In this context, rangefinding is measuring the relative position of a series of points of a static surface from the rangefinder; position sensing gives a rapid time sequence of relative x,y,z coordinate information and additionally gives relative orientation.
A number of physical principles are exploited in doing rangefinding and position sensing. These include optical sensing and triangulation as done in rangefinder cameras, and contrast maximization of an autofocused image, as is done in a modern autofocus single lens reflex camera. Also ultrasound is used for rangefinding in the manner of traditional SONAR. Magnetic field generation and sensing is exploited to produce position sensors. Both AC and DC field systems exist. Another type of optical rangefinder uses a laser and diffraction grating to shine a pattern on the object in question. Looking at the pattern's size and shape gives range information. Mechanical position sensors in which the user is connected to a kind of large extensible mechanical arm are also used.
Before discussing a few of the designs in a little more detail, we want to mention some considerations that are common to all such systems. Many of these concern whole body animation, but are all to one degree or another applicable to facial animation. A few definitions are also forthcoming. (A good review article on the subject of position sensing is [92].)
Before deciding on a position sensing or rangefinding system one needs to consider several things. The accuracy, resolution and registration are all important. Resolution is the smallest measurable change. Accuracy is the range in which a position is correct. Registration is the correspondence between reported and actual position and orientation. Accuracy, if it is relative to a previous position and orientation, does not imply registration since errors may accumulate. Other considerations are update rate - how often a measurement is taken and responsiveness - or how much lag there is between a movement and its reporting. There are also worries of how robust systems are in a real world environment. For example, magnetic trackers may be affected by metal or stray magnetic fields. The range of operation is also a consideration, though more for full body animation than for facial animation.
Systems may be classified into two basic types, orthogonal to the technology used in many cases. These are called inside-out, or outside-in. In an outside-in system the source is attached to the moving object and the sensors are fixed. An inside-out system is just the opposite. An inside-out system can be made to have better resolution by using lots of sources so that the sensors are in close proximity no matter where the object moves. Some systems do not have either sources or sensors on the moving object and thus do not fall into the inside-out or outside-in categories. These include various laser ranging systems as well as optical rangefinders that work by using multiple images and finding common points (and solving a simple trigonometry problem) to find range and the contrast maximizing rangefinders.
Direct optical rangefinding systems must reliably find common physical points from the two separate cameras. This is usually done using ``structured light'' in which a grid or a dot pattern of one or more colors is projected onto the face in order to make the problem of common point identification much simpler. A full pattern recognition on the face to find corresponding points with ordinary lighting is much more computationally difficult.
The laser rangefinding system made by Cyberware Inc. projects a vertical line on the face and uses this line for rangefinding. A separate camera is used to get color information. Either the object is rotated under computer control or the scanner moves around a fixed object in order to get complete three-dimensional information. A complete scan takes less than 30 seconds. We don't know of any facial animation application that uses contrast maximization, but this should be fairly simple to implement. It may be slow in that the scanner has to move in both x and y directions unlike the laser rangefinders.
Other optical systems and ultrasonic systems use multiple sources and sensors. The relationship among the sources and among the sensors is known. Triangulation allows one to calculate the 6 degrees of freedom required (x,y,z and pitch, roll and yaw angles). Some acoustic position trackers [86] in addition to the time of flight information used to triangulate also measure phase of the acoustic signal. However, acoustic systems are sometimes sensitive to environmental noise.
Magnetic systems generate 3 perpendicular fields. AC fields, such as used by the Polhemus Navigation Sciences [125] are the most commonly used. Three mutually perpendicular coils generate the fields. The induced currents are measured in sensor coils and the nutating fields allow the 6 degrees of freedom to be calculated. Eddy currents from metal in the field can be a problem with AC magnetic systems.
Ascension Technology [3] makes a series of DC systems called Bird, Big Bird and Flock of Birds. The DC system (pulsed DC) avoids the problems of eddy currents. DC systems are sensitive to ferrous metals, however and they must carefully subtract off the earth's magnetic field to operate successfully. (Fields produced by these systems are of about the same value as the earth's field.)