58
`
`IEEE TRANSACTIONS ON MAN-MACHINE SYSTEMS, VOL. MMS-11, NO. 1, MARCH 1970
`
`determine the identity and layout in three dimensions
`of a group of familiar objects if we had designed our
`system to deliver 400 maximally discriminable sensations
`to the skin. The perceptual systems of living organisms
`are the most remarkable information-reduction machines
`known. They are not seriously embarrassed in situations
`where an enormous proportion of the input must be
`filtered out or ignored, but they are invariably handi(cid:173)
`capped when the input is drastically curtailed or arti(cid:173)
`ficially encoded. Some of the controversy about the neces(cid:173)
`sity of preprocessing sensory information, the author
`thinks, stems from disappointment in the rates at which
`human beings can cope with discrete sensory events. It is
`possible that such evidence of overload reflects more an
`inappropriate display than a limitation of the perceiver.
`Certainly, the limitations of the system we have been
`working with are as yet attributable more to the poverty
`of the display than to overtaxing the information handling
`capacities of the epidermis.
`
`REFERENCES
`[1] J. C. Bliss and H. D. Crane, "Touch as a means of com(cid:173)
`munication," Stanford Research Inst. J., no. 5, p. 7, 1969.
`[2] J. J. Gibson, The Senses Considered as Perceptual Systems.
`Boston, Mass.: Houghton-Mifflin, 1966, pp. 26~286.
`[3] D. 0. Hebb, Organization of Behavior. New York: Wiley,
`1949.
`[4] J. Holt, How Children Learn. New York: Pittman, 1967.
`[5] K. Koffka, Principles of Gestalt Psychology. New York:
`Harcourt, Brace, and World, 1935.
`[6] W. Metzger, Gesetze des sehens. Frankfort: Kramer, 1953.
`[7] L. G. Roberts, "Machine perception of three-dimensional
`solids," Lincoln Laboratory, Massachusetts Institute of
`Technology, Cambridge, Tech. Rept. 315, 1963.
`[8] W. Schiff, "Perception of impending collision: A study of
`visually directed avoidant behavior," Psychol. Mono., vol.
`79, no. 604, 1965.
`[9] B. F. Skinner, The Technology of Teaching. New York:
`Appleton, 1968.
`[10] J. G. Taylor, The Behavioral Basi.s of Perception. New
`Haven, Conn.: Yale University Press, 1962.
`[11] E. B. Titchener, "Instructors' manual," pt. 2, in Experi(cid:173)
`mental Psychology, vol. 2. New York: Macmillan, 1905.
`[12] H. Wallach, M. Moore, and L. Davidson, "Modification of
`stereoscopic depth-perception," Am. J. Psychol., vol. 76,
`pp. 191-204, 1963.
`
`Optical-to-Tactile Im.age Conversion for the Blind
`
`JAMES C. BLISS, MEMBER, IEEE, MICHAEL H. KATCHER, CHARLES H. ROGERS, MEMBER, IEEE,
`AND RAYMOND P. SHEPARD
`
`Abstract-This paper describes two optical-to-tactile image(cid:173)
`conversion systems being developed for the blind. The first is a
`reading aid in which an area on the printed page about the size
`of a letterspace is translated into a corresponding vibratory tactile
`image. The tactile image is produced by a 24-by-6 array of pins
`driven by piezoelectric bimorphs. The array of 144 pins fits on the
`distal and a portion of the middle phalanges of one finger. The
`piezoelectric bimorphs cause the pins to impact the skin in a non(cid:173)
`linear manner. Precise measurements on this bimorph-finger
`system are given. These measurements also show that shades of
`"grey" can be displayed by sequentially varying the threshold level.
`Three experiments conducted with the reading aid involved
`measurement of legibility, reading rate, and the effect of field of
`view. Legibility in the 92-98 percent range was obtained at the
`design magnification. A reading rate of 50 words per minute was
`achieved with one subject after roughly 160 hours of practice.
`Three other subjects achieved reading rates of over 10 words per
`minute after about 40 hours of practice. Reading rate increased
`markedly as the number of columns in the array was varied from
`one to six.
