ratio begins with the location of the center of the ac-
`tivated area. For this purpose, all nonzero pixels are
`taken to be part of the object. The center and the
`point farthest away from the center determine the
`major axis of the object. The minor axis is taken to
`be perpendicular to this. These axes provide an
`object-relative coordinate system in which if is pos-
`sible to specify, roughly, the location of bumps and
`depressions in the image. The bounding rectangle of
`the object is taken to be the smallest rectangle, with
`edges parallel to the axes, that contains the image.
`The aspect ratio of the object is taken to be the as-
`pect ratio of its bounding rectangle.
`
`Moving the Finger
`
`if the image read m is not satisfactory, as is usually
`the case at first, if is possible to move the finger and
`read another. An important part of the image analy-
`sis Invelves moving the finger so that an optimal
`image is sensed. The offset pressure, for example, is
`adjusted in this manner. Optimally, most of the
`touched area activaies the mid-range of the sensar,
`allowing bumps and depressions to be detected eas-
`ily. This is accomplished by reading in an image,
`computing the median pressure of all points above
`the noise threshold, and readjusting the finger pres-
`sure appropriately. This is repeated several times
`until an acceptable offset pressare is achieved.
`The finger is also moved to measure the stability
`{resistance to rofl) of an object. To measure the ob-
`ject’s stability im a given chrection, the object is
`pressed between the finger and the supporting sur-
`face by applying a fixed force on the ohject normal
`to the plane of the surface. The finger is then moved
`laterally in the desired direction. (The supporting
`surface was a thin layer of soft rubber, to prevent
`sliding.) The stability of the object is indicated by
`the amount of force necessary to meave the finger.
`
`The Matcher
`
`During the hypothesize step, the program must de-
`termine which of the objects it knows best matches
`the known data. Kor a small set of possible objects,
`
`such as the fasteners, it is not really important that
`ihis be done well. For a large set of possible objects,
`the quality of the matcher may be a determining
`factor in the speed of recognition. When the hy-
`pothesis is chosen by selection of the possibility that
`best matches the information given, usually the
`choice that has the largest number of features in
`common with the known facts is the best choice. In
`a system with a large mumber of parameters, other
`factors may also be taken inte consideration. For
`one thing, some features may be more important
`than others, either in general or for that particular
`possibility. Also, the features themselves may not
`exactly match—a bump may be too lurge, a shape
`distorted. In cases such as this, we wish to give the
`possibility only partial credit for a feature match.
`The most obvious way () Unplement such a
`matcher would be to use a numerical scoring aystem,
`with the weighting of factors for feature importance
`and partial maiches. This approach was avoided
`for the following bwo reasons. First, there would
`have to be a degree of arbitrariness in assigning the
`numbers: Is a circle a 50% match to a hexagon? Is
`shape 2.5 times as important as texture, or only
`twice as important? ht is unwise to trust the sums
`and products of numbers if the numbers themselves
`are chosen arbitrarily, The second objection is more
`of a philosophical one-—converting a complex set of
`symbolic structures into a single number causes us
`
`to throw away too much information too quickly. Of
`course, this information must eventually be lost
`the matcher must terminate by selecting a single
`item. But the pruning can be, and is, controlled in a
`More reasoned maruier.
`The implemented matcher takes two possibilities
`at a time and compares them on a feature-by-feature
`basis. If, for a particular feature, both items match
`the image to about the same degree, the information
`is ignored. If one of the items is clearly a better
`match, the feature is counted in favor of the appro-
`priate item. This procedure is repeated for each fea-
`ure and then the features themselves are compared
`ina similar manner. A feature counted toward one
`item will be cancelled by a feature counted toward
`another, if they are of approximate importance.
`In a large artificial intelligence system, the best
`match could be computed in parallel. Parallel
`
`a2
`
`The fnternational Journal of Robotics Research
`
`Page 877 of 1714
`
`SAMSUNG EXHIBIT 1006 (Part 5 of 8)
`
`Page 877 of 1714
`
`
`
`
`
`SAMSUNG EXHIBIT 1006 (Part 5 of 8)
`
`

`

`marker-propagation schemes, such as the one pro-
`posed by Fahiman (1979), would do such a task well.
