1. Window-> Settings/Preferences -> Preferences -> Settings:

   	Linear = centimeter (usually default)
   	Up Axis: Z
   	Angular: radians

2. Create->Polygon Primitives->Plane (now a plane appears, centered at the origin).

3. Beware: Some Maya sessions may be not be set to create planes in the z=0 plane by default. If your plane isn’t purely in the xy plane, create a new one, and use the “custom” box at the right of the menu entry for ->Plane so you can specify the Z-plane radio button.

4. Hit Ctr-A to view its attributes

5. Go to the PolyPlane tab, and adjust the settings for the overall width/height of the plane, as well as the number of subdivisions (should be same as # of squares in your pattern). While you're here, change the Texture setting to "Preserve Aspect Ratio."

6. Go to the pPlane settings tab and type into the Translate (x and y) fields amounts that will move the plane so it fits wholly in the positive_x and positive_y quadrant. That will be half of the pattern's width and height (in cm).

7. Go to the pPlaneShape tab and the settings for Mesh Component Display: here you'll want the backface culling to be off, since we'll be looking at the underside of the plane.

8. To make the pattern look pretty, you'll want to do:

   Window->Hypershade, pick Rendering->Materials, right-click in the empty space on the right, and Create->Materials->Lambert (left-click to put it down). Double-click the new icon to bring up this material's settings. You want to change the color to a special pattern, so click on the pattern-button to the right of the Color's slider. Pick "checker," and under the Placement tab, give it a RepeatUV equal to half of your #boxes_in_width and height.

   (Note, if the number of squares doesn't match up, use the bigger # for both U and V).

9. Now select the plane in the scene and right-click on the checker material's icon to "Assign Texture's Material to Selection."

10. You now have a calibration pattern which should be the same size and look as the one you used to calibrate the extrinsics of your scene. Create->Lights->Point Light will be needed at some point so your nice checker model isn't pitch black. Move it around after you perform the steps below to add the camera. So far, this was the easy part.

1.  Create->Camera

2. Window->Render Globals...

3. Change Image format to tif

4. Change Camera to camera1

5. Resolution: choose Custom, and set to 720 x 480

6. set Pixel Aspect Ratio to: (b/a) (hit Tab to see how this auto-changes the Device aspect ratio to 1.366).

7. Insure 'Channels' is set to "RGB Channel"

8. You'll probably want to set the Anti-aliasing Quality set to High, but it's not critical.

9. Select camera, and hit ctr-A (get attributes window)

10. In the cameraShape, set the Focal Length to a (the 1st or bigger focal length). (Angle of View just changed automatically!)

11. Leave Camera Scale and other fields not mentioned here at default.

12. Change your Near Clip Plane to 0.01 and your Far Clip Plane to 50000. Bigger doesn't hurt, and should be used if things are too far apart.

13. Camera Aperture for (x,y) were calculated above, so these should be 28.346inches and something near 20 (depending on your a and b).

14. Change the Film Fit to "Horizontal"

15. Change the Film Offset to:
    {720/2 - cc(1), (cc(2) - 480/2) } =
    (3.32736mm, -11.77465mm) =
    (0.1309984 inches, -0.463569 inches)

16. What did we just do? We moved the location of the film to match the non-centeredness of the real film/ccd w.r.t the actual lens.

17. Change Display Options to have ON: Display Film Gate and Resolution

18. Add Environment-> Image Plane: Create: Image Name = path to your home image. Image Plane should be "Attached to Camera" What just happened? This camera has a billboard showing your home-image mounted in front of it. If the camera moves, so does the image plane, though objects, like the calib-pattern object, can be placed between the two.

19. Optionally make "Display" set to "in all views."

20. Size = same as Camera Aperture

21. Offset = same as Film Offset

22. Fit: "To Size" and click on "Fit to Film Gate"

23. Note: this should allow you to render through Maya the same image as you had as input. If it’s somehow squished, try a Horizontal fit to match the camera settings.

    The depth of the Image Plane does not affect the size of the background image attached to your camera. However, placing it too close to the camera could mean that it occludes the plane sitting behind it, so type in a depth which accommodates your distance from the origin: less than 50000, but more than the largest translation component of your extrinsics.

24. In main Maya app, pick a window, and Panel->Perspective->Camera1

25. Note: If you only have one panel on the screen and want to have more, use Window->View Arrangement->Four

26. In the same window, pick Shading->"Smooth Shade All"

27. Also under Shading, switch on Hardware Texturing if possible.

28. Fine time to save your work.

29. You should still not be seeing much in the camera's panel. I like to peek at the state of the world by using a different panel and switching it to a perspective view, which I can move around:

30. Panels->Perspective->persp. Alt-Shift middle-mouse-button and Alt-shift left button allow you to translate and rotate the scene, which should currently contain a camera and a pattern both sitting at the origin.

31. If you've saved your work, you might want to play with translating and rotating the camera to make the CG "model" line-up with the image of the model. Alt-r and Alt-t reveal rotation and translation handles you can mouse around with to manipulate the camera or the model. Leave the model alone, but do your best by moving the camera. This is hard, and once you're done wasting your time, you'll proceed to the following instructions, because you'll realize that a little math is easier than ALOT of clicking and dragging.