`The second optical-to-tactile image-conversion system is merely
`an extension of the first to permit information to be acquired from
`
`Manuscript received April 3, 1969; revised October 1, 1969.
`This work was supported primarily by the Social and Rehabilita(cid:173)
`tion Service, under Grant VRA-RD-2475-S-67, and partly by the
`Office ,of Education, under Grant 0-8-071112-2995(032) and the
`Seeing Eye, Inc.
`.
`J. C. Bliss, M. H. Katcher, and R. P. Shepard are with the
`Stanfoi:d :Research Institute, Menlo Park, Calif.
`C. H. Rogers is with Stanford University, Stanford, Calif.
`
`the environment. In fact, ultimately only one system with two sets
`of optics, one appropriate for the printed page and one appropriate
`for environment sensing, would be used. A portable, battery(cid:173)
`operated experimental model is described.
`Two preliminary experiments with this environment sensor
`involved form recognition and pursuit tracking. Performance by
`blind subjects using the tactile display matched performance by
`sighted subjects using a corresponding light display. However,
`several problems must be overcome before this application can be
`satisfied in practical situations.
`
`l. INTRODUCTION
`
`F OR THE PAST hundred years, tactile displays
`
`have been suggested for many purposes, including
`sensory aids for the blind and deaf, sensory feed(cid:173)
`back for remote manipulators and prosthetic limbs, con(cid:173)
`trol and navigational displays for astronauts and avia(cid:173)
`tors, and "feelies." Very few of these suggestions have
`been developed to the point of common usage. However,
`advances in materials, electronics, and computer tech(cid:173)
`nologies now make much more complex tactile displays
`feasible, although many difficulties presently confront
`the designer of tactile displays. There are few commerci(cid:173)
`ally available tactile stimulators, and special designs are
`not always straightforward. Also, little is known about
`optimum stimulus parameters and about the character(cid:173)
`istics and capabilities of the tootile char- Thus, the
`Valve Exhibit 1039
`Valve v. Immersion
`
`

`

`BLISS el al.; OPTICAL-TO-TACTILE !:MAGE CONVERSION
`
`engineering of tactile displays is just developing, and the
`neurophysiological, psychophysical, and perceptual foun(cid:173)
`dations for an engineering design philosophy have yet to
`be assimilated.
`In this paper two recent projects in the design of
`tactile displays are described. Both of these projects in(cid:173)
`volve optical-to-tactile image-conversion devices. The
`objectives of both systems are to enable a blind person to
`read normal printed material and obtain information
`about his surroundings important to mobility.
`
`II. READING Arn FOR 'l'HE BLIND
`
`Our basic design for a direct-translation reading aid
`with a tactile output has been described by Linvill and
`Bliss [1], and Bliss [2]. In this design an area about the
`size of a letter space is imaged on an array of phototran(cid:173)
`sistors. The signal from each phototransistor controls a
`tactile stimulator in a corresponding array of tactile
`stimulators. Thus, a vibratory tactile image is produced
`of whatever is printed on the page.
`A primary consideration in the design of this reading
`aid was the spatial resolution of the tactile image. Of
`course, with appropriate optics a single movable photo(cid:173)
`transistor coupled to a single tactile stimulator could in
`principle permit a person to eventually obtain by scanning
`enough information for letter identification. However, if
`the scan is manual, reading would be extremely slow. If
`it is automatic, perceptual considerations would limit the
`scan rate so that a slow reading rate would also result.
`Therefore some parallel channels are necessary for an
`acceptable reading rate. In this paper, we consider the
`case of completely parallel input for the vertical dimen(cid:173)
`sion and a single horizontal scan. We also consider parallel
`input in the horizontal dimension but, since the scan is
`horizontal, additional columns of channels provide no
`new information and the optimum number depends on a
`tradeoff between perceptual and economic considerations.
