`Kramer
`
`III IIHIIII
`5,592,401
`Jan. 7, 1997
`
`USOO5592401A
`11) Patent Number:
`(45) Date of Patent:
`
`54 ACCURATE, RAPID, RELIABLE POSITION
`SENSING USING MULTIPLE SENSING
`TECHNOLOGES
`
`75) Inventor: James F. Kramer, Menlo Park, Calif.
`73) Assignee: Virtual Technologies, Inc., Palo Alto,
`Calif.
`
`(21) Appl. No.: 395,430
`22 Filed:
`Feb. 28, 1995
`(51) Int. Cl. ...................................... GO1B 21/00
`52 U.S. Cl. .......................... 364/550; 340/524; 340/619;
`340/686; 364/559; 364/449.1; 473/128;
`463/49
`58) Field of Search ..................................... 128/639, 640,
`128/642, 774, 779, 781, 782; 340/524,
`547-549, 551, 573, 619, 686, 853.8; 364/453,
`449, 459, 550, 559; 395/100, 152; 473/128;
`463/49
`
`(56)
`
`References Cited
`U.S. PATENT DOCUMENTS
`3,868,565 2/1975 Kuipers .............................. 324/2O7.26
`4,396,885 8/1983 Constant
`......... 324/208
`4,414,537 11/1983 Grimes ................................ 340/365 R
`4,444,205 4/1984 Jackson ................................... 128/782
`4,542,291
`9/1985 Zimmerman ...
`250,231 R
`4,986,280
`1/1991 Marcus et al. ..
`... 128/774
`5,047,952 9/1991 Kramer et al...
`. 3647513.5
`5,166,463 11/1992 Weber ....................................... 84f600
`5,184,009 2/1993 Wright et al. ...
`250/227.11
`5,280,265
`l/1994 Kramer et al. .......................... 338/210
`FOREIGN PATENT DOCUMENTS
`
`O6408 7/1993 WIPO.
`
`OTHER PUBLICATIONS
`Foxlin et al., "An Inertial Head-Orientation Tracker . . .”
`Virtual Reality Software & Technology, eds. Gurminder et
`al., World Scientific, Singapore, Aug. 1994.
`Raab et al., “Magnetic position and orientation tracking
`system,” IEEE Trans. on Aero. and Elect. Systems, vol.
`AES-15, No. 5, pp. 709–718, Sep., 1979.
`
`
`
`Polhemus tech. note, "Latency-3SPACEQ) MagnetrakG),
`Tracker, Digitizer, and Isotrak09, P.O. Box 560, colchester,
`VT, 05446.
`Polhemus 3SPACE(E) Fastrak0) and Isotrak?) II product data
`sheets, Oct. 1994 and Aug. 1993 respectively, P.O. Box 560,
`colchester, VT, 05446.
`Ascension Technology Corp. data sheet #ATC 3-92, P.O.
`Box 527, Burlington, VT, 05402.
`Ascension Technology Corp. tech. note, "Comparison of
`specifications: Ascension Bird vs. Polhemus Isotrak," P.O.
`Box 527, Burlington, VT, 05402, pub. data unknown.
`Ascension Technology Corp. advertisement, "From head to
`toe, we cover all the motions,” CyberEdge Journal, p. 9,
`Nov./Dec. 1994.
`(List continued on next page.)
`Primary Examiner-Emanuel T. Voeltz
`Assistant Examiner-Hien Vo
`57
`ABSTRACT
`In accordance with the subject invention, devices and meth
`ods are provided for the accurate reporting of movement of
`an entity. Sensors which are accurate, but provide a delayed
`signal ("delayed signal sensors'), which delay is unaccept
`able for many applications, may be used in conjunction with
`fast sensors ("fast signal sensors'), which are usually subject
`to drift and other inaccuracies in providing information
`about a position. Additional sensors which may provide
`even more accurate and/or less signal sensor delay for a
`period of time, but which sensor signal is subject to periods
`of interrupted or undesirable output thereby making it unre
`liable ("unreliable signal sensors') may also be used in
`combination with one or more of the delayed signal sensors
`and fast signal sensors. By using a combination of such
`sensors, accurate, reliable position information is rapidly
`obtained to allow high-resolution and/or real-time analysis
`and depictions of movement. Complex rapid movements
`associated with athletics, music, and other rapid activities
`can be monitored in real-time to provide accurate informa
`tion during the movement. Such information may then be
`used to analyze the movement, for instruction for improve
`ment, for repeating the activity in a virtual setting and the
`like.
