The main concept want to get my head around is what I should be doing with the Surface plain when trying to generate the best possible track in Motor. How can I best apply the search area & tracking settings for the examples I’ve uploaded?
I’m currently trying Motor out on some archive footage of Lulu where she needs to be completely rotoscoped out. A couple of the shots are quite tricky so I was wondering if anybody may have some suggestions to get the best tracking results out of Motor.
I keep getting a server error when I try and upload any stills but will hopefully sort this out asap.
Thanks for the advice. I’ll definitely use separate tracking elements for each body part. The reason why I only had one layer contour in the screen grabs I sent you was because she never lifted the rest of her arms into frame, so thought I might get away with one spline. However, I didn’t think about the fact that the arms are still moving a bit differently to the torso.
In the end that particular shot as been much simplified, I now only need to roto the moments where she does actually raise her hands into shot – which will be straight forward. What I have uploaded to your ftp site are two quicktimes containing shots where we have to completely extract Lulu from the background. I haven’t started a Motor project for these yet.
On the both shots traditionally I’d again do as you suggested and build separate contours/splines for each body part – form-arms, upper arms, torso, legs etc.
It’s not easy to advise you by only looking at these screen shots.
Can you upload the media… low/proxy res be fine. Also send your Motor project file pls.
You can upload to:
This is a shared ftp account. If the media is confidential then we will provide you with a personal account.
Some suggestions nonetheless…
Remember to track elements with identical motion. You may find that on this shot, having seperate tracking layers for each arm, and one for the body… may work better.
Also, not that I can gage from the screen shots, it may help splitting arms into shoulder and forearms…
But let us have the media and have a better look.
usually you should leave the search area settings as they are. The main reason for changing them would be where you have a jumpy track on a slow moving object. Then you could try to reduce the search area and you might get a better result.
One technique I sometimes find useful is to track the “core” of an object, say the body of a person. Then create separate layers for the parts of the object as Jay says, for each of the arms and the head, say. However instead of tracking each of these layers, link the tracking data from these layers to the “core” layer, using the “Link Contours to Track” pull-down menu. This way you can often avoid having to track the other layers. You might find that a partly obscured arm, for instance, is difficult to track. Then it makes more sense to link it to the track of the body its attached to, but roto it separately. Of course you should apply this on a case by case basis. If something is easy for motor to track, go ahead and track it.