`I By considering the spatial spectral content of alpha(cid:173)
`betic shapes as they occur in normal printed material,
`~e showed that a minimum of 24 phototransistors are
`needed in the vertical dimension of the array [2] in order
`to obtain acceptable legibility of alphabetic shapes. Ex(cid:173)
`periments with various numbers of vertical columns, each
`with 24 pbototransistors, indicate that higher reading
`rates can be achieved as the number of vertical columns is
`increased [3]. These considerations have led to the read(cid:173)
`ing aid shown in Fig. 1, which is based on a 24-by-6 array
`of phototransistors and a corresponding stimulator array.
`The optical pickup probe shown in Fig. 1 was developed
`in our laboratories. It contains a monolithic integrated
`array of 144 phototransistors on a single silicon chip 120
`mils by 60 mils. This phototransistor array was especially
`designed and constructed in the Stanford University
`Solid-State Laboratories for this application, and bas been
`described previously by Gary and Linvill [4], and Brugler
`et al., [5]. The phototransistors are operated in the charge(cid:173)
`storage mode [6] with a storage time of about 5 ms. Thus
`
`59
`
`Fig. 1. Portable model of the reading aid. This model, designed
`for personal use by a blind person, incorporates the features of
`the previously described models of the reading aid plus bat(cid:173)
`t.ery operation, adjustable magnification, and reduced size.
`
`the image is updated every 5 ms, several times more
`frequently than necessary to be within the perceptual
`time of the u er.
`As the probe is moved across the printed page, images
`of the print are sampled by the 24-by-6 phototransistor
`array. The phototransistor signals are multiplexed over
`six '\¾ires to the tactile stimulator drive transistors as shown
`in Fig. 2. An automatic threshold-control circuit frees
`the reader from all adjustments as he changes from mat
`to glossy paper. This threshold-control ci:ccuit adjusts
`the black-white decision boundary to a fraction of the
`peak signal from one column of the phototransistor array
`averaged over about 10 seconds.
`
`Tactil,e Stimulator Properties
`Of the various physical possibilities for tactile stimuli
`we chose mechanical vibration because of the convenience
`and simplicity of the piezoelectric bimorph as a stimulator,
`and because a nonpainful sensation is obtained with good
`two-point discrimination. In addition, these stimulators
`require less power than any we have found, and they
`can be closely packed relatively easily.
`A piezoelectric reed mounted as a cantilever is illus(cid:173)
`trated in Fig. 3. Such reeds are constructed of lead zircon(cid:173)
`ate and are commonly used as generators in phonograph
`cartridges. In the reed illustrated in Fig. 3, the upper and
`lo,ver surfaces are coated with nickel and serve as the
`electrical terminals. The center conductor is a thin brass
`sheet. Under application of voltage of proper polarity, the
`upper lead zirconate slab contracts longitudinally, the
`lower one extending. The result is that the reed flexes and
`the end deflects upward. The opposite polarity of voltage
`has the opposite effect.
`In 9ur application a short 10-mil diameter wire is
`fastened to the free end of the bimorphs along a vertical
`axis as shown in Fig. 3. The array of free tips is accu-
`
`

`

`60
`
`IEEE TRANSACTIONS ON MAN-MACHINE SYSTEMS, MARCH 1970
`
`vR/F
`I
`
`: "> - -~
`
`\
`I
`I
`
`I
`I I .-+-+--+---, I
`I I
`I
`11
`I I
`
`I
`I
`
`, - - - - - - - - , , - - - - - 7
`I
`I I
`I
`I
`I I
`I
`I - - - - -
`I I
`I
`II
`I
`11
`U::t=:==i=E~
`"' ~~~il$r
`... ~
`4-<11----1.1---1--'--+-~~~l--..lC::C:ie:::i
`"'"' a:
`
`PHOTOTRANSISTOR
`ARRAY
`
`Bl MORPH
`DRIVERS
`
`11
`I I
`
`BIMORP H
`ARRAY
`
`'1
`I
`
`READI NG UNIT
`
`I
`I
`
`L _____ _J
`
`I I
`I
`I
`I I
`CLOCK
`I
`L _______ _j L _ _ _ _ _J
`CONTROL UN IT
`TACTILE UNI T
`Fig. 2. Simplified block diagram of multiplex electronics. The
`shift register specifies which row of six bimorph drivers is to
`receive signals from the six phototransistors in the correspond(cid:173)
`ing row. Thus, the phototransistor signals are transferred, one
`row at a time, to the bimorph drivers, over six wires.
