Hi all,
I just heard about OptiTrack system and found it quite impressive so far. Congratulations for everyone working on it!
The thing is, I need to develop a hand tracking system. And although there were many hand tracking focused solutions around, my particular case have some pretty though constraints and couldn't find a proper solution among them so far.
The main problem is I am tracking a small hand (5cm x 3cm) and need high resolution in this small space. I want to track all 22 degrees of freedom of a hand if it is possible after all. I do have a well defined fixed space where hand moves, so placing the cameras are quite straight forward.
My main question comes; according to my initial understanding the default tracking range starts from 0.6m which is quite a high value for me. Is it possible to fine tune this value lets say, around 0.1m order?
I was wondering if it is reasonable to expect from an OptiTrack system to track lets say 20 points in a 0.3m sided cube?
And what would be the spatial resolution of the tracking if its possible at all?
Thanks in advance.
Hand tracking
Re: Hand tracking
Unfortunately, 4 inches from the camera lens is going to be a little close.
The FOV from the camera itself is going to really limit the coverage area at that distance, and the camera overlap area would be non-existent at that range as well.
The FOV from the camera itself is going to really limit the coverage area at that distance, and the camera overlap area would be non-existent at that range as well.
Re: Hand tracking
So, is 0.6m really the absolute limit which a camera can be placed? I was hoping that playing with some other lenses may give a chance for me to benefit from OptiTrack.
According to my lame calculations; combining a SLIM:V100 camera with lens M12(1.9mm) should capture a 30cm sided cube completely when placed 10 cm away from one edge. what do you think, is this possible or am I simply making an error, which is quite possible, I don't know much about FOV calculations though.
I may push my psychical design limits a little actually but 0.6m seems still too much for me.
And again, assuming that I have managed to place the cameras far away enough, do you think the system can cope with distinguishing 22 points packed in a small space(hand) -with dimensions 5cm x 3cm x 3cm ?
According to my lame calculations; combining a SLIM:V100 camera with lens M12(1.9mm) should capture a 30cm sided cube completely when placed 10 cm away from one edge. what do you think, is this possible or am I simply making an error, which is quite possible, I don't know much about FOV calculations though.
I may push my psychical design limits a little actually but 0.6m seems still too much for me.
And again, assuming that I have managed to place the cameras far away enough, do you think the system can cope with distinguishing 22 points packed in a small space(hand) -with dimensions 5cm x 3cm x 3cm ?
Re: Hand tracking
Hello:
The problem with using the ultra wide FOV lenses is that it becomes very difficult to correct out the distortion from the various cameras.
This project is probably within the scope of our cameras, with very special considerations, but I don't know how you are going to track all the hand points and sort them. ARENA doesn't do hands and this is probably beyond the RB tools. You could try some SLIM cameras and track some points and see what happens and write your own code to sort the points. Just depends on how much time and effort can be put into the project.
The problem with using the ultra wide FOV lenses is that it becomes very difficult to correct out the distortion from the various cameras.
This project is probably within the scope of our cameras, with very special considerations, but I don't know how you are going to track all the hand points and sort them. ARENA doesn't do hands and this is probably beyond the RB tools. You could try some SLIM cameras and track some points and see what happens and write your own code to sort the points. Just depends on how much time and effort can be put into the project.
Re: Hand tracking
Thanks for your reply Jim,
Hardcore hand tracking research is not my main interest in my PhD project. But indeed a well functioning hand tracking module is a crucial element of it which means I can definitely put some time and effort in it, provided that I will get a solution at the end. I can not afford to spend too much time on something that I am not going to use finally though.
Although it seems obvious that it is not going to be an out of box solution in my case, OptiTrack system seems still quite promising to me due to some of its unique features, i.e. no need to gloves, disposable passive trackers, no magnetic interference, affordable price.. And I am seriously thinking about giving a try to it, but due to the experimental characteristics of the whole idea, I want to have an action plan beforehand for the possible problems. And I think I need to first clarify the points below.
- Can the distortion problem of wide FOV lenses cause Point Cloud Toolkit to malfunction? What can be the order of 3 dof position errors of individual markers in my 0.3m sided cube tracking volume.
- Again how hard is the issue of distinguishability in this case. Tracking ~20 markers packed in my small 3x5x5(cm) volume.
- I am assuming ARENA uses an internal skeletal connectivity information combined with time course evaluation of coordinates coming from the Point Cloud for identifying individual markers and sorting them. I guess the source code for this part is not available, so for a similar calculation of hand markers' sorting I need to write it from scratch. I can not foresee how big an effort is that at the moment, any comments are welcome.
- Besides using `soft` methods, is there any other way for identifying-sorting individual markers? What about active LEDs with different illumination frequencies or passive ones with different reflection characteristics for example.
thnx and cheers.
Hardcore hand tracking research is not my main interest in my PhD project. But indeed a well functioning hand tracking module is a crucial element of it which means I can definitely put some time and effort in it, provided that I will get a solution at the end. I can not afford to spend too much time on something that I am not going to use finally though.
Although it seems obvious that it is not going to be an out of box solution in my case, OptiTrack system seems still quite promising to me due to some of its unique features, i.e. no need to gloves, disposable passive trackers, no magnetic interference, affordable price.. And I am seriously thinking about giving a try to it, but due to the experimental characteristics of the whole idea, I want to have an action plan beforehand for the possible problems. And I think I need to first clarify the points below.
- Can the distortion problem of wide FOV lenses cause Point Cloud Toolkit to malfunction? What can be the order of 3 dof position errors of individual markers in my 0.3m sided cube tracking volume.
- Again how hard is the issue of distinguishability in this case. Tracking ~20 markers packed in my small 3x5x5(cm) volume.
- I am assuming ARENA uses an internal skeletal connectivity information combined with time course evaluation of coordinates coming from the Point Cloud for identifying individual markers and sorting them. I guess the source code for this part is not available, so for a similar calculation of hand markers' sorting I need to write it from scratch. I can not foresee how big an effort is that at the moment, any comments are welcome.
- Besides using `soft` methods, is there any other way for identifying-sorting individual markers? What about active LEDs with different illumination frequencies or passive ones with different reflection characteristics for example.
thnx and cheers.
Re: Hand tracking
Hello:
With such a small volume and wide FOV lenses, it will probably be very difficult to map out the distortion of the lenses, and we have never done it for this type of volume.
Tracking the markers will be very hard I think, because they will not be visible in many consecutive frames, and there is no hand skeleton model to help with tracking. Yes, you could sort them on your own, and that would be what we suggest, but I think that effort should not be underestimated.
I think using the passive markers is the best idea, as active markers will not buy you much with our system for sorting.
As you can tell from the tone of my post, I am hesitant to suggest going with our product, as I think it is a lot of work, and you might be better off with an inexpensive glove for finger tracking, and perhaps our system just to track the top of the hand for the initial reference plane.
With such a small volume and wide FOV lenses, it will probably be very difficult to map out the distortion of the lenses, and we have never done it for this type of volume.
Tracking the markers will be very hard I think, because they will not be visible in many consecutive frames, and there is no hand skeleton model to help with tracking. Yes, you could sort them on your own, and that would be what we suggest, but I think that effort should not be underestimated.
I think using the passive markers is the best idea, as active markers will not buy you much with our system for sorting.
As you can tell from the tone of my post, I am hesitant to suggest going with our product, as I think it is a lot of work, and you might be better off with an inexpensive glove for finger tracking, and perhaps our system just to track the top of the hand for the initial reference plane.