{"title": "VLSI Model of Primate Visual Smooth Pursuit", "book": "Advances in Neural Information Processing Systems", "page_first": 706, "page_last": 712, "abstract": null, "full_text": "VLSI Model of Primate Visual Smooth Pursuit \n\nRalph  Etienne-Cummings \n\nDepartment of Electrical Engineering, \n\nSouthern Illinois University, Carbondale, \n\nJan  Van  der  Spiegel \n\nMoore School of Electrical Engineering, \nUniversity of Pennsylvania, Philadelphia, \n\nIL 62901 \n\nPA 19104 \n\nPaul  Mueller \n\nCorticon, Incorporated, \n\n3624 Market Str, Philadelphia, \n\nPA  19104 \n\nAbstract \n\nA  one  dimensional  model  of primate  smooth  pursuit  mechanism  has \nbeen  implemented  in  2  11m  CMOS  VLSI.  The  model  consolidates \nRobinson's  negative  feedback  model  with  Wyatt  and  Pola's  positive \nfeedback scheme, to produce a  smooth  pursuit  system  which  zero's  the \nvelocity  of a  target on  the  retina.  Furthermore,  the  system  uses  the \ncurrent eye  motion  as  a  predictor for  future  target  motion.  Analysis, \nstability and biological correspondence of the system are discussed.  For \nimplementation  at  the  focal  plane,  a  local  correlation  based  visual \nmotion  detection  technique  is  used.  Velocity  measurements,  ranging \nover 4 orders of magnitude with < 15%  variation, provides  the  input  to \nthe smooth pursuit system.  The  system  performed  successful  velocity \ntracking for high contrast scenes.  Circuit design and performance of the \ncomplete smooth pursuit system is presented. \n\n1  INTRODUCTION \nThe smooth pursuit mechanism of primate visual systems is  vital  for stabilizing  a  region \nof the  visual  field  on  the  retina.  The ability  to  stabilize  the  image  of the  world  on  the \nretina has profound architectural and computational  consequences  on  the  retina and  visual \ncortex,  such  as  reducing  the  required  size,  computational  speed  and  communication \nhardware  and  bandwidth  of  the  visual  system  (Bandera,  1990;  Eckert  and  Buchsbaum, \n1993).  To obtain similar benefits in  active  machine  vision,  primate  smooth  pursuit  can \nbe a  powerful  model  for  gaze  control.  The  mechanism  for  smooth  pursuit  in  primates \nwas  initially  believed  to  be  composed  of  a  simple  negative  feedback  system  which \nattempts  to  zero  the  motion  of targets  on  the  fovea,  figure  I (a)  (Robinson,  1965). \nHowever,  this  scheme  does  not  account for  many  psychophysical  properties  of  smooth \n\n\fVLSI Model of Primate Visual Smooth Pursuit \n\n707 \n\npursuit,  which  led  Wyatt  and  Pola  (1979)  to  proposed  figure  l(b),  where  the  eye \nmovement  signal  is  added  to  the  target  motion  in  a  positive  feed  back  loop.  This \nmechanism  results  from  their  observation  that  eye  motion  or  apparent  target  motion \nincreases the magnitude of pursuit motion  even  when  retinal  motion  is  zero  or  constant. \nTheir scheme  also  exhibited predictive  qualities,  as  reported  by  Steinbach  (1976).  The \nsmooth  pursuit  model  presented  in  this  paper attempts  the  consolidate  the  two  models \ninto a single system which explains the findings of both approaches. \n\nTarget \nRetinal \nMoticn  Motion \n\ne~ G \n\nTarget \nMotion \n\nEye \nMotion \nlee \n> \nI \n\ne~~ \n\nEye \nMotion \n> \n\nee = e t  ~; G  ~ co  G r = 0 \n\nG+l \n\n(a) \n\n(b) \n\nFigure  I:  System  Diagrams  of Primate  Smooth  Pursuit  Mechanism. \n(b)  Positive \n(a)  Negative  feedback  model  by  Robinson  (1965). \nfeedback model by Wyatt and Pola (1979). \n\nThe velocity based smooth pursuit implemented here attempts to zero the relative velocity \nof  the  retina  and  target.  