`One important assumption, even for the parallel
`case, is that the binary comparison operatoris tran-
`sitive. Without this constraint, it would be necessary
`to campare each possible pair of tems, a task that
`grows as the square of the number of items.
`The transitivity of the predicate described above
`can be easily demonstrated, given the transitivity of
`individual feature comparisons. Assume that there
`exist three items, A, B, and C, such that A > B and
`Bo>C. Let fix, vy) be the set of features counted in
`favor of x when compared with y. Since the individ-
`ual feature comparisons are transitive,
`
`fiA, C) = CAA, BDU AB, OD
`
`and
`
`fic, Ay = CAC, Bu AOR, A).
`
`H ®is the feature set comparison predicate (he sec-
`ond stage of the algorithm above), then A > B im-
`plies f(A, B} > (CB, A}. Also, for any sets a, b,c
`and d such that a > b and c > d, it avast be that
`(a Uc) > (b U d}, because features that cancel in
`the individual sets will alsa cancel in the union. The
`assumptions A > Band B > C imply f(A, 8) >
`JOB, A} and f(B, C) > (CC, B) and, by the union role
`GAA, B) U FTR. ©) >CC, By U AB, AD. This
`may be rewritten as f(A, C} > fC, A}, which is the
`criterion for A > C. Therefore, the matching predi-
`cate is transitive.
`This maicher is really overkill for a possibility set
`of six objects with three parameters each, but it may
`be necessary if the program is to be extended to a
`large range of objects.
`
`>
`
`Proposed Kiforts
`
`A program that distinguishes among six obiects on
`the basis of three parameters is not too impressive.
`Even if it only got one bit from cach parameter, i
`should have correctly recognized eight objects. In
`the fisture, tactile recognition programs will have
`much more complex and more precise represen-
`
`tations of tactile images. Three improvements can
`help bring this about.
`The first is texture recegnition. The resolution of
`the tactile array sensor, while high, is not not nearly
`sufficient for measuring textural differences between,
`say, paper and glass. Texture sensing requires mea-
`surmg bulk effects of many tiny surface features. It
`is most easily accomplished if something is slid over
`the surface and a pattern of vibrations is detected.
`This can be likened to sliding a phonograph needle
`ever a record, Sensors uf the future may use em-
`bedded piezoelectric devices, or it may be possible
`io use the ACS directly as sort of a carbon micro-
`phone. However the information is derived, it must
`be processed into a useful characterization ofthe
`texture of the surface. Of interest is the intensity and
`periodicity of the signal. These features may be seen
`directly in the frequency domain. Texture processing
`may bear more similarity to the analysis of sounds
`than to the analysis of visual images.
`Another improvement might involve thermal rec-
`ognition: the difference between paper and glass is
`that glass feels cold. This is not actually because
`giass is lower in temperature, but because it is a
`better conductor of heat and sa it is more quickly
`able to carry away the heat generated by the body.
`We have constructed a small thermal conductivity
`sensor that works on this principle. Is the sensor, a
`resistive heating element is sandwiched between two
`iemperature-semsitive current sources. Any dic
`ference in the temperature of the two sensors is iadi-
`cated by an easy-to-measure difference in the cur-
`rents. The sensor is designed to be mounted on the
`finger in such a way that one temperature sensor
`may comtact the device being tested. As the heat is
`drawn from the object into the sensor, a difference
`in temperatures will develop. The primary disadvan-
`tage of thia first prototype is that it is large
`(G1 6.3 x 0.2 in.), resuiting in a relatively high
`thermal mass. This limits both the response time and
`the minimum size of the object that may be usefully
`tested,
`The third area that shows immediate potential for
`further research is the coordination of soultiple tac-
`tile images into a global picture. This is probably the
`most useful next step in tactile processing. This
`problem was deliberately avoided in the program
`
`Hillis
`
`43
`
`Page 878 of 1714
`
`Page 878 of 1714
`
`

`

`described through the choice of small objects that
`eould be read in a single impression. Such size Hmi-
`tations are probably unrealistic outside the labora-
`tory environment. The first real-world applications of
`iactde sensing wdl sot be in recognizing objects that
`fit on the tip of the finger, bat rather in orienting
`known objects grasped with an entire hand. This will
`require coordinating images from multiple sensors.
