Hello, :)

I was really impressed by some videos of motion capture with the Ipisoft Express software + the Kinect camera. Some of you have had really good results at tracking your movements so i decided to try it.

I bought a Kinect camera just for that.

I have followed the step by step guide here :
http://ipisoft.com/en/wiki/index.php?ti ... ect_Sensor

I have managed to capture a video with IpiRecorder, I have managed to import the video in IpiSoft demo 1.2.0.119

I have followed the step by step guide until the step "Refit pose" which i have tried many times but the skeleton is never positionned correctly.
I'm going mad :P

See an example of a screenshot before pressing the "Refit pose" button :


And after the "Refit Pose" step :


Note that i have repositionned the skeleton as explained in the step by step guide, i have also tried to rotate each bone with a better orientation but the result is worse...
What do i have to do in order that the "Refit pose" step is really fitting the skeleton to the shuman shape of the video ?

Here is a .MoCap file i captured with common character moves. If one of you can try it and tell me how you manage to get it working.
http://www.megaupload.com/?d=CDJMCA3S

My pc specs are (from AIDA64) :
Ordinateur :
Type de système : ACPI x64-based PC
Système d'exploitation : Microsoft Windows 7 Home Premium
DirectX : DirectX 11.0

Carte mère :
Type de processeur : DualCore AMD Athlon II X2 215, 2700 MHz (13.5 x 200)
Nom de la carte mère : eMachines EL1350
Chipset de la carte mère : nVIDIA nForce 6100-430, AMD K10
Mémoire système : [ TRIAL VERSION ]
DIMM1: Samsung M378B5673FH0-CH9 : 2 Go DDR3-1333 DDR3 SDRAM
DIMM2: Samsung M378B2873FHS-CH9 : [ TRIAL VERSION ]
Type de BIOS : AMI (01/28/10)

Moniteur:
Carte vidéo : NVIDIA GeForce 6150SE nForce 430 (256 Mo)
Accélérateur 3D : nVIDIA nForce 6100-430

Multimédia:
Carte audio : Realtek ALC662 @ nVIDIA nForce 6100-430 (MCP61P) - High Definition Audio Controller

Stockage:
Contrôleur IDE : Contrôleur IDE standard double canal PCI
Contrôleur de stockage : Contrôleur ATA série NVIDIA nForce
Contrôleur de stockage : Contrôleur ATA série NVIDIA nForce
Disque dur : Generic- Compact Flash USB Device
Disque dur : Multiple Flash Reader USB Device
Disque dur : WDC WD64 00AAKS-22A7B SCSI Disk Device (596 Go)
Lecteur optique : HL-DT-ST DVDRAM GH60N SCSI CdRom Device

Thanks for your help, :)

Unfortunately, the .MoCap file can't be tested without the .avi file that goes along with it.

The issue you're having is very strange, though. I'm using a Kinect and have never seen that issue, with the legs swiveling during the Refit Pose step.

Until someone else comes along with more knowledge, see in the first picture you posted - how your feet are farther apart than the skeleton's? Try the T-Pose with your feet closer together - just like the default skeleton T-Pose.

Yes it is strange indeed. And i don't know what to do to make this step work.

Another test with the ankles closer :

Before pressing the "Refit pose" button :


After the "Refit Pose" step :


I have tried 10 times with no better results.

One question :
In the step by step guide, it is written :
"Move actor model to left or right to match roughly actor silhouette on video. Note that actor model can look smaller due to its position along axis of view. Don’t pay attention to it on this step. "

That's exactly what i do... Do i have to move the skeleton at the exact position forward or backward and then rotate each bone so it fits perfectly the Tpose of the video ?

I really want to make this work, if you have any suggestions they are welcome

Thanks :)

Well, I'll just go over what I do differently.

1) I don't turn off the "Show Skin" option. When I'm moving the dummy over the AVI, I only look at the whole actor and don't pay attention to the skeletal bones. You can click on "Show Skin" along the top to turn the skin back on, although for me it's on by default.

2) On the "Actor" tab, I adjust the size of the dummy to match myself. Again, I use the skin of the dummy, not the skeletal bones. You can save the "actor" so you don't have to resize him every time.

It really looks like your dummy actor is too tall, but it's hard to tell without the skin showing.

I'm going to try to reduce the actor height, so far i used my real height.

After you have resized the character, do you move the character downward in order to put his feet at the ground level ?

I've never had the dummy actor be off the ground. The program automatically recognizes the ground plane and I've never had it be wrong.

I also haven't paid attention to the numbers in the actor size tab. I just have the dummy's skin on and I adjust its size to silhouette my video properly.

Ok another test with my height - 1 which fits well the character of the video i think

Before pressing the "Refit pose" button :


After the "Refit Pose" step :


I have also tried with my real height -2 - 3 - 4 - 5 with no better results.

I really don't understand why it doesn't work properly. All the videos i have seen so far demonstrated an effortless "Refit Pose" step...



What do you mean by "I adjust its size to silhouette my video properly" ? How do you adjust the size ?
I only modify the actor height

Thanks,

Yeah, there's definitely something wrong that doesn't seem to be related to anything you're doing. Hopefully someone else will read this thread who has had a similar problem.

What I mean by "adjusting the silhouette" is, there are more sliders than just height. I think there are three other sliders. So, hold down the right mouse button and rotate around to the side of your character. Make sure the dummy actor is lined up, but also adjust the size (thickness) for torso, legs and arms.

However, this probably won't make any difference, as the silhouette looks lined up properly in your most recent post.

One more question if you have time :) : When you are capturing the video with the Kinect sensor, what kind of lighting do you use ?

An omnilight in your room ?

A specific directional light ?

The sunlight going through the window ?

Thanks,

With the Kinect, lighting doesn't matter because it shines its own infrared light. You can motion capture in the dark if you want. Apparently it doesn't get much interference from other infrared sources because using two Kinects is a possible new feature.

I've motion captured with natural light as well as multiple lamps on. It hasn't made any noticeable difference in the captures.