The  measured  retinal  velocity,  is  zeroed  by  using  positive \nfeedback to accumulate relative velocity error between the  target  and  the  retina,  where  the \naccumulated  value  is  the  current  eye  velocity.  Hence,  this  model  uses  the  Robinson \napproach  to  match  target  motion,  and  the  Wyatt  and  Pola  positive  feed  back  loop  to \nachieve  matching  and  to  predict  the  future  velocity  of  the  target.  Figure  2  shows  the \nsystem  diagram  of the  velocity  based  smooth  pursuit  system.  This  system  is  analyzed \nand the stability criterion  is  derived.  Possible  computational  blocks  for  the  elements  in \nfigure  I (b) are also discussed.  Furthermore, since this entire scheme is  implemented  on  a \nsingle 2 /lm CMOS chip, the method for motion detection, the complete tracking  circuits \nand the measured results are presented. \n\nRetinal \nMotion \n\nEye \nMotion \n\nFigure 2:  System Diagram of VLSI Smooth  Pursuit  Mechanism.  er \nis target velocity in  space,  Bt is projected target velocity,  Be  is  the  eye \nvelocity and Br is the measured retinal velocity. \n\n2  VELOCITY  BASED  SMOOTH  PURSUIT \nAlthough figure  I (b) does  not  indicate  how  retinal  motion  is  used  in  smooth  pursuit,  it \nprovides  the  only  measurement  of the  projected  target  motion.  The  very  process  of \ncalculating retinal motion realizes negative  feed  back  between  the  eye  movement  and  the \ntarget  motion,  since  retinal  motion  is  the  difference  between  project  target  and  eye \nmotion. \nIf  Robinson's  model  is  followed,  then  the  eye  movement  is  simply  the \namplified version of the retinal  motion. \nIf the  target  disappears  from  the  retina,  the  eye \nmotion  would  be  zero.  However,  Steinbach  showed  that  eye  movement  does  not  cea~ \nwhen the target fades off and on, indicating that memory is  used  to  predict  target  motion. \nWyatt and Palo showed a direct additive influence of eye movement on  pursuit.  However, \nthe computational blocks G' and  a of their model are left unfilled. \n\n\f708 \n\nR.  ETIENNE-CUMMINGS, J. V AN  DER SPIEGEL, P. MUELLER \n\nIn  figure 2,  the gain G models  the  internal  gain  of the  motion  detection  system,  and  the \ninternal representation of retinal  velocity is  then V r.  Under zero-slip  tracking,  the  retinal \nvelocity is  zero.  This is obtained by using positive feed  back to correct  the  velocity  error \nbetween target, er,  and  eye,  ee.  The delay  element  represents  a  memory  of the  last  eye \nvelocity  while  the  current retinal  motion  is  measured.  If the  target  disappears,  the  eye \nmotion  continues  with  the  last  value,  as  recorded  by  Steinbach,  thus  anticipating  the \nposition of the target  in  space.  The  memory  also  stores  the  current eye  velocity  during \nperfect pursuit.  The internal representation of eye velocity, Ve, is  subsequently  amplified \nby H and used to drive the eye muscles.  The impulse response  of the  system  is  given  in \nequations (I). Hence, the relationship between eye velocity and target velocity is  recursive \nand given  by equations (2).  To prove the stability of this system, the retinal  velocity  can \nbe expressed in  terms of the target motion as given  in equations (3a).  The  ideal  condition \nfor accurate performance is for GH = 1.  However, in practice, gains of different amplifiers \n() \n-.f..(z) = GH--_-)  (a);  ~(I1) = GH[-8(11) + u(n)]  (b) \n(}r \n(}e(n) = (},(n) - (}r(n) = GH[-8(n) + u(n)] * (}r(n) = GHL(},.(k) \n\n1 - Z \n\nz-) \n\n(I) \n\n() \n\n(}r \n\n(2) \n\nn-) \n\nk=O \n\n() r ( 11)  = (),( n ) (1  - GH)  =>  () r( 1l )  = 0  if  GH = 1 =>  () in) = (),( 11  ) \n()  (n) \n\n)  0  if 11  - GH I < 1 =>  0 < GH  < 2 for  stability \n\n11  ~ 00 \n\n11 \n\nr \n\n( a) \n\n( b) \n\n(3) \n\nare rarely perfectly matched.  