`
`31 Claims, 5 Drawing Sheets
`
`META 1030
`META V. THALES
`
`
`
`5,592,401
`Page 2
`
`OTHER PUBLICATIONS
`Kistler Instrument Corp., "Advanced instrumentation for a
`world of applications,' 75 John Glenn Dr., Amherst, NY,
`14228–2171, Gen. Catalog K2.006, 3rd Ed, Jul. 1993.
`Analog Devices, "Accelerometer data sheets and application
`notes,” P.O. Box 9106, Norwood, MA 02062-9106, Apr.
`1994.
`Gyration, Inc., "Gyration open loop directional gyroengine,
`model GE9300C,” 12930 Saratoga Ave., Bldg. C, Saratoga,
`CA 94070, data sheet C-1003, Sep. 1994.
`BioVision, "State of the art motion capture,” 2882 Sand Hill
`Road, Suite 116, Menlo Park, CA 94025, Porduct brochure,
`Aug. 1994.
`Adaptive Optics Assoc., Inc., "Capture motion, release
`creativity,” 54 Cambridge Park Dr., Cambridge, MA
`02140-2308, Product brochure MT202-CM, Aug. 1994.
`
`Tekscan, Inc., "Tekscan corporate capabilities,” 451 D.
`Street, Boston, MA 02210,
`Interlink Electronics, "FSRTM integration guide & evalua
`tion parts catalog,' 546 Flynn Road, Camarillo, CA, 93012.
`Wavefront Technologies, "Synthetic characters wish to
`exist,' 530 E. Montecito Street, Santa Barbara, CA 931.03,
`X-IST product announcement, Aug. 1994.
`B. A. Marcus, W. Lucas, EXOS, Inc., P. J. Churchill, A.D.
`Little, Inc. "Human hand sensing for robotics and teleop
`erations,' Sensors Magazine, Nov. 1989.
`J. A. Hall, "The human interface in three dimensional
`computer art space." Master's Thesis, Massachusetts Insti
`tute of Technology, Oct. 1985.
`
`META 1030
`META V. THALES
`
`
`
`U.S. Patent
`
`Jan. 7, 1997
`
`Sheet 1 of 5
`
`5,592,401
`
`
`
`O 1 2 3 4 5 6 7 8 9 O
`d
`T
`t
`
`alalauala-le
`
`st-e
`
`2
`
`F. G.
`
`META 1030
`META V. THALES
`
`
`
`U.S. Patent
`
`Jan. 7, 1997
`
`Sheet 2 of 5
`
`5,592,401
`
`252
`
`25O
`
`266
`
`
`
`META 1030
`META V. THALES
`
`
`
`U.S. Patent
`
`Jan. 7, 1997
`
`Sheet 3 of 5
`
`5,592,401
`
`
`
`META 1030
`META V. THALES
`
`
`
`U.S. Patent
`
`Jan. 7, 1997
`
`Sheet 4 of 5
`
`5,592,401
`
`
`
`META 1030
`META V. THALES
`
`
`
`U.S. Patent
`
`Jan. 7, 1997
`
`Sheet 5 of 5
`
`5,592,401
`
`
`
`F. G.7
`
`META 1030
`META V. THALES
`
`
`
`5,592,401
`
`1.
`ACCURATE, RAPID, RELIABLE POSITION
`SENSING USING MULTIPLE SENSING
`TECHNOLOGES
`
`TECHNICAL FIELD
`The field of this invention is rapid, accurate determination
`of position of an entity.
