Hey guys

I just wanted to ask a general question about motion capture with the Kinect, not just with Ipi but with all solutions (its for my Uni project).

I have never developed a piece of software for the Kinect and wanted to know just how limited developers are by the Kinect and its drivers. Is everyone kind of working within the same realms of possibility when it comes to hardware or is there some magical ways that people are pushing the physical limitations with software. The reason I am asking is that I need to justify choosing Ipi over other solutions and I figured at the time that the real difference comes down to what you DO with the data not necessarily getting better raw data.

To attempt to clarify, are solutions (lets say Brekel and Ipi) running under the same limitations of the hardware but using software to interpret and use the data to varying effects? Or is the software being written capable of say pushing the Kinect to do something it shouldn't.

I hope this made sense, it's a difficult question to formulate in my head properly.

Thanks to anyone who contributes!

Oli

The Main difference between iPi and Brekel and or Nuicapture is that iPi finalized output is more accurate.
Brekel and Nuicapture are amazing programs in their own right however, they are for the real time enthusiast and are suited for folks that don't need high accuracy.
Brekel and Nuicapture are also realtime programs designed to capture both face and body mocap data in realtime, Brekel also has a streaming plug in to MotionBuilder, the Brekel and Nuicapture programs are for fast and easy fun with the kinect, and or for realtime facial animation use.

iPi Mocap is also the only package that allows you to use two Kinects sensors simultaneously. This, combined with a sophisticated post-tracking process, is responsible for the extra accuracy in body capture but it also allows you to capture more complicated motions. (Full rotations, for example.)

That said, Brekel and NUI Capture each offer facial capture, which I understand iPi Soft is not especially interested in supporting at this time. In the case of Brekel, face capture is being offered as a separate product called Brekel Pro Face.

I have an interest in trying face capture for a future project so I have invested in a license of Brekel Pro Face, but I have not been particularly interested in using a single Kinect realtime body capture system because the results I've seen so far don't compare to what I've been able to capture and track using dual-Kinects in the iPi package. This is strictly my opinion of course.

I have to admit, I am a little biased towards iPi Soft, having used their software since its early beta days, but I also own two other single-Kinect based programs that I can compare against. Both of these products have unique appealing features but for body capture accuracy and flexibility, iPi DMC still wins out for me.

iPi is also the only system that also supports PS3 Eye, Wii Motion Plus and PS Move.

G.

_________________
Greenlaw
Artist/Partner - Little Green Dog | My Demo Reels (2013,) (2015,) (2017,) and (2019)

Watch a one minute excerpt on Vimeo now!

Hey Greenlaw!

Thanks for your opinions, they certainly help me to construct a valid argument to my peers. I have tried Brekel pro body and Face capture and personally found that real time tracking is not a great for motion capture. I often found that if the Kinect lost tracking of a limb for even a few seconds then the clean up time in another program was a monumental task. I much prefer to track and clean up my data in post, I think you put it best by saying that the other solutions could be seen as fast and fun as opposed to accurate and usable. I have already stated in my documentation that Ipi is (what I believe to be) the only software to use the PSeye which immediately gives it an advantage over other solutions. I did forget to mention the motion controller support, so thanks for that :)

I'll try and re iterate the technical side of what I was asking, knowing very well that I may confuse matters further.



Above is an image of IClone5, Nui and Brekel. As you can see, the points in which is appears to be tracking are pretty much the same across the board, which got me to wondering. Are they all using the same basic tracking algorithms (for lack of a better way to put it) and basically just skinning the experience.



The image above is from the tracking settings on an Xbox, now the image is rather small but the skeleton is pretty much the same.

http://www.dotnetfunda.com/articles/art ... inect.aspx

The above article even goes into tracking the skeleton from a Kinect, I don't understand the code one bit but the diagram at the top clearly shows the same points being used. I do respect the fact that the skeleton that is laid out in these solutions is probably the best it can offer but the minimal variation between them all led me to this conclusion.

So to kind of move the question forward, are the other pieces of software just all doing the same thing using pretty much the same code to drive it? Can the Kinect and its tracking even really be modified that much to get better results? And finally, other than the clean up processes that programs may include, are we looking at the same results packaged differently at different prices?

Thanks for all the input

Hi, ManyMotion

You are right in your conclusion - all the Kinect solutions above rely on the same skeletal tracking algorithm based on predefined poses matching. It is provided by Microsoft Kinect SDK or OpenNI libraries, depending on what drivers are used for working with Kinect. This algorithm is optimized for real time usage in game consoles and PC applications as UI input. And currently it is limited to have only single device as its source - that is why none of these solutions provide multiple Kinects support. Eventually, as Kinect SDK evolves, we may see some improvements in accuracy, this depends on Microsoft's vision on further development of Kinect as a platform.

In iPi Mocap Studio, we use a completely different approach of our own, which is optimized for motion capture from the beginning. And we continue to develop it with the goals of improving the accuracy, enlarging the capture space (more than 2 Kinects), and increasing the performance (hoping for real time some day).

More than 2 kinects, when?

I knew someone will pick on this words :) No estimates, just plans to do a research in this area later this year. More news when we get some results.

Thanks for the reply Vmaslov, that information really helped in the presentation I had to give today! :)

3 depth sensors are supported in Mocap Studio v2.5.0.158. Currently, with the purpose of eliminating occlusions rather than extending the capture area.

Whaaa? That's awesome! Can't wait to test this.

Does the additional sensor improve accuracy? Yes, I understand that less occlusion will improve tracking accuracy, but I'm specifically wondering if the data also reduces jitter and feet tracking?

Thanks in advance for any info and keep up the amazing work guys! :)

G.

_________________
Greenlaw
Artist/Partner - Little Green Dog | My Demo Reels (2013,) (2015,) (2017,) and (2019)

Watch a one minute excerpt on Vimeo now!

Can I put in my request for sample data again for 3 kinects? Still waiting on some sample 2 person capture data too... :)

You guys at ipiSoft must have some somewhere!