind her. I struggled with this and ended up with tons of keyframes, so I basically did it manually. I had little time and I know there’s a couple of frames where the mask slips a little, but I’d rather do it from scratch the right way.

So I just watched Mary’s Siggraph demo of the woman with the gun, but that one’s only a zoom. My shot she’s walking, the coat deforms and there is some twisting to her shoulders and arms. I understand the layering process know and to break it up into the paper doll thing… head, torso, arms and legs all being different shapes.

I really only have to worry about the middle of her body being accurate. So one question is, do I use the fine roto layers to generate the track as well, or try to track each section separately and link it to the masks?

:15 sec to :21 sec.

https://vimeo.com/tier10studios/review/200214031/cb85fae3dd

 

Have you considered trying a Chroma key (outside Mocha), that jacket has quite a distinctive hue so you might get a good edge which you could combine with Mocha rotos to make the final matte?

Ah crap. I was just adding the mask for the hair to exclude it from the other calculations and AE Mac crashed. Whew! Took 6 tries to get AE to open the file again. Would just hang and beachball.

Hi Alex,

I second a garbage matte and key for that jacket and hair… really distinct colors and I am super lazy about roto.

BUT! To answer your question… For roto you can break those tracks up into major tracks and then link smaller shapes to them to get all the details you need. No need to redo the tracks. With “paper doll” roto, your mileage may vary depending on the subject. Sometimes you need more than just breaking up at the joints, break them up by planar data too.

Let me know if that works for you.

Cheers,
Mary