|
Class : Apprenti
Posts : 18
Registered on : 14/09/2009
Off line
|
Hello,
I would like to know technicly wich system is better, taking in consideration:
Field of View
Precision
Thanks!!
I'm expecting an answer based on technical arguments, I'll build one system for me and I would like to know wich one is the BEST... Thanks!!
|
|
Class : Habitué
Posts : 208
Registered on : 19/07/2008
Off line
|
In the help section (of free-track) you can find some arguments for choosing a clip or cap.
It is also a personal choice, I don't like racing with a cap, that's why I made a headphone solution for the cap model. I also made the clip model, but cannot say which one is the best.
With a cap-model your webcam /Wiimote is placed in the centre of your screen, with the clip it is off-centre.....
Point model
Point model should point towards the camera, tilt accordingly.
Model size
The ideal model size is dependent on the distance from the camera and the viewing angle of the camera lens. At the desired distance, the model should be about a third of the height and a third of the width of the camera frame, for the best compromise between rotation accuracy and translation range.
2dof vs 6dof in 2dof program
2dof only measures the image coordinate of the point in the camera frame and uses this to blindly guess rotation. It doesn't know anything about the real head orientation and location and cannot distinguish head rotation from translation. 6dof uses real world coordinates and is able to acccurately measure rotation and translation and scale for distance. So 6dof can be a benefit in 2dof programs being able to distinguish head translation for rotation despite sacrificing some precision. 6dof can also benefit from the zoom smoothing multiplier which adds extra smoothing when leaning forward.
Two degrees of freedom
Single point
A single point provides two degrees of freedom, typically Yaw and Pitch for hands-free mouse control and mouselook abilities. One downside is that head rotation and translation cannot be distinguished but it gives the most return for the least effort. It can also be more precise than other tracking methods if the tracked point moves across a large part of the camera frame. This can be achieved by placing the point far from the head pivot, like on the rim of a cap.
Mapping onto other axes, like X and Y may be useful for some 3D applications where view rotation is already controlled by other means such as mouse or joystick hat-switch. It allows the viewpoint to be translated smoothly in order to see around visual obstructions, common in flight simulators and first person environments.
Six degrees of freedom
FreeTrack staarted using 4p
One point defines a point, two points defines a line, three points defines a plane and four non-coplanar points defines a volume. At least four non-coplanar points are required to track rotation in 3D space for all orientations. However this can be reduced to three if the view is restricted, three points cannot adequately define rotation for all orientations, since there are two possible solutions that are mirrored about the camera plane. Three point is recommended, it is easier to make, easier to power. Four point may be useful for cases where the point model orientation needs to be tracked
The highest point is used as a reference for cap tracking and the second heighest for clip tracking.
Three point
Clip
The clip point model is typically mounted on either the left or right side of a pair of headphones, roughly opposite the ear. Some people have managed to mount it onto BlueTooth headsets. This is particularly convenient for gamers who wear headphones anyway, unlike a cap it is less likely to give you hat hair and a hot head.
Since the points cannot be identified, they are listed from top to bottom and the second point in the list is used as a reference. Consequently tracking can be confused when the points are close to being horizontal. The middle point is used for as the translation reference.
In order to get accurate translation, the offset of the point model from the centre of rotation of the head must be estimated. The offset needs to be adjusted until translation during rotation is minimised, if you change your hardware position you might need to tweak the offsets accordingly.
Vertical translation and rolling may be limited if the model appears too large in the frame.
Cap
Since the points cannot be identified, they are listed from left to right and the top most point is moved to the top of the list and used as a reference. Consequently tracking gets confused when other points move higher than the intended reference point. The topmost point is used for measuring translation.
the model has significant width so x-axis translation is limited if the model is large in the camera frame.
Middle point forward of other points
A model with a middle point positioned forward of the other points can be tracked correctly by making axis adjustments. The model dimensions cannot be changed to reflect the model shape because it would require a negative measurement which is not possible. Instead the Yaw and Pitch axes need to be inverted. The centre of rotation calibration would need to be adjusted because it uses the middle point as a reference.
This model configuration is somewhat inferior to the conventional configuration. When the head rotates, points further from the centre of rotation move more than those closer. In this case, instead of two points moving significantly, there is only one point which means less tracking information is available. Consequently rotation tracking accuracy is compromised.
Four point
Four point tracking is mostly for back compatibility because the model is more difficult to construct and power with no benefit in most cases. Unlike 3 point it doesn't have any degenerate orientations because the points don't belong to the same plane but this normally isn't a problem if the model orientation is
When using active LEDs, the extra LED compared with 3 point tracking means more power is required which can be difficult to satisfy when using batteries. The yawing range is limited by the distance between the points, when the middle point intersects another point, tracking is lost. Like three point cap tracking, the model has significant width so x-axis translation is limited if the model is large in the camera frame.
Since the points cannot be identified, they are listed from left to right and the top most point is moved to the top of the list and used as a reference. Consequently tracking gets confused when other points move higher than the intended reference point. The topmost point is used for measuring translation.
|