tracking summary in short !!!!

[adolf]

hi
herez a brief discription on how trackers work!

2D TRACKER

1)1point- PAN or TILT shot...
when there is only horizontal or vertical movement in a shot. from point a to point b in x n y axis !
2)2point- ZOOM and ROTATION
when your shot scales or rotates u need to track 2 points in space to calculate the zoom and rotation changes in the shot.
3)4points- CORNER PINNING
when u need to track a surface to a plane in perspective
u corner pin the four suraface edge points and track them.

3D TRACKER-

1)NODAL PAN- it gives u a 2.5d tracking solution to pan shots with a very little parallax bcoz the matchmove considers the shot as good as a 2d shot n hence thinks that all the objects are at a same z space n as a result it gives u a point cloud with a spherical plane due to the rotation of camera.......
2)FREE MOVE SHOTS- such shots include parallax in them
ur matchmove engine will solve them on the basis of the filmback, focal length, and the survey data u provide it. these are generally complex shots. u need to do an automatic tracking first (manual optional), the track engine will generate locators in 3d space and give u depth.
locators are track markers or reference track points u analyse. then after tracking u need to select the locators and calibrate them, refine them and define a coordinate system with origin and give 2 axis reference to the engine.
1 origin point and distance between 2points in x axis and y axis.
after your coordinates are set u define relations between points as stated earlier.
example points that share same x , y axis. then export it to an animation package and test by making checkerboard plane for each of the xyz axis!
if u get it right here then u can do set layout and finally composite it in a compositing package!

here are some oints i would like to highlight..
i know its pretty basic but it will help out beginners n others who are confused in this tracking logic

1)focal length n camera aperture together gives us the field of view
camera aperture is the opening in the camera measured in inches
focal length is the opening in the lens measured in mm
it is also te distance from the optical centre of the lens to the sensor...
together they give u a field of view measured in degrees...

2)parallax ( zoom shots ) - it is the expansion n contraction f the objects in a relationship
check out the fg n bg objects ... notice how they change in relation to each other
nodal pan - ( pan shot ) - there is no parallax as none of the objects are expanding or contracting

3) to do a successful 3d track u need at least 9 points.. on 3 planes....

4)coordinates - it is used to tell the camera the orientation of the scene
relations - it defines the relationship between track points

5)tracking problems

please see that there are at least 9 trk points visible
use a 3d track point instead of 2d track point in 3d track
+ sign tracker works well as u can track the corners of the +
a circle changes shape as it rotates or defocus
triangles works well for a 3d as it froms a triangulation
try to increade the contrast of the footage before tracking
increase the size of the search n reference areas in case of moton blur
track the plate from the frame it is in focus incase if there it shifts focus... u may have to do forward n backward tracking
track in parts for occlussion.. then join the track curve from various trackers

6)automatic tracking
please do mask out the fg or bg object you dont wish to track ..
u can import image based mask from other applications
predictions show the 3d position of the 2d features that were used to calculate the camera
yellow points are 2d features visible in current frame
cyan are the features not found in the current frame
predictions exist for the entire length of the sequence even when the features they represent are out of shot
bojou has a y up coordinate system...

7) target tracking
when something goes out of the automatic trackers reach then u go for target tracking
or if u have sufficient survey information then go for target tracks

8) on set information -
please shoot grid for distortion reference ( optics compensation ls )
please place the track markers at places where it will not overlap the foreground object otherwise it will take days n days to
mask out the tracking markers
there should b at least 9 track markers visible
there should b three sides ( planes ) with 3 trackmarkers for correct tracking...
putting the wrong type of trackmarks at wrong places will make a shot impossible to composite even if
it might appear very simple
for shots requiring 3d track you must place or arrange the track markers in 3d space that is some before the
artist n some after the artist in various depths n sizes.. half cut tennis balls are preferable for outdoor tracking
please measure the distance between 2 track markers horizontally n vertically
please take care of tracking when the plates are being defocused n rack defocus

9) camera info

film back n focal length of each vfx shot
the distance of the camera from the subject
the lens aperture, the shutter angle n shutter speed, camera angle, height of the camera from ground
the type of camera used
any special type of lens, rigs, steadicams
the lens used
filters used if any
the exposure n the iso value of the camera
tilt angle
put a 3d cube or any object in scene but plese measure the size of that object
camera model
camera mount
film stock

10)lighting

please tell us the light direction
you can actually take pics of all the lights so that the compositor will know the light direction
please mention the type of lights that is hardlight, softlight or diffused fill
there should b no shadows on screen

11)planar tracking is very different from point tracking
mocha monet do planar tracking
here u use mask to calculate the perspective n the position of any planar object like screens or banners or billboards or mobile screens
u have to draw splines n then track them ... later u can do adjust ....

so planar tacker is used to track planes n surfaces whereas point tracker is used to track objects n track the scene....

PLEASE DO FEEL FREE TO CORRECT ME
I M PROVIDING THIS INFO ON THE BASIS OF SOME RnD I DID FOR YEARS n thanks to fxphd and MR. BEHRAM PATEL, who taught me .... he is currently woking for pixion studios, mumbai and he is one of indias best faculty for vfx....
to b very honest three years ago i didnt knew a thing about vfx....
n today i supervise vfx shoots...
thanks fxphd....

for some this might b pretty basic but for others it might help... those who want to know basics of tracking....
adolf
yash_2001us@yahoo.com

[blafla]

If you permit, I would precise the 2D tracker.
2 points for the evaluation of a zoom is amho too risky. It's only valid if the object is perfectly aligned in front of the camera and in the right axe, and in the case of a digital zoom, not a physical one.
For beginners, I would advise to use 4 points to track an object in case of physical zoom, just for "security" reasons. So I tell to my students which mostly work on real shots, not digital/computer-made ones.

[milkey]

Clarification on #2
The definition of "parallax" would be like this: "The apparent displacement or difference of orientation of an observed object along two different lines of sight". In other words parallax is the distortion of orientation between two observation points. So this means parallax = motion in space.
In static zooming shots the observer experiences a different phenomenon called "perspective distortion". It's definition would be: "The warping or transformation of an observed object at different angles of view. (or FOV)".
So static zoom shots do not exhibit parallax but perspective distortion.

The dolly zoom or "vertigo shot" (Alfred Hitchcock 1958) shows both phenomenons at work. Perspective distortion varies the size of the subject and scene but by modifying the distance of the camera (thus introducing parallax) the perspective distortion effect is highly increased and keeping the subject at the same size in frame.

[milkey]

"to do a successful 3d track u need at least 9 points.. on 3 planes...."

It is not a rule. It's more like a guiding rule or best practice rule.
Depends on how many variables of the camera you have to solve. In theory the algorithm would need only 3 points (thus describing 3 planes) with known coordinates in space for the case where camera variables are known. If those 3 points do not have known coordinates you will need parallax to solve them but for an accurate solve the algorithms would need 6 points (it needs to evaluate distance between them). The number of points has to be increased when the intersection angle of all frame projections (trackers in each frame) is low. Also adding zoom to all those variables will increase the tracker number exponentially.