`
`Fig. 4. Position of the bimorph with small voltage spikes super(cid:173)
`imposed to show interval during which bimorph is in contact
`with the finger. Horizontal scale: 1 ms/div. Vertical scale:
`2.5 mils/div.
`
`4 0~ - - - - -- - - - - - - - - - - - - - ---,
`
`35
`
`f O 180
`
`_,,,., BIMORPH UNLOADED
`
`LIGHT FINGER
`PRESSURE
`
`V
`
`Fig. 3. Piezoelectric bimorph reed mounted for use as a tactile
`stimulator.
`
`rately positioned with respect to a perforated plate that
`is curved to fit the finger. The pins move through a 40-mil
`hole to impact the skin. This mode of tactile stimulation
`has been studied by Rogers and some results follow.
`The most intense sensation is felt when the rest position
`of the skin is slightly above the rest position of the bimorph
`pin tips. Under this condition the bimorph tip impacts the
`skin, and contact between the skin and the pin is broken
`each cycle of bimorph vibration as shown in Fig. 4. This
`figure shows that the peak-to-peak swing of the bimorph
`is 8.3 mils, but only a small portion of this occurs while
`the bimorph is in contact with the finger. The depth of
`skin indentation was about 2.6 mils or 65 microns. The
`duration of the skin contact under the conditions illus(cid:173)
`trated was approximately 0.8 ms out of the total period
`of 4.3 ms or about 67 degrees. This was roughly the
`maximum indentation that could be obtained on this
`subject's index finger.
`Fig. 5 illustrates the resonant characteristics of thi
`stimulator in both loaded and unloaded conditions. For
`' these curves, the bimorph stimulator was driven with a
`0-to-30-volt pulse, 2.6 ms in duration, and the period
`was varied in order to give fundamental frequencies frolfi
`12.5 to 250 Hz. _As shown in Fig. 6, the bimorph. in(cid:173)
`variably responded to these pulses by ringing at or 'near
`
`--'- - --'---:-'-:--:-'-::--::-':-:----,J
`o,L-__ ---1.__,,1..__J__J,,_ _
`100
`200
`300 400500
`1000
`10
`20
`30 40 50
`FUNDAMENTAL FREQUENCY OF INPUT (Hz)
`Fig. 5. Bimorph deflection as a function of the repetition rate
`of 2.6-ms driving pulses.
`
`_
`
`its major resonant frequency. In Fig. 5 it is the peak-to(cid:173)
`peak amplitude of the first cycle of response after each
`drive pulse that is plotted for each -driving frequency.
`The upper curve in Fig. 5 is for the bimorph vibrating
`in free air. The lower curve is for the same bimorph when
`it ,rns allowed to contact a finger during a portion of its
`upward swing. Note that loading the bimorph stimulator
`greatly reduces the peak-to-peak amplitude and also raises
`the resonant frequency. Also, notice that the peak in the
`response versus frequency plots do not occur at ex" ct
`integral divisions of the fundamental. In particular, the
`peaks are closer together with the bimorph unloaded and.-<
`farther apart under load than would be expected by
`integral divisions of the corresponding observed funda-(cid:173)
`mental. (Figs. 4, 5, and 6 are for one bimorph and slightly
`different results would be expected from other bimorphs.)
`
`Method for Displaying Several Intensity Levels
`The existence of these peaks in response below reso(cid:173)
`nance suggest a method for achieving a graded intensity
`of stimulation corresponding to "tactile grey." For ex(cid:173)
`ample, in our reading aid, if the threshold level dis(cid:173)
`tinguishing between black and white was sequentially
`varied through four discrete levels, stimulators corre(cid:173)
`sponding to fully black portions of the image would be
`pulsed every cycle and thus near their resonant frequency;
`.stimulators corresponding to gray portions of the image
`would be pulsed every other cycle or every third cycle of
`the clock frequency, depending on which of the four
`
`

`

`BLISS et al.: OPTICAL-TO-TACTILE IMAGE CONVERSION
`
`61
`
`Fig. 7. Tactile stimulation array. The 24-by-6 array of tactile
`stimulators fits on one fingertip. The stimulator pins are spaced
`50 mils apart along the finger and 90 mils apart across the
`finger. The perforated surface is curved to fit the finger.