`We are enthusiastic about the future prospects of
`automated tactile sensing. What has been described
`here-—-the sensor, the finger, and the progragi-——is
`only an initial approach.
`
`Acknowledgments
`
`i would like to thank the following people for their
`help, ideas, and enthusiasm: Mike Brady, Tom Cal-
`lahan, Fred Dreneckhahn, Richard Greenblatt, John
`Purbrick, Gerald Sussman, Iohn Hollerbach,
`Michael Dertouzos, Margaret Minsky, Laurel
`Simmons, Patrick Winston, and, most of all, Marvis
`Minsky.
`
`REFERENCES
`
`Broit, M. 1979 (March). The utilization of an ‘artificial
`skin’’ sensor for the identification of solid objects. Proc.
`Sth far, Symp. industrial Robotics,
`
`Harmon, L. BD. 1982. Automated tactile sensing. far. J.
`Rebatics Res. 1(2):00-00.
`Hiulis, W. BD. 1981 (April, Active touch sensing. A. f.
`Memo 629, Massachusetts fnetitite of Technology Arti-
`ficial Intelligence Laboratory.
`Pablman, S. 1979. NETL: A system for representing and
`using real-world knowledas. Cambridge, Mass.: MIT
`Press.
`Larcombe. M. H. B. 9976 (March). Tactile sensors sonar
`and parallax sensors for robot applications. Paper deliv-
`ered at 3rd Conf. Indusitnal Robot Tech.
`Okada, T. 1979. Object handling system for manual in-
`dustry. (EEE Trans, Syst.. Man, Cybern. 92).
`Okada, T.. and Tsuchiya, 8. [977 (Oct.). On a versatile
`finger system. Prac. 7th Int, Svmp. fadustrial Robars.
`Purbrick, J. A. (1981). A force transducer emmptoying con-
`ductive silicone rubber. Paper delivered at Ist Rabot Vi-
`sion and Sensors Conf.
`Stoilikovic, 4.. and Clot, J., 1977. lategrated behavior of
`artificial skin. FEE Traas. Biomed. Engineering
`24{4): 396 3945,
`Storace, A., and Wolf, 8. 1979. Fonetional analysis of the
`role of finger tendens. J, Blomechanics 12.
`Weinreb, D.. and Moon, 0. 1979. Lisp Machine manual.
`Cambndge. Mass.: Massachusetts Institute of Technol-
`ogy Artificial Intelligence Laboratory.
`
`4%
`
`The international Journal of Robotics Research
`
`Page 879 of 1714
`
`Page 879 of 1714
`
`

`

`APRIL 19885
`
`A MULTLTOUGH THREE DIMENSIONAL TOUCH-SENSITIVE TABLET
`
`SK. Lee, W. Buxton, K.C. Smith
`Computer Systeme Research insiliute
`University of Taronia
`Torents, Gniario
`Ganade, MBS tA4
`
`(416)-976-8220
`
`ABSTRACT
`
`A projetype touch-sensitive tablet ie presented. The tablet’s main
`innovation is that H is capable of sensing more than one paint of
`contact at a time.
`in addition to being able to provide peaitien ooor-
`dingtes,
`the tablet also gives a measure of degree of contact,
`indeperxiantly for each point of coriact.
`in arder to arable mdi
`touch sensing, the iablet surface ls divided inte a grid of discrete
`points. The points are scanned using a recursive area subdivision
`aigortnm.
`in order to minimize the resolution lest due te the
`disergie natura of the grid, « novel interpolation echerne has been
`developed.
`Finally,
`the paper briefly discusses how mufthiouch
`sensing, interpolation, and degree af contact sensing can be com
`bined to expand our vooabulary in human-computer Interaction,
`
`% INTRODUCTION
`
`Rapid advancement of computer technology has opened a variety
`of new spolications. New applications and users mean demands for
`new modes of interaction. One consequence of this ls « growing
`appreciation of ihe imporiance of using aporepriate inpik technole-
`giss (Buxton, 1982}. Positioning devices are sean io be essential te
`graphics applications, Image transducers are required for pattern
`recognition in medical diagrosia, touch screens are usefis for ihe
`educalion of young children, and the QWERTY keyboard remains
`the usual standard for text processing. However, the range of inpul
`devices available is silt quite limited, as is our understanding af how
`to use ihern in the most elective manner.
