Hello all:
Here is a link to a great video from FluidImages, an OptiTrack Arena alpha tester. They have been doing lots of captures and this video shows their setup at a remote location with capture and rendering.
http://media.naturalpoint.com/video/opt ... _Demo2.mov
Demo video of Arena setup and capture
Re: Demo video of Arena setup and capture
Hello Jim,
I just saw the video looks really good...however it felt a little jerky at times, the other videos i saw (tennis and boxing etc) looked way better in terms of fluid movement...was it because of the amount of cameras used? How many cameras were used in this Fluid images one...you know?
I pre-ordered the system with an 8 camera setup a week ago...I CANT WAIT to try it out:):), decmber 3rd seems such a looong way still:)
Thx for posting this:)
Nildo
I just saw the video looks really good...however it felt a little jerky at times, the other videos i saw (tennis and boxing etc) looked way better in terms of fluid movement...was it because of the amount of cameras used? How many cameras were used in this Fluid images one...you know?
I pre-ordered the system with an 8 camera setup a week ago...I CANT WAIT to try it out:):), decmber 3rd seems such a looong way still:)
Thx for posting this:)
Nildo
Re: Demo video of Arena setup and capture
As a general rule, its hard to judge mocap data from final animation output. There are just way too many things going on in between the system and the final animation to judge.
Were filters applied to the opticals? Should they have been? Were gaps filled? How? Were there swaps? Were they resolved? When the data was applied to a kinematic model, was the calibration accurate? What were the retargetting parameters? Was the skeletal data filtered? Should it have been?
What you really want to see to judge the performance is raw c3d data and then c3d data that has been cleaned via the system's more automatic means. The raw data should be well formed (it will have some noise, but it should be in the half a centimeter or less range.) It will have gaps. And it probably will have swaps. Then, once it has gone through basic automatic-ish cleanup, the noise should be gone. no gaps less than 1/4 of a second should exist. Swaps might still exist but should be infrequent.
If an optical system does that, then its doing pretty well. And having used Arena for the past month or so on Alpha, and having spent a lot of time with a pretty expensive Vicon system of the same caliber, I personally think the OptiTrack system creates clean optical data when set up correctly. I've been experimenting with different lenses and camera setups, as the principal limiting factor in the system is the resolution of a single camera (lensing can play into that.) However, I've been able to get pretty solid results with an 8 camera system with a functional volume about 10' square at human scale, which is not too shabby for 8 cameras. I intend to go up to 12 within a few months. 8 is enough to test and do basic moves. If you are expecting a volume you can do runs and stunt work in (20' square at least), you're going to need to get more cameras. The angular coverage from 8 cameras is ok for that kind of work (not great, but ok). The problem is that you have to pull the cameras back to cover that kind of space, and then their resolution does start to factor in. The only way to deal with that, is to get more cameras... and at the OptiTrack pricepoint, thats always an option. Whereas with other solutions, its not an option.
I think there is a bit more work to be done on the software in the area of reconstruction and smoothing. But honestly its nothing that you can't filter out later with a heavy hand, if you need to. And its still Beta after all. Also, I feel the software is still a little too linear with its data pipe where it should be non-linear. But again, its Beta. We'll see where we are at 1.0 and 2.0.
Were filters applied to the opticals? Should they have been? Were gaps filled? How? Were there swaps? Were they resolved? When the data was applied to a kinematic model, was the calibration accurate? What were the retargetting parameters? Was the skeletal data filtered? Should it have been?
What you really want to see to judge the performance is raw c3d data and then c3d data that has been cleaned via the system's more automatic means. The raw data should be well formed (it will have some noise, but it should be in the half a centimeter or less range.) It will have gaps. And it probably will have swaps. Then, once it has gone through basic automatic-ish cleanup, the noise should be gone. no gaps less than 1/4 of a second should exist. Swaps might still exist but should be infrequent.
If an optical system does that, then its doing pretty well. And having used Arena for the past month or so on Alpha, and having spent a lot of time with a pretty expensive Vicon system of the same caliber, I personally think the OptiTrack system creates clean optical data when set up correctly. I've been experimenting with different lenses and camera setups, as the principal limiting factor in the system is the resolution of a single camera (lensing can play into that.) However, I've been able to get pretty solid results with an 8 camera system with a functional volume about 10' square at human scale, which is not too shabby for 8 cameras. I intend to go up to 12 within a few months. 8 is enough to test and do basic moves. If you are expecting a volume you can do runs and stunt work in (20' square at least), you're going to need to get more cameras. The angular coverage from 8 cameras is ok for that kind of work (not great, but ok). The problem is that you have to pull the cameras back to cover that kind of space, and then their resolution does start to factor in. The only way to deal with that, is to get more cameras... and at the OptiTrack pricepoint, thats always an option. Whereas with other solutions, its not an option.