Equations (3b) shows that stability is assured  for  O<GH<  2. \nFigure 3 shows a plot of eye motion versus updates  for  various  choices  of GH.  At  each \nupdate, the retinal motion is  computed.  Figure 3(a) shows the eye's motion  at  the  on-set \nof smooth  pursuit.  For  GH = 1,  the  eye  movement  tracks  the  target's  motion  exactly, \nand  lags slightly  only  when  the  target accelerates.  On the  other hand,  if GH\u00ab \nI,  the \nIf GH  ->  2,  the  system  becomes  increasing \neye's  motion  always  lags  the  target's. \nunstable,  but converges for GH < 2.  The three  cases  presented  correspond  to  the  smooth \npursuit system being critically, over and under damped, respectively. \n\n3  HARDWARE  IMPLEMENTATION \n\nUsing  the  smooth  pursuit  mechanism  described, a  single  chip  one  dimensional  tracking \nsystem has been implemented.  The chip  has  a  multi-layered  computational  architecture, \nsimilar  to  the  primate's  visual  system.  Phototransduction,  logarithmic  compression, \nedge  detection, motion  detection  and  smooth  pursuit  control  has  been  integrated  at  the \nfocal-plane.  The computational  layers  can  be  partitioned  into  three  blocks,  where  each \nblock is  based on  a segment of biological oculomotor systems. \n\n3.1 \n\nIMAGING  AND  PREPROCESSING \n\nThe  first  three  layers  of  the  system  mimics  the  photoreceptors,  horizontal  cells  arx:l \nbipolar  cells  of  biological  retinas.  Similar  to  previous  implementations  of  silicon \nretinas,  the  chip  uses  parasitic  bipolar  transistors  as  the  photoreceptors.  The  dynamic \nrange of photoreceptor current is compressed with a logarithmic response in  low  light  arx:l \nsquare root response in bright light.  The  range  compress  circuit  represents  5-6  orders  of \nmagnitude of light intensity with 3 orders of magnitude  of output  current  dynamic  range. \nSubsequently, a passive resistive network is  used to realize a discrete implementation  of a \nLaplacian  edge detector.  Similar  to  the  rods  and  cones  system  in  primate  retinas,  the \nresponse time,  hence the maximum detectable target speed, is ambient intensity  dependent \n(160 (12.5)  Ils  in  2.5  (250)  IlW/cm2).  However,  this  does  prevent  the  system  from \nhandling fast targets even in dim ambient lighting. \n\n\fVLSI Model of Primate Visual Smooth Pursuit \n\n709 \n\n20 \n\n15 \n\n10 \n\n5 \n~ \ng  0 \nu \n> \n\n-5 \n\n-10 \n\n-15 \n\n-20 \n\n20 \n\n15 \n\n10 \n\n5 \n\n~ \n]  0 \n\n\" > -5 \n\n\u2022  Target \n\n- -Eye:  GH=I 99 \n-Eye  GH=IOO \n__ .Eye:  GH=O_IO \n\n0 \n\n50 \n\nUpdates \n(a) \n\n\u00b7 10 \n\n\u00b7 15 \n\n-20 \n\n100 \n\n150 \n\n500 \n\n600 \n\n900 \n\n1000 \n\n700 \n\n800 \n\nUpdates \n(b) \n\nFigure 3:  (a)  The On-Set  of Smooth  Pursuit  for  Various  GH  Values. \n(b)  Steady-State Smooth Pursuit. \n\n3.2  MOTION  MEASUREMENT \n\nThis  computational  layer  measures  retinal  motion.  The  motion  detection  technique \nimplemented here differs from those believed to exist  in  areas  V 1  and  MT  of the  primate \nvisual  cortex.  Alternatively,  it  resembles  the  fly's  and  rabbit's  retinal  motion  detection \nsystem  (Reichardt,  1961;  Barlow  and  Levick,  1965;  Delbruck,  1993).  This  is  not \ncoincidental,  since  efficient  motion  detection  at  the  focal  plane  must  be  performed  in  a \nsmall  areas and using simple computational elements in  both systems. \n\nThe  motion  detection  scheme  is  a  combination  of  local  correlation  for  direction \ndetermination, and pixel transfer time measurement for speed.  In  this  framework,  motion \nis defined as the disappearance of an  object, represented  as  the  zero-crossings  of its  edges, \nat  a  pixel,  followed  by  its  re-appearance  at  a  neighboring  pixel.  