`
`The intent of the research presented in this paper fs to increase the
`vocabulary thet can be uilized in human-computer imteraction. Our
`approach has been ta develop a new input technology that enlarges
`fhe demain of human physical gestues that can be captured for
`sonirol purpases.
`in whai follows, we will deserlbs the technology,
`what
`evolved from, and some aspects of how H can be used.
`
`& GVERVIEW
`
`The wansducer that we have develoged is a touch-sensitive tablet;
`fhat is, a flat surface that can sense where it is Being touched by
`the operators finger. This In Usell ls not new. Several such dev-
`coe are commercially available from a number of manufaciurers
`isee Appendix A}. What is unigue about our tablet is that K com-
`
`Permission to copy without fee all or part af this material is granted
`provided that the copies are not made or distributed for direct
`commercial advantage, the ACM copyright notice and the title of the
`publication and its date appear, and notice is given that copying is by
`permission of the Association for Computing Machinery. To copy
`otherwise, or to republish, requires a fee and/or specific permission.
`
`@
`
`7985 ACM 0-89791-149-0/85/004/0021
`
`$00.75
`
`Zi
`
`bines two additional features. First, it can sense the degres of oor
`tact in @ continuous manner. Secarcd, 8 can sense the amount and
`focation of a number of simulianeous pointe of contact. These iwo
`jeatures, when combined with touch sersing, are very important in
`respect is the types af Interaction thal we can support. Some of
`these are discussed below, bul see Buxton, HU, and Rowley (1985)
`and Brown, Buxton and Murlagh (1985) for more detail. The tablet
`which we present Is 8 continuation of work done in our lab by
`Sasaki at al (1989) and Metha (1282).
`in the presentation which follows, wa focua mainly on issuss relat-
`ing to the iransducar’s implementation. Two important contribullons
`discussed are our methed of scanning the tablet surface, and our
`method of maintaining high resalutlon despite ihe surface belng
`partitioned into a discrete grid. Additional technical details can be
`found fry Lee (1984).
`
`3%. WHY BLATETOUCH?
`
`Toueh sensing has a number of important characieristles. There is
`no physical stylus or puck ta get fost, broken, or vibrate out of pose
`ton. Touch tableis can be molded so as to make them easy io
`clean (theretare making them useful in clean environments ike has-
`pitaia, or dirty environments like factories), Since there is no
`mechanical intermediary between hand and tablet, there is nothing
`ia prevent multi-touch sensing. Ternplates can be placed over the
`tablet fo define epscial regions and, since the hand is being ued
`directly, these regions can bs manually sensed, thereby allowing the
`trained user to effectively "touch type’ on ihe tablet.
`Wihout pressure sensing, however, the ulllty of touch tablets is
`quite limited, One can move a tracking symbal around the sores,
`jor exampis, but when the finger Is aver a light bution, ihere is noth.
`Ing equivalent to ihe bution on & mouse fo push In order io make a
`selection. Yas, we could Hit the finger of the tablet, but that would
`be more like pulling (rather than pushing) the bution. And what i
`we wanted to drag an Rem being pointed al, or io indleate thal we
`wanted to start Inking? Lifting our finger woud leave our finger off
`ihe tablet, just when we want i in contact with f the most. There
`are ways around this problem, bul they are indirect. 9, however,
`ihe tablet has pressure sensing, we can push @ virtual button by
`giving an extra dH of pressure to signal a change in siate,
`fine
`Pressure has other advantages. One example is to soniral
`thiekness in a paint program. Bul why do we want multisie point
`sensing? A simple example would be W we had a templaie placed
`over ine iablet which delimited three regions of & cm by 2 cm.
