iPi Mocap Studio Kinect v2 requirements

Snapz · **Posted:** Fri Nov 27, 2015 7:27 pm

...

Yes, that camera positioning on calibrating, (with the board anyway), happens a lot, I found just by re-clicking the calibrate button again, sometimes took several re-tries, they would flip to correct positions on-screen, not sure why this happens, but after you should get a PERFECT result, with actor point clouds matching very close and tight.

(I got tired of re-recording the calibration video, and even doing so it didn't really help, but the re-clicking button trick did eventually after a few tries).

Also on the frame drop on the slave, I saw this dropped a bit also, until it was connected to the Master, then the frame rate stabilized at 30 fps, but I don't think even 29+ fps should be an over-all issue.

Also you can try the minimizing the Slaves recorder window to the task bar, see if that helps, you don't need to see that comp while recording anyway.

...

Greenlaw · **Posted:** Fri Nov 27, 2015 8:03 pm

Okay, thanks for the feedback.

Originally, the UI allowed you to choose whether the sensors were in 90- or 180-degree configs and this was never a problem back then. The problem first appeared when the ability to auto-sense the positions was added. That was a couple of years ago though and the error was promptly fixed by the developers (for Kinect v1 of course.)

I'm calling this a bug--it really shouldn't be happening.

G.

Greenlaw · **Posted:** Fri Nov 27, 2015 9:07 pm

Nope, I just can't get Mocap Studio to position the sensors properly with that calibration data. The sensors are always placed on top of each other.

I'm going to try a new calibration recording later this evening.

I would like to comment on the new distributed recording system. I have to confess, I was a little skeptical at first but having tried it now, I think it's brilliant. I thought a big drawback would be in waiting for the footage to download and get merged but after each recording, the data was instantly merged at the Master computer and waiting to be tracked. Amazing!

Note: I have an ethernet connection to the tablet using a USB 3.0 ethernet adapter. Having this adapter attached doesn't seem to interfere with the performance of the KinectOne sensor--the sensor still maintained a steady 30 fps. I haven't tried distributed recording with wifi yet but my feeling is that you won't see anywhere near this kind of performance with it. I'll give it a shot later just to see what the difference is like.

Greenlaw · **Posted:** Sat Nov 28, 2015 12:33 pm

I got some helpful advice from from Pavel about calibration.

When I was using triple Kinect (v1), I had gone back to using a Maglite for calibration as required, which worked really well, but I wasn't able to do this with the recent dual KinectOne setup because I had to disable video to sustain a steady 30 fps on my workstation. But I think the workstation will be able to sustain 30fps with video during actual recording so I'm going to give it another shot.

If I am to rely on the board tracking, he suggested letting the calibration video run longer, cover more of the performance space with the board, and re-position the sensors.

He said the calibration accuracy with either method should be comparable when done correctly. I personally liked the board method but I'll try both today, assuming I can record with video enabled.

I was wondering, does anybody have suggestions for a really cheap laptop, mini-desktop or tablet that can support KinectOne for iPi Recorder? The Wacom Cintiq Companion tablet I have now works perfectly with KinectOne but I can't possibly afford a second one. I was thinking something in the $250 - $400 range, or cheaper if that's even reasonable. The important thing is that the computer is certain to work with KinectOne, iPi Recorder, and networkable.

If a reasonable cost for this computer has to be higher than $400, I may as well wait until I've saved up for a better workstation, which could be a year from now. That's okay too but I want to explore all my options.

Thanks in advance for any info.

G.

Greenlaw · **Posted:** Sat Nov 28, 2015 2:08 pm

I tried another series of dual Kinect One calibration tests this morning, using both a light and a board. I only had time for calibration tests but it was successful and informative.

The light tracking calibration failed because the room is simply too bright during the day even when our blinds are closed. However, I did learn that the old workstation is capable of recording both depth and rgb sufficiently at 30fps. Only one rgb frame was dropped out of three separate tests, and depth data was perfectly recorded in each test. No frames at all were dropped form the tablet's recording. That's pretty good.

I should point out that I was using the PS Move controller as a light source. Normally I use a Maglite, which probably would have worked better but maybe just barely. The room is simply too bright during the day, which why I normally record mocap only at night.

The board calibration ran perfectly! This time I covered a much broader area and let the recording run longer than usual. Normally, I record about 700 frames or a little over 20 seconds and use a 400 frame segment of that. This time I let it run for about 1500 frame or roughly 45 seconds and calibrated the whole range. This took a bit longer to calibrate than expected but I guess there's quite a bit more data there than with the Kinect v1 data. Anyway, the results were perfect so that's all that matters. I still think I can make calibration work with about half the frames but for now I was pleased to see this feature working as expected.

I'm going to continue with dual Kinect One motion tests tonight and perhaps start recording new motions for our short film using the new setup.

Also looking into some cheap laptops. It sounds like an i3 processor may be adequate so long as other requirements are met. I'll probably still want at least an i5 processor, but this does narrows down the choices in my target price range. Maybe I can take advantage of one of the 'Black Friday' sales going on now. :)

G.

Snapz · **Posted:** Mon Nov 30, 2015 6:38 am

...

If you're just wanting the comp to only handle iPi Recorder and do no other running of other programs, or need many features, I would suggest today on eBay, or other seller, for a refurbished laptop 15 in. with a minimum i3 for the hyper-threading. (Although you won't find one cheap with an SSD probably).

Cyber Monday today, you may get a very good deal from a factory refurbisher, some offer 1 yr warranty included, I am sure you are looking.