The (dis)appearance  of \nthe  zero-crossing  is  determined  using  the  (negative)  positive  temporal  derivative  at  the \npixel.  Hence,  motion  is  detected  by  AND  gating  the  positive  derivative  of  the  zero(cid:173)\ncrossing of the edge at one pixel with the negative derivative at a neighboring  pixel.  The \ndirection of motion  is  given  by  the  neighboring  pixel  from  which  the  edge  disappeared. \nProvided that motion  has  been  detected  at  a  pixel,  the  transfer  time  of the  edge  over the \npixel's finite geometry is inversely proportional to its speed. \n\nEquation (4) gives the mathematical representation of the motion  detection  process  for  an \nobject  moving  in  +x  direction.  In  the  equation.  f,(l.'k,y.t)  is  the  temporal  response  of \npixel k as the zero crossing of an edge of an  object  passes  over  its  2a  aperture.  Equation \n(4) gives the direction of motion,  while equation (5) gives the speed.  The schematic of \n\nmotion  _ x = [ f f,( l: k, y, t)  >  0] [ f f  t(l.'  k  + J,  y,  t)  < 0] = 0 \nmotion+x=[~f,(l.'k-J,y,t)<O][~f/l.'k , y,t\u00bbO] \n\n( a) \n\n(b) \n\n(4) \n\n2a(k-n)-a \n\n= 8[t  -\n\nv \nx \n2a(k -n) -a \n\n]8[x - 2ak] \n\nDisappear \n\n2a(k -n) +a \n.'  t d  = --~--\u00ad\n\nvx \n\nMotion.'  t  = m \n\nv x \n\nv x \n2 a \n\nJ \n- t \n\nSpeed  + x  = t \nd \nthe  VLSI  circuit  of the  motion  detection  model  is  shown  in  figure  4(a).  Figure  4(b) \nshows reciprocal of the measured motion pulse-width for  1 D motion.  The on-chip speed, \net, is the projected target speed.  The measured pulse-widths span 3-4 orders magnitude, \n\nm \n\n(5) \n\n\f710 \n\nR.  ETIENNE-CUMMINGS, J.  VAN DER SPIEGEL, P.  MUELLER \n\nOne-Over Pulse-Width vs On-Chip Speed \n\nO.R \n\n\u2022 \n\n~  0.4 \n~ -0.0  +--------::II~-----__+ \nM \n~ -0.4 \n\n\" \n\n-0 .8 \n\n---e-- \\IPW_Lefi \n\n- -. - - IIPW_ Rlght \n\nLeft \n\nRight \n\n(a) \n\n- 1.2  +-'----'--''-+--'--'--'--t---'--''--'-+-'--'--'-t-'--'--'-t---'--'--'-+ \n12.0 \n\n8.0 \n\n00 \n\n-40 \n4.0 \nOn-Chip Speed rcml~J \n\n-12.0 \n\n-R.O \n\n(b) \n\nFigure  4: \nMeasured Output of the Motion Detection Circuit. \n\n(a)  Schematic  of  the  Motion  Detection  Circuit. \n\n(b) \n\ndepending  on  the  ambient  lighting,  and  show  less  than  15%  variation  between  chips, \npixels, and directions (Etienne-Cummings,  1993). \n\n3.3  THE  SMOOTH  PURSUIT  CONTROL  SYSTEM \n\nThe  one  dimensional  smooth  pursuit  system  is  implemented  using  a  9  x  I  array  of \nmotion  detectors.  Figure  5  shows  the  organization  of the  smooth  pursuit  chip.  In  this \nsystem, only diverging motion is computed to reduce the size of each pixel.  The  outputs \nof the  motion  detectors  are  grouped  into  one  global  motion  signal  per  direction.  This \ngrouping  is  performed  with  a  simple,  but  delayed,  OR,  which  prevents  pulses  from \nneighboring motion  cells  from  overlapping.  The  motion  pulse  trains  for  each  direction \nare XOR  gated,  which  allows  a  single  integrator  to  be  used  for  both  directions,  thus \nlimiting  mis-match_  The  final  value  of the  integrator  is  inversely  proportional  to  the \ntarget's  speed.  The  OR  gates  conserve  the  direction  of  motion.  The  reciprocal  of  the \nintegrator voltage is  next computed using the  linear mode  operation  of a  MOS  transistor \n(Etienne-Cummings,  1993).  The  unipolar  integrated  pulse  allows  a  single  inversion \ncircuit to  be used for  both directions of motion, again limiting mis-match.  The output  of \nthe  \"one-over\"  circuit  is  amplified,  and  the  polarity  of  the  measured  speed  is  restored. \nThis analog voltage is proportional to retinal speed. \n\nThe measured retinal speed is subsequently  ailed to  the  stored  velocity.  