`Where we touch each region could conirol the setting of & parame-
`ter associated with each region. ¥ we wanted to simulianesusly
`adjust all ihree parameters, then wo would have io be able io sense
`ail threa regions. An even easier example is using the tablet to
`emuate a plane keyboard thet can play polyphonic munic,
`
`Page 880 of 1714
`
`Page 880 of 1714
`
`

`

`
`
`APREE19 8
`
`& HARDWARE DESCRIPTION
`
`the fast multicle-touch-
`the hardware of
`A Brief description of
`sensitive input device (FMTSIO} ts introduced here. The design of
`the hardware ia based on the requremenis of the fast scanning
`algorihm and on tradeaiis between software and hardware, Many
`sensors have been examined for our particular application, however
`(Hursi, 74; Hillis, 82; TSD, 882; TASA, 9860; JSAC, 98:
`Matha, 882) none seemed te have the properties that satisfy the
`requrements of a FMYSID. The hardware basically consists of a
`sensor mairix board,
`row and column selection registers, A/D
`eanverting clrevis and a contrailing GPU,
`The design of the sensor matrix is based on the technique of cana-
`chance measivermert between a finger
`tip and a metal plats. To
`mintnize hardware, the sensors are accessed by row and column
`selection. Row selection registers select one ar more rows by set-
`ting the corresponding bits to a high state in order to charge up ihe
`sensors while the column selection registers select ane or more
`columns by turning on corresponding analog ewltches to discharge
`the sensors through uming resistors. The intersecting region of the
`selecied rows and the selecied columns represents the selected
`seneers a8 a group.
`AVD
`converting
`cireulia messure the
`discharging time interval of the selected sensors.
`A University of
`Torome 6809 board is used as a contrafling CPU. The touch sur-
`face of the senaar board cansisis of number of amall metal-costed
`reclanguarshaped areas serving as senser olate capacitors. The
`design of ine metal plate area of a uni senser depends on the
`measurable capaciiance change that resiiis when ihe srea is
`covered by @ fingay tip, and on the rasciutien that can be imple-
`mented.
`
`Charging dicde
`
`Discharging dieds
`
`& sow Line
`
`c
`
`.. volum line
`
`a
`
`R
`
`wee Mone
`Eh a column selection
`switch
`
`Fig. i A model of a selected senser in the senser matrix.
`
`in order fo select a sensor by row and column access, two diedes
`are wed with each sensor. One diode, connected te the row fine, ls
`used io charge up the sensars in the row. it is referred to a8 the
`Charging Olode (CD) as shown In Figure 1. The CD aise serves io
`block the charge flewing back to ihe row line when the rew line vot-
`tage is dropped to zere. The other diode called the Discharging
`BDiode(DD), sonnected to the column fing, snablea discharging of the
`seiscied row sensors to a virtual ground. Also ithe BD blecks
`charge flaw from the sensors in the selected row to the sensors in
`the unselected rews during the discharging period. The selection of
`rows, by the row selection procedure, causes the sensora to be
`charged, The sensars in the solumr are then discharged through
`associated iiming resistora eannecied io the eclumn seisctien
`switches.
`
`The charges stored in the selecied row({s) flow down through the
`selected switches te the virtual ground of @ fast operational
`amplifier. All tha discharging currants are correspondingly addad ta
`produce a signal frorn which the discharging time of all the selected
`sensors is found by comparison with @ threshold voltage.
`Pressure sensiiivily is incorporated by two measures: Firel there is
`the affect, here miner, af compression of ihe overlaying Insulator.
`Sacond there is the effect of intrinsic spreading of the compressible
`finger tip as pressure is increased,
`
`Page 881 of 1714
`
`The software in the controlling CPU utilizes communication with the
`host carmpuier
`io accommodate the Interpolation scheme. The
`dock rate (10 MHz} allows about % courts fo correspond to the
`sensor oapaciianes changs due t @ touch. But, ef courses, the capa~
`clance of all the ckeuiry atiached to the solumn line during the
`discharging period is much farger than the sensor capacilance.
`Thus before scanning the tablet fer a iouch, f is acanned com-
`pletely in all possibie resolution modes when mot touched, The
`yalues so obtained sre stored as references. Touches are
`idenitied by the diferences between ihe reference values and the
`values measured during use.
`The capaciance change corresponding te the touch by more than
`ene finger (or by the whole hand) is very large. Thus the number of
`bits in the counter should be enough lo measure tye maximum
`capantance, However ho
`is unnecessary aither fo have suificlont
`hits io measure the entira capacitance inchsding the surrounding
`capacitances, ar to stere the corresponding “complete” counter
`values as yeferences.