`
`BACKGROUND
`Recently there has been interest in motion data capture for
`use in such fields as graphical animation for movie produc
`tion and virtual reality. Such motion data is also of interest
`to sports trainers and athletes to analyze athletic perfor
`mance. There is also interest by the medical community, for
`example, in gait analysis. There are two main technologies
`presently being used to capture the motion data: (1) optical
`tracking; and (2) electromagnetic tracking.
`Optical tracking can provide a high sample rate which is
`required to capture rapid movement with high resolution.
`One disadvantage of optical tracking is the potential for
`obscuring the light sources to the degree that no position
`solution can be determined. Optical tracking is primarily
`accomplished using triangulation and multiple light sources,
`which often requires time-consuming calculations to resolve
`multiple or visually obscured light sources and detectors.
`Therefore, the raw motion data is often recorded real-time
`but analyzed and positions computed off-line at a later time.
`Electromagnetic tracking technology does not suffer from
`the problem of obscured sensors in a metal-free environ
`ment. Determination of position and orientation from the
`electromagnetic sensors can typically be performed in near
`real-time, however, with just enough delay (typically 4-50
`ms) to provide unacceptable disparity between the physical
`and measured motion. The delay is largely due to the physics
`of the technology which requires the electromagnetic signal
`be transmitted, received, filtered and analyzed. Current
`devices provide a maximum sample rate of only about 120
`Hz, while a sampling rate of from about 200 to 300 Hz is
`required to have an acceptable data resolution for rapid
`movement. The delay has made the electromagnetic devices
`unacceptable in applications requiring the tracking of rapid
`movements in real-time, e.g. conversion of human motion
`to music, while the slow sampling rate means the electro
`magnetic devices provide data with unacceptable resolution
`for tracking rapid athletic motion, such as a golf, tennis or
`batting Swing.
`There is, therefore, a need for a system which does not
`suffer from the shortcomings of the present optical and
`electromagnetic tracking systems, while providing a high
`rate of orientation and position data in real-time.
`
`10
`
`15
`
`20
`
`25
`
`30
`
`35
`
`45
`
`50
`
`55
`
`RELEVANT LITERATURE
`1. F. H. Raab, E. B. Blood, T. O. Steiner, H. R. Jones,
`"Magnetic Position and orientation tracking system,'
`IEEE Trans. on Aero. and Elect. Systems, vol. AES-15,
`No. 5, pp. 709-718, September 1979.
`2.J. Kuipers, "Object tracking and orientation determination
`means, system, and process," U.S. Pat. No. 3,868,565,
`Feb. 25, 1975.
`3. Polhemus Technical Note, “Latency-3SPACEQ Magne
`65
`trak(E), Tracker, Digitizer, and IsotrakCE)," P.O. Box 560,
`Colchester, Vt., 05446, publication date unknown.
`
`60
`
`2
`4. Polhemus 3SPACE(E) FastrakG) and Isotrak0) II product
`data sheets, October, 1994 and August 1993 respectively,
`P.O. Box 560, Colchester, Vt., 05446.
`5. Ascension Technology Corporation data sheet #ATC
`3–92, P.O. Box 527, Burlington, Vt., 05402.
`6. Ascension Technology Corporation Technical note,
`"Comparison of specifications: Ascension Bird vs. Pol
`hemus Isotrak," P.O. Box 527, Burlington, Vt., 05402,
`publication date unknown.
`7. Ascension Technology Corporation advertisement, "From
`head to toe, we cover all the motions,” CyberEdge Jour
`nal, p. 9, November/December 1994.
`8. C. Constant, "Device applicable to direction finding for
`measuring the relative orientation of two bodies,” U.S.
`Pat. No. 4,396,885, Aug. 2, 1983.
`9. Kistler Instrument Corporation, "Advanced instrumenta
`tion for a world of applications,” 75 John Glenn Drive,
`Amherst, N.Y., 14228-2171, General Catalog K2.006, 3rd
`Edition, July 1993.
`10. Analog Devices, "Accelerometer data sheets and appli
`cation notes,' P.O. Box 9106, Norwood, Mass. 02062
`9106, April 1994.
`11. Gyration, Inc., "Gyration open loop directional
`gyroengine, model GE9300C,” 12930 Saratoga Ave.,
`Bldg. C, Saratoga, Calif. 94070, data sheet C-1003,
`September 1994.