I think there is a bit more work to be done on the software in the area of reconstruction and smoothing. But honestly its nothing that you can't filter out later with a heavy hand, if you need to. And its still Beta after all. Also, I feel the software is still a little too linear with its data pipe where it should be non-linear. But again, its Beta. We'll see where we are at 1.0 and 2.0.
-
seanboyette
- Posts: 164
- Joined: Wed Jan 17, 2007 10:31 am
Re: Demo video of Arena setup and capture
Nildo,
I have had excellent results with the overall setup - most of the jerk that you are seeing, is because of the filter applied to the video - this is really a demonstration of the hardware and setup, rather than animation itself. Also keep in mind that software setup was early alpha, and the optitrack guys have made leaps and bounds enhancing the package.
My setup is with 8 cams and Arena, I export the c3d to Motionbuilder and apply characterization in MB. The most notable example of motion is the drummer - it has been one of my best captures. Some notes, once you get your setup, spend some time to understand the capture volume and get a solid volume, it will make your results that much better! You really need space!
If you have any questions, let me know.
Sean
I have had excellent results with the overall setup - most of the jerk that you are seeing, is because of the filter applied to the video - this is really a demonstration of the hardware and setup, rather than animation itself. Also keep in mind that software setup was early alpha, and the optitrack guys have made leaps and bounds enhancing the package.
My setup is with 8 cams and Arena, I export the c3d to Motionbuilder and apply characterization in MB. The most notable example of motion is the drummer - it has been one of my best captures. Some notes, once you get your setup, spend some time to understand the capture volume and get a solid volume, it will make your results that much better! You really need space!
If you have any questions, let me know.
Sean
Re: Demo video of Arena setup and capture
Hi Guys,
Thanks alot for the info...sorry if my questions seem abit harsh hehe, it will my 1st time that im gonna use a mocap system, so i dunno anything about it...and YES the drummer is AWESOME!
and thanks alot Brad for that explanation, it cleared alot of my doubts.
hmm hehe what are swaps??
Nildo
Thanks alot for the info...sorry if my questions seem abit harsh hehe, it will my 1st time that im gonna use a mocap system, so i dunno anything about it...and YES the drummer is AWESOME!
and thanks alot Brad for that explanation, it cleared alot of my doubts.
hmm hehe what are swaps??
Nildo
Re: Demo video of Arena setup and capture
a marker swap is when the system has properly reconstructed a marker as it moves through space but has confused the identification of which marker it actually is temporarily. Often, this is not just a confusion of a single marker, but a combined confusion about 2 or more markers. You'll see them swap their IDs for a short time.
Its actually a little more visually evident in ViconIQ, where the markers are colored. When they are colored, you see the points exchange colors.
If you take the example of a three point rigid object, and the system accidentally swaps two of the points, the rigid will flip upside down and then when the swap is over, it will flip back upright. Within a more complex kinematic model, a swap will often result in a pop of a joint rather than a full inversion, depending on the implementation of the kinematic system.
I've not worked with Arena enough to give an overall opinion of swaps and how to deal with them within the software. However, within IQ, my approach is usually to remove all the labels, seed the system with some manual labeling in the trouble areas, and have it try to figure it out again taking my data into account. That particular method isn't currently available in Arena, though there is a manual tool to fix swaps after the automatic ID has taken place.
Its actually a little more visually evident in ViconIQ, where the markers are colored. When they are colored, you see the points exchange colors.
If you take the example of a three point rigid object, and the system accidentally swaps two of the points, the rigid will flip upside down and then when the swap is over, it will flip back upright. Within a more complex kinematic model, a swap will often result in a pop of a joint rather than a full inversion, depending on the implementation of the kinematic system.
I've not worked with Arena enough to give an overall opinion of swaps and how to deal with them within the software. However, within IQ, my approach is usually to remove all the labels, seed the system with some manual labeling in the trouble areas, and have it try to figure it out again taking my data into account. That particular method isn't currently available in Arena, though there is a manual tool to fix swaps after the automatic ID has taken place.