`f is necessary only to have ane mere BR
`than the number of bits required is count the valus of change in
`the sapackance rather than the complete value in order ta megaure
`the diferences of capacianes cue to fousk. Thus only an 8 bit
`ecounler is implemented. The counter enables the measurement of a
`7? bR capaclianse change regardless of the degroe of evarfliow in
`the counter.
`
`A factty is alse provided for Kentiying lempletes appiled to the
`surface of the tablet.
`
`& SCANNING ALGORITHM
`
`Gre ides of some significance that oan be introduced fe te avold
`scanning of all the pixels in the tablet which contain no information.
`For example, seanning all 2048 pointe of a tablet Raving a regaiu-
`tien 84 by 32 for ewer than 1 poinis is really quite a rkieuious
`idea.
`in jact, f the number of points to be searched fe comparably
`small, hen an improved aigarithm, here called recursive area subd
`vision, can be used. A particular
`implementation example f¢
`desoribad as follows.
`
`Consiter a tablet with reaciution & by 8 te be searched for 3 touch
`point as shown in Figure 2. First, check the tablet for touch as a
`whole region as shown by the area ABCD In the figura. ¥ touch ie
`detected, divide ihe tablet nic two equal regions shawn by the line
`EF and check sack of
`the iwo regione ABEF and EFCD for
`touchadness, Select the touched region, region EFCD in this cae,
`and divide ihis into hwo aqual regions as shown by the division line
`GH. Continue ils process on ihe touched region untll no further
`division is possibis, that fa, untill a unit sensor, designated as the
`region PKMO In Figure 2, Is nsached. The figure also shows the
`sequence of subdivision in the recursive subdivision scheme.
`
` G
`
`rv
`
`L
`
`C
`
`in}eoSaquencs of subdivision in binary operation.
`
`Fig. 2 Recursive subdivision operation for & by 8 tablet.
`
`22
`
`Page 881 of 1714
`
`

`

`PROCEEDINGS
`
`APRIL 1985
`
`Using this algorithm, a search for one palmi on 3 tablet having &
`resolution 64 by 32, requires 22 scanning ilmes, that Is
`
`Z* flog sub 2} (G4 ° a9} = 22
`
`® thera is no overhead in the recursive subdivision process and
`scanning begins at the “lop of the tree” Ghat is, with a region in
`which all pixels are grouped tegether}, then using this scheme, the
`number of touched pointe thal can be identified In the time that ft
`would take te detect one touch directly (ihai is, f all pixels are
`stanned one by one sequentialiy) fe
`
`N = {{64 ° 32} ovar 22} = 186.
`
`Thig shows immediately that the recursive subdivision schame Is
`much superior fo sequential scanning Hf the number af points to be
`scanned is fewer than 16.
`
`72 Spailel Resokslon
`One possible and immediate interpolation scheme fa is interpalzie a
`"ouchad’ point with all adjacent values which may not be large
`enough to be reporied as iouthed. A local array of 9 by 9 pointa
`ean be used for this interpolation. Some examples drawn on a
`laser printer (consequently having no intensity scale} are shown in
`Figuwe 3. These pictures are produced without feedback, that fs,
`drawn without the operator looking af ihe ouipul sereen. This does
`not allow the operator io compensate, that is, te select poinis where
`data are sparse In comparison with the Infended figure, but rather
`iakes dire input from the leeation of the figure drawn on the Input
`davice. The first picture {a} is drawn by moving 3 finger in a
`aivaight line (guided by a ruler} for various angles and the second
`one ib} is drawn by moving a finger in 4 fine guide by s circle drawn
`en 3 template. These teste show thal
`interpolation actually
`increases the spatial resolution ag well as the locatability of a fine
`point on a screen.
`
`8.