`12. BioVision, "State of the art motion capture,” 2882 Sand
`Hill Road, Suite 116, Menlo Park, Calif. 94025, Product
`brochure, August 1994.
`13. Adaptive Optics Associates, Inc., "Capture motion,
`release creativity,"54 Cambridge Park Drive, Cambridge,
`Mass. 02140-2308, Product brochure MT202-CM,
`August 1994.
`14. Tekscan, Inc., "Tekscan corporate capabilities,” 451 D.
`Street, Boston, Mass. 02210.
`15. Interlink Electronics, "FSRTM integration guide & evalu
`ation parts catalog,' 546 Flynn Road, Camarillo, Calif.,
`93012.
`16. Wavefront Technologies, "Synthetic characters wish to
`exist,' 530 E. Montecito Street, Santa Barbara, Calif.
`93103, X-IST product announcement, August 1994.
`17. B. A. Marcus, W. Lucas, EXOS, Inc., P. J. Churchill, A.
`D. Little, Inc., "Human hand sensing for robotics and
`teleoperations,” Sensors Magazine, November 1989.
`18. J. F. Kramer, P. Lindener, W. R. George, "Communica
`tion system for deaf, deaf-blind, or non-vocal individuals
`using instrumented glove,” U.S. Pat. No. 5,047,952, Sep.
`10, 1991.
`20. J. F. Kramer, P. Lindener, W. R. George, "Strain-sensing
`goniometers, systems and recognition algorithms,' U.S.
`Pat. No. 5,280,265, Jan. 18, 1994.
`20. J. F. Kramer, "Determination of kinematically con
`strained multi-articulated structures,' International Appli
`cation No. PCT/US93/06408, Jul. 6, 1993,
`21. T. G. Zimmerman, "Optical flex sensor,” U.S. Pat. No.
`4,542,291, Sep. 17, 1985.
`22. J. A. Hall, "The human interface in three dimensional
`computer art space.” Master's Thesis, Massachusetts
`Institute of Technology, October 1985.
`
`SUMMARY OF THE INVENTION
`Apparatus and methods are provided for quickly, accu
`rately and/or reliably determining position of an entity by
`employing a combination of individual devices where the
`combination provides a position measurement which is
`superior to any of the individual devices comprising the
`
`META 1030
`META V. THALES
`
`
`
`5,592,401
`
`3
`combination when taken separately. In one embodiment, a
`fast inaccurate position sensing device (“fast device' or "fast
`position sensor') and a more accurate position sensing
`device which includes measurement latency or some other
`signal delay ("slow device' or "delayed signal sensor') are
`employed in combination. The position sensing process
`continuously takes samples from both the slow and fast
`devices. The sample rate of the fast device may be higher
`than the slow device. The subject invention correlates the
`dam from the two devices such that the speed advantage of
`the fast device is combined with the accuracy benefit of the
`slow device. Such benefits include real-time performance
`with reduced position sensor latency, and off-line perfor
`mance with increased position sensor sample resolution.
`Another embodiment provides a combination of a very
`accurate but unreliable device which may be fast or slow
`("unreliable device' or “unreliable position sensor”) with a
`moderately accurate but slow device ("slow device') and a
`device ("fast device') which is even less accurate but faster
`than the slow device. The subject invention correlates the
`data from the devices comprising the combination and
`produces a superior position estimate such that combination
`device will produce a signal value, nominally (a) as reliable
`as the most reliable signal value of the fast device and slow
`device; (b) as accurate as the most accurate signal value of
`the unreliable device and slow device; and (c) with as little
`sensor signal delay as the shortest delay of the unreliable
`device and the fast device.