`
`100
`
`90
`
`80
`
`~ 70
`t:
`8
`
`! 60
`,e 50
`>-
`u
`<(
`er 40
`::,
`u u
`
`<(
`
`30
`
`0
`
`Fig. 6. Lightly loaded bimorph driven by rectangular pulses 2.6
`ms in duration and 80 ms apart. Horizontal scale: 10 ms/div.
`Vertical scale: 2.5 mils/div.
`
`threshold levels their phototransistor signal exceeded.
`This four-level grey scale should improve the image quality
`of the letter shapes as has been hovm in facsimile y -
`terns. We are presently experimenting with this system.
`Our experiments have also shown that the bimorph
`response to various electrical waveforms can be predicted
`by Fourier analyzing the drive waveform, if the bimorph
`response to purely sinusoidal waveforms is known. Since
`the bimorph essentially filters out all drive frequencies
`other than the fundamental, as a good approximation only
`this component need be considered. This means that
`drive pulses of very short duration will give the same
`deflection waveform as a purely sinusoidal drive signal,
`and this property is u eful in multiplexing signals into an
`array of stimulators because each bimorph does not need
`a separate storage element.
`Many special techniques for construction of piezo(cid:173)
`electric bimorph stimulator arrays have been worked out
`by J. A. Baer and J. P. Gill. The bimorph reeds are indi(cid:173)
`vidually tested and carefully selected on the basis of
`resonant frequency. The method of mounting the reeds
`has progressed through many stages to the present tech(cid:173)
`niques based on mounting the reeds in epoxy, which per(cid:173)
`mits 40-mil wide reeds to be positioned on as small as
`45-mil centers.
`An example of a finished array using these technique is
`shown in Fig. 7. This is a 24-by-6 array with the rows
`on 50-mil centers and the columns on 100-mil centers.
`For several months several complete reading aids have
`been operational. A description of some test with these
`reading aids follows.
`1) Legibility: To verify directly the design resolution
`requirements for the reading aid, a legibility experiment
`was performed. Random strings of upper-case letters and
`rumbers and lower-case letters were printed in four ize
`of Mid-Century typescript. (This printed material was
`identical to that used by Arps et al., [7]). Each letter and
`number was manually scanned with the reading aid by
`two sighted and two blind subjects. All four subjects were
`instructed to take as much time as they needed to make
`each identification. The sighted subjects made their
`identifications by observing the light display and the
`blind subjects used the tactile array. The performance of
`each group is shown in Fig. 8. Legibility in the 92-9
`percent range was obtained at the letter-space height
`for which the reading aid was designed (i.e., 160 mils or
`24 samples across the height of the letter space). Since
`the size of the letters on the light display in no way taxed
`visual acuity, the sighted subjects' performance primarily
`
`-;,.--::.::
`
`/
`
`/
`
`,, ,,
`,, /
`~ /
`/
`I
`/'( /
`I
`I
`I
`
`I
`I
`I
`I
`
`" I
`
`I
`I
`I
`I
`I
`I
`I
`I
`I
`I
`I
`
`I "
`
`-0- SIGHTED UC
`-0- SIGHTED LC
`-♦- BLIND UC
`- •
`- BLIND LC
`
`160
`71
`106
`LETTERSPACE HEIGHT (mils)
`
`236
`
`Fig. 8. Reading-aid-output legibility as a function of letter(cid:173)
`space height. Recognition accuracy on random strings of upper(cid:173)
`case letters and numbers UC, and on lower-case letters LC,
`was measured for sighted subjects observing the light display
`and for blind subjects using the simulator display.
`
`indicated sampling rate influences on legibility. For
`sampling rates less than the design value, legibility
`dropped rapidly. For example, with the 71-mil letter(cid:173)
`space height (equivalent to about 10 photosensors acros
`the height of the letter space), a visual lower-case legi(cid:173)
`bility of 81 percent was obtained. Tactile performance was
`significantly worse, probably because of the smaller size
`of the letters, as well as the poorer resolution. Although
`some reading is possible with 1 percent legibility, it is
`slower, less accurate, and generally unsatisfactory.