`
`INTERPOLATION
`
`i may seem thai the resolulion of the hardware Is foo low for use in
`graphics applications. Mewever touch infenaity and mulll-touch ser-
`sitivity can be used fo enhance resolution, This is possible because
`the center of a touch can be most secwately estimated by an inter-
`polation utilizing ihe values of the adjacent sensor Iniensities,
`Direct
`intarpaistion schemes for a few cases haa been imple-
`menied. One of interest is lo interpolate an array of 3 by 8 sensors
`using & touched point In the center. Another is to interpolate all
`points on the table. The later one obviously provides the highest
`resolion bul a3 a result
`it simply emulates a single touch tablet
`with very sigh resalulian.
`7, PERFORMANCE
`
`
`
`
`
`74 Seemer
`An ideal sensor matrix for a FAYTSIO weuld be one that has uniform
`and small reference values over a grouping level, a large variation
`of intensity due to e touch, and fast measurement time. The senor
`matrix of the prototype, however, has a relatively wicks ranges of
`reference values, However these values de nei change very much
`over extended periads of time. The resus show thet doubling the
`number of sensorsin a group in the column direction Increases the
`reference value by a iactor of about 1.5. This corresponds weil to
`theoretical estimates. As well the resulta show ihat increasing the
`number of sensors in a group in the row direction, in contrast, doas
`not Increase the reference value in gonsral, even H the number of
`the sensore is doubled in a group. The reference value ranges
`from 49 (or a single sensor in a group) te S86 (for the entire array
`of 64 by 22 sensors considered as a group).
`hh order ta account fer {ime and ether variations of the reference
`values, a thrashoid is included which must be overcome In order for
`&@ iowh te be detected. The threshold used ranges from 2 lo 7
`counts depending on group size. Using these threshold valuss the
`CPU does nol raporl uviouched poinis wrongly over Intervals of at
`wast & hours in either sequential or recursive subdivision modes.
`The recurshe subdivision scheme uses 6 ciffereant ihreshalds, cor
`sequently 4 is very unlikely fo repor « wrong polmt whereas the
`lingar acanning movie using only a single inreshold Is likely to be
`mare sensiiive.
`
`fa} Straight lines drawn by the tablet using 3 by 3
`sensor array interpolations
`Ths scales shown represent
`the boundaries of the
`aatusl sensors.
`
`coke
`jacale
`
`}
`i
`3
`i
`
`j
`
`3 i
`
`i
`;
`i
`“
`
`ca
`
`a
`
`wat Shen .
`wh
`é
`*
`
`{
`°
`2
`°
`-
`
`,
`
`*
`e,
`om
`
`eaten nee OY
`
`roe seals
`
`(o) A circle deawn by the tablet using 3 by 3
`sensor array Lntarpolutione
`Whe seales chown represent
`the toundariea of the
`Sctusl sensorss
`
`Fig 3 Poinis drawn by the tablet using an interpolation mathod.
`
`The Intensity of a single touch far a single sensor group varies over
`the lablet buf usually ranges above the threshold value by as much
`as 5. For @ single 64 by 32 sensor group, the intensity varies from
`person tO parson bul H ranges from the threshold to 124. This max-
`run is obtained when a palm rather than a finger touches the
`tablet. Anoiher
`inieresiing feature is
`that
`ihe raspanse time
`becomes fastey as the number of sensors in a group becomes
`larger, ared furihermare thai for the 64 by 32 sensor group, it ls pos
`sible tc datect af a hand merely placed in the vicinity of ihe tablet.
`
`Page 882 of 1714
`
`(ULLAddiedebtddiabhhlaintadainaulaSdhabatatzinbdataaiantamndicokumn
`
`
`
`5asa
`*a.
` FON acAle
`
`ae
`
`eee he
`
`*
`
`>
`
`>
`
`Page 882 of 1714
`
`

`

`CHE '85 PROCEEDINGS
`
`
`
`Since ihe spatial resoludien in the local interpolatian scheme |s lin-
`fed by the number of bite avaliable from the intensities of an array
`of 3 by 3 sensors, other scheme was considered.
`in this scheme,
`ai the points from a complete scan of a tablet are imerpolsted
`allewing ihe potenilal resolution to be almesi infinite, However this
`process simply emulaies a projective device and accordingly
`reporis only single point, which ts interpolated from all the points on
`the tablet. However wih this achame, there are a great mary ways
`of pointing fe @ specific location on a display screen, & feature with
`some Intriguing apolicatiion possiblities.
