Thursday, 1 March 2012

Bowling ball drop rotoscope

Taking a brief break from fcurves, thought I'd have a bit of fun playing around with Softimage's rotoscope options. You can basically attach an image to the back of any viewport to use as modelling reference. That in itself isn't especially useful for my purposes, but in addition to static images, Softimage allows you to attach image sequences as well.


I found a video on Youtube of a guy dropping a bowling ball in front of the camera and decided to try playing around with it.

Mucking around with trying to convert the file to an image sequence appears to have messed up the frame rate (it was probably originally shot at 29fps, come to think of it) so the final version appears to be on fast forward, but hey ho.


To attach the sequence, click on the display option dropdown in the top right of any viewport and click on 'Rotoscope' which will bring up the 'Rotoscopy' property page.


Clicking on New > New from file then opens up a dialogue box - by default, Softimage looks in the 'pictures' folder for all rotoscopy images. I used the terribly-converted bowling ball image sequence, which Softimage then pins to the back of the scene.



Scrubbing the timeline will play back the image sequence! Magical. The only problem is scale - as you can see, simply dropping in a primitive sphere reveals how tiny the image is, which isn't always ideal. In most cases you could simply scale down your object but in instances where you have a complex character rig, you don't really want to scale it down too much.


The other option is to enable the 'attached to camera' checkbox in the rotoscopy property page. This basically fixes the rotoscope to the lens of the camera so you're able to pan around with the camera and line up your object over the rotoscope without altering its position. You can zoom in and out of your object without effecting the scale of either the object or the rotoscope.


When you're happy with the positioning, you can turn on 'enable pixel zoom' (the square magnifying glass at the top of the viewport' which locks the position of the rotoscope and object together. 

Sorry if that made no sense. I'm still very tired!

Anyway, I didn't get too fancy with it. I just wanted to see how it worked, mostly, so literally all I did was trace the position of the ball with keyframes - no fancy fcurve editing (yet)!

video

Wireframe:

video

No distracting background:

video


Of course direct rotoscoping never usually works - for things like ball bounces the results can be decent enough, but in the case of more organic or lively motion (particularly walk cycles) the resulting animation can lose a lot of the original personality and weight. Rotoscope and motion capture can be a great starting point for overall timing and to get a very basic animation down, but they usually require so much cleanup afterwards you may as well be animating from scratch!

Anyway, I actually think that the rotoscope looks mostly OK - there's certainly a feeling of some kind of weight, but there's a lot of tweaking and refining that could be done. I think the overall speed is probably a bit too fast, mostly due to the problems in converting the frame rate - so I'll probably sort that out with the region select tool in the dope sheet. Once the speed is sorted out it should be a little easier to pinpoint the exact problems with it, but I think generally the series of successive bounces could use some work. Looking at the original footage there is a noticeable pause at the peak of each bounce before it comes back down, which mine definitely lacks. This should be easy enough to fix.

As an aside, I also think that the ball rolls away a little too long at the end, but that could simply be due to the fact that the rest of the animation is too fast in comparison.

No comments:

Post a Comment