mocap's accuracy
Page 1 of 1

Author:  NicoDownBeat [ Wed Dec 19, 2018 2:32 am ]
Post subject:  mocap's accuracy

Hello everyone, is there a way estimating mocap's accuracy?
Mocap is a measurement to all effects, so how to estimate the uncertainty?

Thank you all

Author:  margarettteS [ Thu Dec 20, 2018 5:40 pm ]
Post subject:  Re: mocap's accuracy

Will it be possible? I haven't try to do it, I hope that somebody can shed some light on this problem.

Author:  Snapz [ Sat Dec 22, 2018 9:03 am ]
Post subject:  Re: mocap's accuracy


What would you be using the mocap for?

How much experience do you have with motion capture recording technology?

Accuracy is a little bit of a subjective term for mocap depending upon what you need, there are many videos on YouTube showing this program, although not many yet for the version 4 release, which to me is better than version 3 for use with my 6-cam PS Eyes set up, several improvements in tracking quality overall.

Mocap is a measurement to all effects, so how to estimate the uncertainty?

This iPi program is going to track the videos, or Kinect point cloud you record pretty well, so not much uncertainty, you can see what it is doing, as it isn't a real-time solution, although even before that, any system must be set up correctly for best results, that depends on the camera models used, as well as following the proper process for each system, iPi has some documentation for each to read through for basic start up and usage, but nothing is as good as experience when using any product and the better the machine used for tracking, the faster and better the result will be, especially with a PS Eyes system.

If using this for medical, athletic, or gate of human motion analysis diagnostics, the PS Eyes system will track that fairly well, the Kinect sensor not as well, but there is more to use of this system than just pressing the track forward button and expecting a perfect outcome, although zero mocap tracking systems are perfect, or to use your term "Accurate", a lot more relies on the users, performers and set up prior to any recording and tracking performance, than the tracking itself.

Do yourself a favor, if using the depth sensors, use them in a dual configuration, a single depth sensor will not perform as well, read the iPi docs for requirements first.

A user must know how to set up and correctly use any system for it to perform at its best, so anyone's first attempt with any solution will probably not turn out that good anyway, most of the reviews of this software are only using the trial of the program and will show lesser quality results, but most are inaccurate in many ways and don't demonstrate the full achievable quality of this system, so take those with a grain of salt, look for more experienced user reviews, if using these to make judgement prior to actually using the program yourself.

It would also be up to the performance to convey the action better, such as human gate accuracy, versus animation, versus character applied to and depending on which camera system is used, which would more determine the quality of the accuracy... Kinect, less accuracy, but easier to set up and smaller volume area use 4 m x 4 m.... PS Eyes, more accurate, but more strict recording requirements, more complex to set up, better used in larger volumes up to 7 m x 7 m, by rule of thumb description of each system.

(Action cam use has been added in version 4, but I can not comment much on this system, you can read more about it in the documents for more information, but there is no real user feedback to go on, as of yet). A little more complex system, as it uses external cameras to record and then building them inside iPi for tracking use.

If you are trying to compare this to a marker based system with a higher frame rate recording, that is apples to oranges, even those high end systems are not 100% accurate either, but if you are trying to compare it to lower cost IMU sensor tracking, or even the Vive/Orion, it is much more accurate than those solutions overall, due to it will be tracking the actual performer on the video, not guessing at a location in space to lock a key and letting the software interpolate what isn't actually occurring.

A good thing with tracking an actual video, is you can see what you are tracking, if you didn't like the recorded performance, you don't have to track it and it is much faster to tell this from an actual video review first, because the system is going to track what it on screen, so if the performer is stiff and clumsy, the animation will look that way as well, so if you want to include that in the term of accuracy, then yes, it would be accurate to what it is tracking, though again, more to it than just that.

No sense tracking a poor performance for most reasons, unless just going to use the exported animation for pose to pose break down of key poses, where you would be changing the animation completely anyway, the system can be used in this respect as well to help speed up making key poses and even setting basic hand poses, which can later be transferred to a studio library in other 3D editing software, for quick access later.