`7.2 Response Ties Boley
`The response time delay ls the time delay trom the beginning of &
`touch to an output recelved either by local terminal or by an outpid
`devices allached fo the hest computer, For mutiply touches, this
`delay will Increase with ihe number of touches. The prototype used
`with & 9800 baud-rate terminal
`to measure time delays. Actual
`response times were measused several times and averaged fer
`various cases and are tabulated in Tabis 1.
`
`Case
`
`
`{a} pts/sec
`
`masec /pt
`
`‘{b) plis/see
`
`rasec/ph
`
`
`
`
`
`
`
`TABLE < Actual Respense Tims Relays
`
`The cases in Tabls one are to be interpreted as follows:
`a@ ene sensor touched continuously
`kb
`twe sensors touwhed at the same time continuously
`6 four sensors toushed at the same time continuously
`
`8 CONCLUSIONS
`
`A prototype of a fasi-scanning multiple-touch-sensitive inpia tablet
`having both the adaptability and flaxitilty for a broad range of appl
`ealions kas been designed and implemented. Capactansa mesa
`wement of Individual sensor{s) which can be unlgusly addressed
`wing two diodes ser sensor, makes ik possible fo sense both the
`positions and iniensties of ane or more simultaneous touches
`without ambiguity. The sensor matrix is cuntrolled by Univeralty of
`Foronto 6809 board whose serisi port is connected to one af the
`VO ports of a host computer. Solware that ullizes the recursive
`subdivision algeriihim for fast scanning an array of 64 by 32 sensora
`on the tablet, and thai communicates wih the host computer, has
`been implemented and tested.
`
`§. ACKNOWLESGRMENTS
`
`The research described in this paper has been funded by the
`Natural Sciences and Engineering Research Council of Canada.
`This support is gratefully acknowkxiged,
`%. REFERENCES
`
`Brawn, (E., Burton, W. & Murtagh, . (885). Windews on Tablets as
`& Meane of Achieving Virnal Input Devices. Computer Sya-
`tams Research institute, Univeralty of Teranta.
`
`Buxsion, WW. (082), Lexical and Pragmatic Considerations of Inout
`Sirustures, Computer Graphics, WF (4%, 31-3.
`
`Burton, W., Hill, A. & Rowley, P. (1985). Issues and Techniques in
`Touch-Sensitive Tablet
`input, Computer Systerns Research
`Instiiute, University of Torenta.
`
`Hillis, W.D. (1982), A High Resolution imaging Touch Sensor, leter-
`national Journal of Robotics Research, 1 (2), 33 - 44.
`
`(9874) Electragraphic Sensor for Determining Planar
`Hurst, G.
`Coordinates, Uniied State Paient 3,798,370, March 1 974,
`Elagraphics, incorporated.
`
`JSR (3963, Pressure-Gensiive Conductive Rubber Data Sheet,
`Japan Synthetic Rubber Ce, New Product Development
`Department, JSR Bullding, 2-71-24 Tukijl, Chuo-Ku, Tokyo #4,
`danan.
`
`(1954), A Fast Aultiple-Foush-Sensitive input Device,
`Les, S.
`MASc. Thesis, Depariment of Electrical Engineering, Univer-
`sity of Taranto.
`
`Metha, N. (1862), A Flexible Machine interface, MASs. Thesia,
`Department of Electrical Engineering, University of Toronte.
`
`Sasaki, .., Fedarkeaw, G., Buxton, W., Retterath, C., & Smith, KC.
`(8h. A TouchSensitive Inpul Device. Proceedings af the
`Fifth international Conference on Comoutar Musie, Norih
`Texas Siats University, Denton, Texas, November, 1981.
`
`TASA (B80), Medel: x-y 3600 and sy controller, Model: FR-15
`Date Shest, Touh Activated Switch Arrays
`ine.. te
`Lawrence Siation Road., Sule G., Sunnyvale, CA S4088.
`
`TSO (1982), Touch Screen Digiizer Daia Sheet, TSG Display Pro-
`duxts inc. 35 Orville Brive, Bohernia, NY 1718.
`:
`
`VW. APPEMDON A: TOUCH TABLET SOURCES
`
`Big B