`
`BRIEF DESCRIPTION OF THE DRAWINGS
`FIG. 1 is a curve illustrating the relationships between the
`position as determined by a fast device, a slow device, and
`the actual position of the entity;
`FIG. 2 is an elevated perspective view of a golf player
`wearing a device according to this invention;
`FIG. 3 is a diagrammatic view of a device on an arm
`comprising a fast bend sensor and a fast translational sensor;
`FIG. 4 is a diagrammatic view of a device on an arm
`comprising a slow electromagnetic sensor and a fast angular
`SensOr,
`FIG. 5 is a diagrammatic view of a device on an arm
`comprising a slow electromagnetic sensor, a fast angular
`sensor and an unreliable optical tracking sensor;
`FIG. 6 is a diagrammatic view of a device on a hand
`comprising a slow electromagnetic sensor, a fast angular
`sensor, an unreliable optical tracking sensor and fast bend
`Sensor,
`FIG. 7 shows the relationship of slow, fast and estimated
`sensor outputs for one type of estimator.
`
`4
`least delayed as the most proficient aspect of the two
`SCSOS
`In one embodiment, the apparatus comprises a combina
`tion of a "fast' device which provides for relatively fast,
`relatively less accurate or provides fewer degrees-of-free
`dom of information (i.e., less complete information) about
`the position of the entity as compared to a "slow' device,
`where the information about the position of an entity at a
`particular position in time is not available for use by the
`measurement system until at least one sampling instance
`after position information is available from a fast device, but
`where the slow device provides more accurate or more
`degrees-of-freedom of information (i.e., more complete
`information) about position than the fast device. For a fast
`device which provides inaccurate data, the inaccuracy is
`typically due to a sensor output which drifts slowly as
`compared to the sampling rate of the sensor, i.e., which does
`not change appreciably from sample to sample such that the
`output from the device remains relatively accurate over a
`few sampling instances. Often, sensor drift may be due to
`sensor noise, transverse sensitivity, temperature sensitivity
`and the like. Other fast devices might provide position data
`which is incomplete, meaning that one or more of the 6
`degrees of freedom (DOF) of the sensor location are not
`measured.
`In general, we want to determine the true position, P(t), of
`an entity, e.g., a body part, at an arbitrary time, t. We assume
`that the samples from the fast device are available immedi
`ately, i.e., without delay, but the position value produced is
`P(t)=P(t)-E(t), where P is the position value produced by
`the fast device, P is the true position desired and E is the
`error deviation from the true position, which error is prima
`rily due to an accumulating error such as sensor drift. The
`position values from the slow device, on the other hand, are
`assumed to be precisely P without error, but dt seconds old,
`i.e., P(t)=P(t-dt), where P is the position value produced by
`the slow device. We then correct or re-zero our current fast
`measurement by subtracting out an error term.
`FIG. 1 provides an example of the relationships between
`these parameters in the form of points and curves plotted
`versus time. As seen in the figure, the points corresponding
`to one DOF from the fast device, P(t), are aligned in time
`with true position trajectory points P(t), but are offset in
`value by error E(t). The data points from the slow device,
`P(t), are exactly equal to the true data points P(t), but are not
`available until dt seconds later. As illustrated in FIG. 1, is it
`assumed that the present sample time is t' and the sample
`period is T. (Note that the dashed portion of the curve and
`hollow sample points for P(t) represents future values
`which have yet to occur, and the dashed portion with hollow
`circles of P(t) represents the corresponding portion which
`must be estimated.) Since dt is actually a continuous time
`value, e.g., seconds, we define an integer sample time index
`as n=t/T and so the difference in integer sample time indices
`is dn=dt/T. Note that n is only valid when dt corresponds to
`a sample time, and dn is only valid when dt is an integer
`multiple of the sample period, T. FIG. 1 shows the exem
`plary case where T-1 second, dit=2 seconds and thus dn=2.
`dE(t) is defined as the error accumulated only over the last
`dt seconds, assuming that P(t)=P(t-dt), and dE(t)=E(t)-
`E(t-1) is the error accumulated only since the previous
`sample. Note that dE(t) is only determinable fort at least dt
`seconds ago.