`2) Reading Rate: We have made reading-rate determi(cid:173)
`nations at many stages in the development of this reading
`aid. With an early computer simulation of the reading aid,
`rates of 30 correct words per minute were obtained [l].
`With an early complete reading aid, four subjects read
`at rates greater than 10 words per minute and two of
`
`

`

`62
`
`IEEE TRANSACTIONS ON MAN-MACHINE SYSTEMS, MARCH 1970
`
`these subjects read at rates greater than 20 correct words
`per minute [2].
`Our most recent reading-rate determinations, taken
`with the reading aid in its present configuration, are shown
`in Fig. 9. These measurements were taken with the sub(cid:173)
`ject operating the reading aid in a natural way under
`standardized conditions. The data reported are for one
`subject and materials of similar difficulty level. 1 Each of
`the sessions lasted approximately 2 hours. The subject
`read silently at her own speed and scanned the printed
`page in any fashion she desired. She paused after each
`major paragraph, usually three times per page, and re(cid:173)
`lated the contents of the story to the experimenter. Her
`comprehension was always judged to be equal to or better
`than that of a good-sighted reader's understanding of
`the material.
`These experiments can be viewed as exercises in which
`we attempted to assess tl:e "actual" operating character(cid:173)
`istics of the entire device. While earlier investigations have
`explored various design and theoretical aspects, this was
`the first extended usage of the complete unit under normal
`conditions. Thus, these data include several sources of
`variation not accounted for in the earlier computer-simu(cid:173)
`lation experiments. An additional burden imposed was
`the manual tracking task. In early experiments, a track(cid:173)
`ing aid was used, which provided very free horizontal
`travel and an optional locking brake for vertical move(cid:173)
`ment that held line registration once it has been estab(cid:173)
`lished. However, with the probe shown in Fig. 1 and a
`trained subject, this tracking aid was found to be of
`little or no value. However, the tracking task does im(cid:173)
`pose the considerable burden of keeping its scanning rate,
`position, and direction coordinated with the decoding
`process. Closely related to this condition are the less
`perfect images of letters produced by phototransistor
`sampling of the printed page as opposed to the perfectly
`registered letters produced by a computer-driven display.
`We were greatly encouraged by the subject's apparently
`steady increase in reading rate from 20 to approximately
`51 words per minute over the seven-month interval that
`comprised roughly 128 hours of reading practice and 32
`hours of highly abnormal experimental manipulations.
`3) Window Width and Mapping: Once the reading aid
`was in the new configuration of Fig. 1, it was possible to
`recheck some of our earlier data [1] relating to the number
`of columns of stimulators required for adequate reading.
`In this set of exploratory experiments with the reading
`aid we duplicated those earlier manipulations, but sub(cid:173)
`stituted hand tracking, bimorph stimulators, and lower(cid:173)
`case letters for computer-driven bimorph stimulators and
`block capital letters.
`Since it is possible to consider the field of view of the
`photo~ransistor array as a "window" that moves past
`the letters, it is reasonable to ask how large this window
`should be and how many points are needed within the
`window for adequate resolution. Using the same subject
`
`1 Malcolm X. New York: Grove Press, 1966.
`
`50
`
`.. "5
`~
`.. -0
`8_40
`~
`w 30
`!q
`a::
`C) z
`15
`i5 20
`a::
`
`10
`
`50
`
`40
`
`30
`
`20
`
`10
`
`w
`~ a::
`
`C)
`z
`0
`<t
`w
`a::
`
`J:
`5
`i
`~
`0
`z
`i
`
`>-
`C)
`t: z
`do~
`<Dzu
`i3<t<f
`a::
`w
`I-
`...J
`
`C)
`z
`a:
`a.
`<t
`::.
`
`w
`~
`a::
`
`C)
`z
`0
`<t
`w a::
`
`0
`
`SEP
`OCT
`NOV
`DEC
`JAN
`FEB
`MAR
`Fig. 9. Reading-rate measurements and experiments conducted
`on one subject from September 1968 to March 1969.