32. It's important that you realize that what follows IS part of the assignment, but, strictly speaking, isn't necessary for rendering CG to go with your footage. Why? Your virtual camera now projects 3D objects in almost the same way as your real camera did (more later on why not 100%). That means the two cameras' intrinsics match. Since the camera didn't move, you could (theoretically) start inserting CG flying elephants and they would enter and leave the field of view in the right fashion, assuming you modeled them to scale (using cm units as we specified earlier). This means that you could theoretically get by without extrinsic calibration at all. So who needs the external parameters of the camera (rotation and translation)? Doing a match-move certainly requires it, and actually needs the R and t for each frame of footage. But even when the camera is static, it's nice to do the following steps, because then you can line-up the ground-planes (or wall planes if that's where you put the calibration pattern) with the cardinal directions within Maya. You can texture and place new CG objects in the scene, and they should line-up in each camera for which you calculate the extrinsic parameters. That's more than just a little useful if you intend to replace walls or match up foot-falls. It's especially handy if you've calibrated multiple cameras and want them all in the same CG scene.

33. To remove ambiguity about rotation order pre/post multiplication, we're just going to enter our transformation matrices directly. Note that Maya matrices are transposed versions of the Foley et al. notation, so, for example, translation in the x directions would go in the bottom left of a 4x4 matrix.

34.  Window->General Editor->Script Editor

35. Note: type the following commands in the bottom panel, and use Edit->Execute or Ctrl-Enter to execute them. The help contains more on MEL-scripting, but all you'll need is given below.

36. Click-select the camera.

37. type:

    xform -m 1 0 0 0 0 1 0 0 0 0 1 0 0 0 0 1;

38. setAttr "camera1.rotateX" 3.1415926;

39. <execute the above calls> (-m stands for matrix)

40. What just happened? The camera was sent back to the origin, and back to pointing in the default direction: -Z. Our calibration was done with the model origin in the upper left and assumed the camera was looking toward +Z, so the second call just rotated the camera about the x axis to make that true for the virtual camera too: we're still sitting at (0,0,0), but facing +Z. Notice that the panel showing the view through the camera has color-axes in the lower left, and x and y are pointing to the right and down respectively. Now dust off the inverted version of the camera's extrinsic transformation matrix. Transpose it, and perform the following relative (hence the "-r") xform:

41. type (substituting your own matrix in place of my 4x4):

    xform -r -m    0.772569  0.634323 -0.0277706 0   -0.158799  0.235386  0.958841  0    0.614752 -0.73636   0.282582  0 -138.148   180.645  -100.554     1;

42. Save your work. You should be seeing a CG model (possibly with a checker-texture) sitting almost precisely on top of the part of the image where the real calibration model projected onto. If it's not perfect, don't get upset: the calibration optimized the camera extrinsics after modeling lens-distortion (kc in the intrinsic results). That's the one part of the calibration we can't get Maya to reproduce. The absence of lens-distortion means that your CG object may be SLIGHTLY off, but with the DV cameras we're using, won't explain gross mismatches, so at this point, you really should be seeing a rather good match.

43. Render this out (using Render->Render to New Window, and File-Save Image).

    

    Figure 2: Alignment is off because the extrinsics were estimated to include lens distortion, which is absent in this rendering.

44. Now select your camera, and make a 2^nd camera by doing Edit->Duplicate. This camera has all the same parameters (intrinsic and extrinsic) as the first, but you’ll reposition it to the other “home” camera position:

    	First clobber the extrinsics of camera2 (make sure #2 is the only one selected): xform -m 1 0 0 0  0 1 0 0 0 0 1 0 0 0 0 1;
    	setAttr "camera2.rotateX" 3.1415926;

45. Then apply the other set of extrinsics:

    	xform -r -m
    	(the 4x4 matrix which comes from doing an extrinsic calibration of your camera when it was placed at a second pose, looking at the same calib pattern)
    	Render this out also.

46. To get it perfect, render out a second pair of images after translating (ONLY!) very slightly in the x and y directions (that's the dimension affected by distortion). The hand-adjusted offset should be less than a cm for either axis. Render these images out also: both are part of your submission for this part of the assignment.

    

    Figure 3: Minor (< 1cm) hand adjustment of the camera position (not rotation!) has compensated for the lack of lens distortion. Lighting has also been adjusted slightly, and now the scene is ready for new CG elements/action.
     

47. Now model and render whatever you like, and render out a movie. Insert a simple CG object (like torus or cone) into the scene so it “belongs”: on floor or wall etc. Light and render to output a still/movie seen from both cameras, and submit it to the HW2 directory. For Maya help in general, hit F1 or look through the help-menu at the top right. The section on rendering (especially Ch.10) will help you through rendering out a sequence of images. Doing a search in the help on "background" will reveal a nice section which lays out how to composite and texture 3D objects on top of live-action images (like the home image) and image sequences (including movie files).