`The error term, E(t), which if we knew we could add to
`P(t) to provide P(t), may be estimated in a variety of
`manners. From the estimate of E(t) and P(t) we produce an
`estimate of P(t), denoted P(t). In general, we let
`
`10
`
`15
`
`20
`
`25
`
`30
`
`35
`
`40
`
`45
`
`50
`
`55
`
`DESCRIPTION OF THE SPECIFIC
`EMBODIMENTS
`In accordance with the subject invention, apparatus and
`method are provided for determining the "true position' of
`an entity, e.g., a body part. By employing a combination of
`position-sensing technologies, position information from the
`combination provides a more reliable, accurate, and/or less
`60
`delayed measurement system than any of the devices com
`prising the combination when taken separately. By employ
`ing a pair of sensors, where each sensor has different
`deficiencies and/or proficiencies, the deficiency of a signal
`value from one sensor may be improved with the signal
`value of the other sensor. As a result of the combination of
`sensors, the signal value will be as accurate, as reliable and
`
`65
`
`META 1030
`META V. THALES
`
`
`
`5
`
`5,592,401
`
`6
`Dynamic Systems, by Franklin and Powell, Adaptive Filter
`ing Prediction and Control, by Goodwin and Sin, Random
`Signals: Detection, Estimation and Data Analysis, by Shan
`mugan and Breipohl, Linear Estimation by Arthur Gelb and
`Discrete-Time Control Systems by Ogata provide good ref
`erences for such estimation techniques.
`The slow device may also be reporting data at a slower
`data rate than the fast device. In such a case, the slow device
`sampling time determines atype of "frame rate,” i.e., the rate
`at which the more accurate data is being made available. In
`general, the slow data provided at each frame corresponds to
`a known previous inter-frame instance. Thus, when the fast
`sensor has a higher sampling rate, the fast sensor may also
`be used to fill in inter-frame position measurements where
`drift error in the fast sensor measurement has been accu
`mulating only since the known previous inter-frame instance
`corresponding to the most recent slow measurement.
`Various devices may be used as the fast and slow devices.
`In some instances a device will not have general applica
`bility as to the positions and degrees of freedom which can
`be measured, but may be used in combination with other
`devices which have general applicability (but have one or
`more of their own shortcomings). For the fast devices, one
`will wish to measure translational and/orangular movement
`rapidly. Present day available devices include accelerom
`eters, gyroscopes, and the like. Accelerometers are available
`from Kistler Instrument Corp., Amherst, N.Y. and Analog
`Devices, Norwood, Mass. Directional gyroscopes are avail
`able from Gyration Inc., San Jose, Calif., however, rate
`gyroscopes may also be employed. Other devices which find
`use include semiconductor-based orientation and transla
`tion-sensing devices. Accelerometers may be used for both
`translation and angular movement measurements. Gyro
`scopes measure the rotation of an entity, e.g., a body part
`about one or more axes. One may use one or a combination
`of accelerometers and gyroscopes, for example, for more
`complex joints (i.e., having a plurality of revolute intersect
`ing joints, as exemplified by the shoulder and thumb tra
`peziometacarpal joint).
`When accelerometers are employed, typical operation is
`as follows. At time zero, the accelerometer is at rest and its
`position known. From this point the accelerometer is
`allowed to move. By doubly integrating the signal from an
`accelerometer (where a single integration of the accelerom
`eter signal yields the velocity), one can determine the
`position and/or orientation of the accelerometer sensor
`which may be affixed to a portion of the body. The double
`integration promotes the introduction of drift into the posi
`tion signal calculation by continual accumulation of all
`minor acceleration errors, for example, errors introduced by
`sensor noise, transverse sensitivity, temperature sensitivity
`and the like. In some instances, sensor drift can be reduced,
`e.g., in the case of accelerometer temperature sensitivity, the
`accelerometer's temperature can be kept relatively constant
`using commercially available crystal ovens. Regardless of
`whether the sensor drift is reduced, over relatively short
`periods of time, the error accumulation might not reach a
`significant level. However, over longer periods of time, the
`error accumulation might become significant, in which case
`the determined position may need to be re-zeroed to elimi
`nate this error. The re-zeroing may be accomplished by
`comparing the output which includes the error drift to a .
`known or otherwise estimated reference value to determine
`the size of the error drift which may then be subtracted from
`the sensor output.