`
`0
`
`as the earlier computer experiment we varied the number
`of columns of data in the window from 1-6 (5 columns
`were not tested). The results are shown in Fig. 10. 2 As
`with the earlier experiment we found that reading rate
`increases as the field of view increases. It is important to
`note, however, that in both of these experiments the
`maximum window width did not take in any more than
`one character. In fact, the subject was required to per(cid:173)
`form some degree of temporal integration in all but the
`6-column condition. The next logical step is to increase
`the field of view to take in more than one letter at a time.
`However, simply increasing the number of channels
`would impose complexity, cost, and size problems rela(cid:173)
`tive to the present reading aid. An obvious alternative is
`to reduce the density of the stimulators by distributing
`them over a large area. We were able to obtain some
`data bearing on this issue by separating the active columns
`in the 2- and 3-column conditions, as shown in Fig. 10.
`While these exploratory data are not precise enough to
`support a definitive determination, there is no apparent
`decrement within the range of separations tested.
`The closely related question of the effect of various
`mapping configurations was also investigated during this
`period. The results of these configurational changes are
`also plotted in Fig. 10. With the exception of the 2-
`column condition, which did not constitute a very great
`distortion of the character font, none of the abnormal
`mappings tried seemed promising. In fact, they all seemed
`to produce approximately similar results. While some of
`
`2 The appar1:n.t de~rease in reading rate between the 4- and
`6~column cond1t10ns 1s probably due to an equipment malfunc(cid:173)
`ti4:m that precluded effective stimulation by the outer columns of
`shmulators and probably introduced some "noise" into the image.
`Som~ strength for this interpretation is provided by the results
`obtamed when those outer columns were driven with the same
`d!tta as their immediate inboard neighbors. In this cas3 the com(cid:173)
`bme1 effect pr«;>duces one of the best results obtained. lending to
`the mterpretat10n that a completely functional 6-column array
`~an ~e expec~ed to produce greater reading efficiency than realized
`m !his expenment. Subsequent experience has verified this expec(cid:173)
`tation.
`
`

`

`BLISS et al.: OPTICAL-TO-TACTILE IMAGE CONVERSION
`
`63
`
`30
`
`25
`
`&20
`
`w 15
`a:
`"' z
`0
`~ 10
`
`5
`
`'
`I
`o~~ -~ -~ - -~ -~ --~
`2
`3
`4
`6
`5
`FIELD OF VIEW (columns)
`
`Fig. 10. Reading rate as a function of field of view and photo(cid:173)
`transistor-to-stimulator mappings. Line and vertical bars indi(cid:173)
`cate results as the number of phototransistor columns is varied.
`Right-hand dots indicate phototransistor columns and left-hand
`dots indicate bimorph columns. Connecting lines indicate photo(cid:173)
`transistor-column to bimorph-column mappings.
`
`the variability in the data wa due to the contextual
`problems associated with connected prose, requiring the
`subject to learn new character mappings in each condition
`also contributed to the variability.
`
`III. Ai.~ ENVIRONMENTAL SENSOR
`A straightforward modification of the reading aid to
`extend its range of application into environment sensing
`can be achieved by changing the optical system. To
`investigate this possibility, we constructed the optical-to(cid:173)
`tactile image-conversion system shown in Fig. 11. In this
`system the image formed by the lens falls on a 12-by-12
`array of phototransistors. The phototransistors are func(cid:173)
`tionally connected, one-to-one, to an identical array of
`tactile stimulators, which are in a 1 ¼-inch square in the
`handle of the device. Illumination of a phototransistor
`(above a threshold level) results in the vibration of the
`corresponding tactile stimulator. The threshold level is
`automatically adjusted so that reasonable operation over
`a 400-to-1 range of average ambient light intensity is
`obtained.
`The field of view of the system is approximately 30°.
`Because the receptor array is 12 by 12, the maximum
`spatial frequency the device can transmit is 6 cycles/30
`degrees or ¼ cycle/ degree.