One other good thing, is you can pre clean your tracking very well and fairly easy with iPi, right within the program itself, which gives a better post clean up operation, but again, there is a learning curve to any of this, you are not going to just plug things in and expect it to just be perfect with the click of a button, that is a pipe dream for any system, you have to learn how to get good with any system, if you don't want to, then no system will be even close to "accurate".

Another plus is, you can import many characters into iPi studio and view the resulting animation right on a character for previz.

There are videos in the Videos Board Index of both camera systems used with iPi and the resulting animations, some exported and used in other 3D programs, some directly recorded from the iPi studio view port, have a look.

As far as determining the accuracy of the program, there are just too many caveats to that term, but the system works well with ease of use and to get the hang of fairly quickly, so, iPi is a good choice for motion tracking without an exorbitant costs of other systems for the Basic version, the Pro version comes at a higher cost, but for some use reasons that version is necessary, that would depend on the use case of the particular user.

I have been using the Basic system with 6-PS Eyes now for 5 years with very good results and the development has come a long way, the developers are responsive to features and to update the system regularly and making it better, so no one can go wrong with using this system, it just depends as much on the user, as it does on any system used and no one but the user can determine if any system will work for their needs until it is used properly to get the full performance achievable.

This software solution is solid for what it was designed to do, and to be a very good lower cost option to mimic the higher end systems and does a good job at it, in my opinion.

Just some of my thoughts on this question, take them as you see fit and good luck with your choice for mocap.


Author:  NicoDownBeat [ Tue Jan 08, 2019 4:16 am ]
Post subject:  Re: mocap's accuracy

Thank you for the answers, specially to @Snapz.
I am using ipisoft for biomechanical studies.

I'm going to solve this point with Arduino. In few words:
I am going to compare the Arduino's results (like angles and positions of the arm, for example) with the bvh file obtained with ipisoft studio.
Since Arduino's accuracy is known and I think is better then ipisoft, I can use Arduino like a sample measuring instrument.
We'll see.....

Author:  Snapz [ Tue Jan 08, 2019 8:57 am ]
Post subject:  Re: mocap's accuracy


It would be interesting to know the accuracy, or differences between the two systems for this purpose, maybe you could post your comparisons when you have finished.

Both systems will still require an accurate skeleton scale, with correct actor height, hip, arm and leg lengths to be set to closely match the subject recorded.

iPi tracks an actual video recording from several angles, where and IMU makes a best guess as to world position compared to it's calibrated position, so I don't know how an IMU can be claimed as known accuracy, it may have a shown accuracy by the software, but unless overlaid on a video of the performance, how would you be able to measure actual accuracy?

This has been done with several other IMU tracking systems and the results were not so accurate to a reference video recorded, so it would be interesting to see your findings.

I don't believe the full processing of the auto cleaning process from iPi, using improper Jitter Removal settings, or excessive Trajectory Filtering, would bare the same accuracy, as those processing steps do slightly alter the motions performed, and can smooth them too much, though for character animations this processing works well, so just something to keep in mind for your purpose, though I think if the Refining processes in iPi are run optimally, the angles and trajectories of any body part wouldn't be affected severely enough to not be close to the actual angles from a correctly set up and calibrated PS Eyes cam configuration, there is just many more caveats to any optical systems set up, than there would be for any IMU recording, but you don't have the magnetic interference issues with optical systems.

iPi uses a form of IMU sensor when using the PS Move controllers and the data accuracy from those is fairly consistent to the video recording when the hand and head bones are correctly set, though the data only uses the rotational values, not the positional values of the bones they are attached to, so for correct head, or wrist rotational tracking, these controllers would have to be used, even if the head tracking was used during the tracking process for the head, as it is close, but not as accurate as when a head mounted Move controller is used.

Good luck in your tests and I hope you will post some results.


Page 1 of 1 All times are UTC - 8 hours [ DST ]
Powered by phpBB® Forum Software © phpBB Group