`Other sensors which may find use in conjunction with the
`above combinations include bend sensors which measure the
`
`10
`
`15
`
`20
`
`25
`
`30
`
`35
`
`45
`
`50
`
`55
`
`65
`
`One simple estimate is to set the error term, E(t), of the
`present fast measurement equal to the difference between the
`present slow measurement (i.e., the present slow measure
`ment which is determined to represent the placement of an
`entity dt seconds ago) and a prior fast measurement deter
`mined to correspond to that same sample instance dt seconds
`ago. In this estimate example, the slow measurement is
`effectively used as the accepted true value for dt seconds
`ago. The preceding estimate of E(t) is represented in equa
`tion form for time instance t' as
`
`Such a simple estimate, where dE(t)=0, may be acceptable
`if we do not wish to attempt to estimate and remove any
`trends in the fast sensor signal drift.
`A slightly more complex estimate of the error term, E(t),
`of the fast measurement at time t', which incorporates a
`non-Zero value for dE(t), is given as
`
`where the error over the most recent calculatable single
`sample period
`
`is averaged over a desired number m (where mid=1) of
`previous sample values of dE(t'-dt) to yield dE(t-dt).
`Here we assume that in addition to the error known to exist
`at the sample dtago, E(t-dt), the fast sensor has also drifted
`approximately the same amount during each sample period
`for the total (dt+m-1) time period, i.e., for a total of
`(dn+m-1) sample periods. The multiplication factor of dn is
`added since this average drift is assumed to occur over each
`of dn sample periods, from time (t'-dt) to the present time,
`t'.
`
`A yet even more complex estimator may (instead of just
`averaging, i.e., zero order growth, or zero-slope first order
`growth) extract first, second or other order growth tenden
`cies for dE(t), for example, by fitting a polynomial (e.g.,
`a line or quadratic) through m (e.g., m=dn+1) previous
`values of dE(t) and then determining the value of the
`polynomial for dE(t) at time t, which when added to
`E(t-dt) yields our estimate for E(t). For instance, if we fit a
`straight line, e.g., using least squares techniques, through 3
`previous samples of dE(t), where dE(t-4) was 1, dECt
`3) was 2 and dE(t'-2) was 3, we would estimate that dE(t)
`as 5. Thus, our estimate for E(t)=E(t'-dt)+5, and our estimate
`for P(t) is P(t)=P(t)+E(t'-dt)+5.
`One may use other techniques for estimating the error
`terms or P(t) directly, by employing many of the accepted
`practices known in the fields of Estimation Theory and
`Kalman Filtering. Such techniques may incorporate knowl
`edge of the dynamics and other physics of the sensing
`devices, along with a sequence of previous measurements, to
`produce a more elaborate error-correction algorithm. Esti
`mation/prediction algorithms typically use one or more
`previously measured sensor values and errors (for example,
`deviations between prior slow and fast measurements) as
`inputs to an estimation filter which produces a prediction of
`the present sensor value or of the error at the present sample
`time. Such estimator algorithms may be implemented in
`software on a data processor. Using any of a variety present
`value or error-term estimators/predictors, we may estimate
`the amount necessary to re-zero the fast device to reduce the
`expected error for the current fast sample. Digital Control of
`
`META 1030
`META V. THALES
`
`
`
`5,592,401
`
`10
`
`15
`
`20
`
`25
`
`30
`
`35
`
`45
`
`50
`
`7
`bend of abody part about a single axis (as exemplified by the
`elbow or knee). Bend sensors may be of many varieties,
`where typically bend sensors may be resistive bend sensors,
`fiber optic bend sensors, potentiometers and the like, as
`described in U.S. Pat. Nos. 5,047,952, 5,280,265, 4,542,291
`and Marcus et al. Depending on the particular bend sensor
`and structure used, the bend sensor may not provide accurate
`information at a complex joint or about the absolute orien
`tation of an elongated body part around its long axis.