`The normal human visual system, under optimal con(cid:173)
`ditions, can resolve a grating of approximately 60 cycles/
`degree. By definition, this level of visual acuity corre(cid:173)
`sponds to 20/ 20 vision. If this human terminology is
`applied to this device, the device may be said to have a
`visual acuity of about 20/6000. That is, it can at best
`resolve at 20 feet what the human visual system can
`resolve at 6000 feet. Obviously, since this is much more
`
`Fig. 11. Optical-to-tactile image-conversion unit for environment
`sensing. In the operator's left hand is the optical unit and the
`tactile stimulator array. Battery-operated electronics are carried
`under the right arm.
`
`than legal blindness, only extremely crude images are
`produced.
`Evaluation of the potential usefulness of such a device
`is a particularly difficult problem. Simple stimulus-re(cid:173)
`sponse tasks, which are easy to interpret, do not properly
`measure the complex man-machine interaction that can
`be achieved with the device. Complex tasks are difficult
`to interpret by anything more than observation and
`anecdotal description, and these are higbly subject to
`bias. Because of this we have attempted to develop
`quantitative tests and
`two exploratory experiments
`follow, one on form perception and the other involving a
`tracking task.
`
`Form Detection
`The purpose of this experiment was to determine how
`large an object had to be in order for it to be recognized
`on the tactile display and to compare that to the mini(cid:173)
`mum size which could be recognized on the visual dis(cid:173)
`play. Differences between these two sizes were assumed
`to reflect the superiority of one modality over the other
`in making use of the available information, while the
`absolute minimum detectable sizes were assumed to re(cid:173)
`flect the limitations of the device.
`The experiment consisted of presenting 44 figures (9
`triangles, 7 diamonds, 3 crosses, 9 circles, 9 squares, and
`7 rectangles) to the subject one at a time. The figures
`were white, of varying size, and taped up on a black
`board about 8 feet in front of the subject. He was allowed
`up to 1 minute to examine the figure and then asked
`to which of the 6 categories it belonged. He was not told
`whether or not he was correct and therefore, it is assumed
`that little or no learning occurred. The procedure was
`repeated three times for each of two subjects.
`
`

`

`64
`
`3
`2
`I
`0
`
`3
`2
`
`0
`
`3
`2
`I
`0
`
`0
`
`4
`
`8
`
`12
`I
`
`16
`I
`
`0
`
`4
`I
`
`8
`I
`
`12
`I
`
`16
`
`~~>< -X /
`
`_/ O'vfjlv
`◊~ a~ • -
`□ x-x,X
`□2;::r, \.
`◊_,;e
`f x,x
`
`3
`2
`
`0
`
`3
`2
`I
`a
`
`X
`
`-
`
`a
`
`0
`
`I
`4
`
`I
`12
`
`I
`16
`
`8
`
`I
`0
`
`I
`4
`
`I
`8
`
`I
`12
`
`I
`16
`
`x-x TACTILE
`0-----0 VISUAL
`
`Fig. 12. Number of correct identifications (ordinate) as a func(cid:173)
`tion of pattern size (abscissa) for six different_geometric shapes.
`Abscissa scale is a in degrees of field view. Tactile points are
`for the blind subject and visual points are for the sighted
`subject.
`
`The frequency with which each figure was correctly
`identified is shown in Fig. 12 as a function of its size for
`the blind and sighted (using visual display only) subject.
`It will be noted that as the size of the object increases,
`the probability of a correct identification also increases.
`However, while this result is unequivocal for the sighted
`subject it is only marginal for the blind subject.
`Although it is not surprising to find that tactile per(cid:173)
`formance is inferior to visual performance, it is surprising
`to find that figures as large as 18° on a side (i.e., covering
`almost ¾ of the display) could not be reliably recognized
`by the blind subject. This unexpected finding, however,
`probably reflects no deficit at all in the tactile system but
`rather is due to two defects in the tactile display.
`The first defect was that while the phototransistors
`and neon bulbs of the visual display were arranged in 12
`rows in perfect register, the bimorphs in the tactile dis(cid:173)
`play are arranged in 12 staggered rows. Con