When I used the board for calibration with dual Kv2, I usually ran about 1000+ up to 2000 frames many times and the cameras still were set wrong, it's just whatever the program gives you is what you get to work with, you may get some to go perfect, others won't, maybe you can offer your tips to get it to be consistent on every reasonable calibration attempt where you know you did it fine and should have worked.

If room is too lighted, can you switch camera to dark or extra dark video view setting when trying the light probe technique, I can't remember if that setting is available for Kinects using video.

Good to read you are finally making some dual Kv2 progress now though.

...

Greenlaw · **Posted:** Mon Nov 30, 2015 9:28 am

Luckily I found a brand new HP convertible laptop/touch tablet with Intel i5, 8GB RAM, USB 3.0, DirectX 11 support, 1GB Ethernet and SSD for about $540 this weekend. That's actually a bit more than I really wanted to spend but it seemed like too good a deal to pass up. I was able to justify the purchase because my daughter really needs a new laptop and she's been wanting to become more involved with the mocap too. Hopefully, the laptop will do what I need it to do with Kinect One but if not, well, at least my daughter is getting the new laptop she needs.

If the laptop works out, I may try to save up to get another one but it will have to wait till after the holidays--I've already gone past my personal project budget for the year a couple of months ago. I have a feeling that deal was just for this past weekend but I'm sure there will be other deals later.

I ran a few more calibration tests last night using the board and I'm getting perfect results every time. The main difference I'm doing is that I'm covering a much broader area and rotating the board a little more than usual to make sure Mocap Studio can differentiate the sensor locations. I noticed that when I bring the board close to a sensor, it may become clipped from its FOV but Mocap Studio still tracks it because the other sensor can still see it--I think this 'error' especially informs the software that the sensors are in two very different locations.

The board I'm using it my usual 30" x 40" sheet of foam core that I used to use with my Kinect v1 sensors. I think it's important to have a fairly large board and foam core is rigid, light weight, and pretty cheap. (About $4 - $5 a sheet.) Shooting a range of about 1000 frames but I think I get get away with tracking fewer frames--with multi-Kinect v1, I typically used anywhere from 300 to 600. A 1000 frames is actually fine--it takes about 3 minutes to track with the new graphics card but I'll experiment with lower a number of frames tonight.

I'm guessing that recording at night is helping too. During the day, I get a quite a bit of IR pollution from the sun because the room is so bright, even with blinds closed. The computer room and adjoined living room are small but between the two of them, there are nine windows and the blinds are not meant for blocking light. It's nice space if you like a lot of light but not so good for mocap recording during the day.

Anyway, with Kinect (v1), I could only reliably record at night. During the daytime, the results could be hit or miss unless I put heavy blankets over some of the windows. That said, the Kinect One (v2) sensors do seem better able to filter out the pollution--I think it's still not reliable enough for daytime use in this room though so will probably only use it at night.

When I was storing the gear away this morning, I realized there was a feature I miss from when using Kinect (v1). While it's nice to have a tripod mount built right into Kinect One (v2) sensors, the tripod adapters I was using with the v1 sensors were easily detachable, like the quick release plate on a good tripod head. Since I'm attaching to v2 sensors directly to med-duty light stands, I no longer have this 'quick-release' feature. Sigh! I guess you can't have everything, can you. :)

Anyway, yes, it's pretty exciting to see this finally coming together. I'll post more info when my daughter's laptop arrives and I'm able to run a triple-Kinect One test.

G.

Snapz · **Posted:** Mon Nov 30, 2015 10:20 am

...

Awesome, sounds like it's working out now, yeah $200 to $300 on a new laptop would have been quite a feat anyway for an i5 and SSD.

Maybe this recent update helped a bit with the calibration by board, but as long as it's working for you, that's the ticket.

Can't wait to see some of your animations with them :)

...

Greenlaw · **Posted:** Mon Nov 30, 2015 11:06 am

Yes, I found a few good i5 adn i3 candidates for under $400 but with only 4GB RAM and hard drive, no SSD. I think that would have been adequate but I didn't want to take a chance going too cheaply. Already been burned twice: once with the cheap USB 3.0 cards and again with the cheap cables.

Greenlaw · **Posted:** Sun Dec 06, 2015 3:21 pm

I meant to mention this a week ago but I tried out my daughter's new convertible tablet with a Kinect v2.0. This computer is an i5 dual core with 8GB RAM and USB 3.0. It's graphics are the standard Intel chipset and DirectX 11+ compatible. I got this computer mainly because it was a good deal for the price and it seemed to meet the Kinect 2.0 specs--oh, yeah, and my daughter needed a newer computer too. :p

The Kinect works reasonably well--not nearly as well as is does with my Wacom Compantion 2 but about the same as my old workstation. It can do 30 fps with the RGB disabled and it's just managing to do it--it anything else is going on, it can start dropping frames.

For my purpose, her computer is powerful enough to use for testing the triple Kinect One setup. However, if I need to use a triple Kinect One setup routinely, I think I'd like to go with an i7 tablet or laptop like the Wacom. (I'm just not in the position to spend that kind of money right now.)

Anyway, I hope to do a triple Kinect One setup very soon. I'm guessing I'm not going to see a huge difference in quality over a dual Kinect One setup but it will be interesting to try a multi-actor capture with the triple setup--that's probably the one thing I haven't explored with Mocap Studio yet.

G.

iPi Mocap Studio Kinect v2 requirements

Who is online