`However, since bend sensors can provide fast, non-drifting
`angle measurements about a single axis, the fast angle
`information may be used in combination with slow sensors
`to produce faster measurements of certain selected important
`single-axis degrees of freedom. In such cases, the sensor
`values for the remaining orientation degrees of freedom, if
`measured only by the slow sensors, may lag. If it is unde
`sirable to have some degrees of freedom with different
`latencies than others, a fast orientation-sensing device (e.g.,
`an accelerometer or gyroscope) may be used (in addition to
`the bend sensor) solely to provide non-lagging orientation
`information which must then be re-zeroed as described
`above.
`Depending on its use, the fast sensor may vary in size.
`Particularly, for mounting on a human body, one would
`desirably have sensors which have their largest dimension
`less than about 4 cm, preferably less than about 2.5 cm. The
`sensor will usually be packaged in a form convenient for
`mounting on the body and with minimal interference with
`movement of the body part, and weight of only several
`ounces. The packaging may assume any convenient shape,
`particularly in association with a particular body part. The
`sensor may be mounted to various straps which may take the
`form of an elastic band or fixed to other wearable articles of
`clothing which retain their relationship with the point of
`interest.
`For slow devices, one may use position sensors, such as
`electromagnetic sensors, as supplied by Polhemus, Inc.,
`Colchester, Vt. and Ascension Technology Corp., Burling
`ton, Vt. Other sensors which may find use, include optical
`tracking devices (including such systems as used by BioVi
`sion, Menlo Park, Calif. and Adaptive Optics Associates,
`Cambridge, Mass.), acoustic sensors (including such ultra
`sonic sensors as provided by Logitech, Freemont, Calif.),
`Global Positioning System sensors (GPS) or Differential
`GPS (DGPS) (including such sensors as provided by
`Trimble Navigation, Sunnyvale, Calif.), and the like.
`Usually the optical tracker system will comprise one or
`more light sources and one or more light receivers, where
`the light sources will usually be mounted on the entity, and
`the light receivers mounted at fixed positions. The light
`sources and receivers may also be mounted at fixed positions
`and reflectors mounted on the entity, or sources and receiv
`ers mounted on the entity and reflectors mounted at fixed
`positions. In the following discussion of optical trackers, the
`case where the light sources are mounted on the entity and
`receivers mounted at fixed positions will be used to illustrate
`many of the considerations, however, the ideas provided are
`not to be limited to such case and may also pertain to a case
`which uses reflectors.
`Optical trackers may be considered delayed signal sensors
`for a number of reasons. There will usually be at least two
`light sources and two light receivers for following the
`movement of the entity in 3 dimensions and determining its
`6 degrees-of-freedom in space relative to a fixed reference
`coordinate system. In most instances there will be a plurality
`of light sources and a plurality of light receivers. Therefore,
`there will be multiple measurements which have to be sorted
`
`55
`
`60
`
`65
`
`8
`and analyzed to determine which light source is being
`received and the site of the light source, which may require
`triangulation. (The art of triangulation procedures for deter
`mining the location of a point based on multiple measure
`ments of the point relative to fixed known locations are well
`documented in the literature and will not be discussed in
`detail here.) While the signal is transmitted very rapidly
`from the light sources on the moving entity, the complexity
`and ambiguity of the signal will usually lead to slow
`processing. Ambiguity occurs when the light receiver or
`tracker is subject to being blinded during motion, so that
`there are periods of reduced intensity or erratic signal.
`Depending upon the motion path of the entity and barriers in
`the path, as well as whether the movement through the blind
`area is slow or fast, the data processor may be unable to
`determine the position of the entity for an extended period
`of time and may be uncertain as to the position of the entity
`when it exits the blind area. The uncertainty will include
`being unable to determine which sensors are in which
`positions in relation to their prior positions, before the light
`transmission became obscured. Therefore, with optical
`trackers, one is not only interested in using a second sensor
`to provide a rapid signal rapidly processed, but one is also
`interested in being able continuously to determine the posi
`tion of the entity, without periods of unreliability and
`uncertainty. The second position sensor would still be desir
`able, even if the optical tracking information could be
`processed rapidly to provide the position information during
`the periods of signal interruption.
`If rapid processing of optical tracking information could
`be achieved, there would still be a need for a second position
`sensor so that there would be an accurate position at the time
`that the optical tracker became unreliable. If