Figure  6  shows \nthe schematic for the retinal velocity accumulation (positive feedback) and storage (analog \n\nWave Forms \n\nMotion Pulse Integration \n\nand \"One-Over\" \n\nV =  GIRetinal Velocityl \n\nPolarity \n\nRestoration \n\nRetinal Velocity \n\nAccumulation \n\nand Sample/Hold \n\nFigure 5:  Architecture of the VLSI Smooth Pursuit System.  Sketches \nof  the  wave  forms  for  a  fast  leftward  followed  by  a  slow  rightward \nretinal motion are shown. \n\n\fVLSI  Model of Primate Visual Smooth Pursuit \n\n711 \n\nmemory).  The output of the XOR gate in  figure  5 is used  by  the  sample-and-hold  circuit \nto  control  sampling  switches  S I  and  S2.  During  accumulation,  the  old  stored  velocity \nvalue,  which  is  the  current  eye  velocity,  is  isolated  from  the  summed  value.  At  the \nfalling  edge of the XOR output, the  stored  value  on  C2  is  replaced  by  the  new  value  on \nCl.  This  stored  value  is  amplified  using  an  off chip  motor  driver  circuit,  and  used  to \nmove  the  chip.  The  gain  of  the  motor  driver  can  be  finely  controlled  for  optimal \noperation. \n\nMotor \nSystem \n\nRetinal \nVelocity \n\nAccumulatiun \n\nTwo Phase Sample/Hold \n\nTarget \nVelocity \n\nFigure 6:  Schematic Retinal Velocity Error Accumulation, Storage  and \nMotor Driver Systems. \n\nFigure 7(a) shows a plot of one-over the  measured  integrated  voltage  as  a  function  of on \nchip target speed.  Due to noise in the integrator circuit, the dynamic range  of the  motion \ndetection system is reduced to 2 orders of magnitude.  However, the matching between left \nand  right  motion  is  unaffected  by  the  integrator.  The  MaS  \"one-over\" circuit,  used  to \ncompute  the  analog  reciprocal  of the  integrated  voltage,  exhibits  only  0.06%  deviation \nfrom  a  fitted  line  (Etienne-Cummings,  1993b). \nFigure  7(b)  shows  the  measured \nincrements in  stored target velocity as a  function  of retinal (on-chip) speed.  This is  a test \nof all  the circuit  components  of the  tracking  system.  Linearity  between  retinal  velocity \nincrements  and  target  velocity  is  observed,  however  matching  between  opposite  motion \nhas  degraded.  This  is  caused  by  the  polarity  restoration  circuit  since  it  is  the  only \nlocation  where  different  circuits  are  used  for  opposite  motion.  On  average,  positive \nincrements are a factor of 1.2 times larger than negative increments.  The error bars shows \nthe  variation  in  velocity  increments  for different  motion  cells  and  different  Chips.  The \ndeviation  is  less  than  15  %.  The  analog  memory  has  a  leakage  of  10  mV/min  and  an \nasymmetric  swing  of  2  to  -1  V,  caused  by  the  buffers.  The  dynamic  range  of  the \ncomplete smooth pursuit system is measured to be  1.5 orders magnitude.  The maximum \nspeed of the system is  adjustable by varying the integrator charging time.  The maximum \nspeed is ambient intensity dependent and ranges from 93 cmls to 7 cm/s on-chip speed in \n\nIntegrated Pulse vs On-Chip Speed \n\nVelocity Error Increment  vs On-Chip Speed \n\n24 \n\n~ 16 \n\n.' \n\n._ \n\n~  8 \n~  0  -t--------\",/II!...------+ \nil \n\u00a3 \noS \n::  -16 \n\n-8 \n\n-24 \n\n-e--lnlPuI~_l..xft \n\n_ _ \u2022  _  JntPlllo;e_Rl~hl \n\n-32  -t-'---'---'-'--+-'--~~-t--'\"-'-~_t_--\"--''---'---\"-t \n10.0 \n\n-100 \n\n-5.0 \n\n0.0 \n\n5.0 \n\nOn-Chip Speed lemlsl \n\n(a) \n\n1.4 \n\n1.2 \n\n~ \nl'!  1.0 \n\" e \n~  O.R \nu .s g 0 .6 \nLLl \n.::;. \ng 04 \nOJ \n> \n\n02 \n\n0.0 \n\n0 \n\n----. Nc~_Jncrt~nl \n__ \u2022  _ _ Po,,_Incremclll \n\n4 \n\n6 \n\nOn-Chip Speed  lem/s) \n\n10 \n\n(b) \n\nFigure 7.  (a)  Measured  integrated  motion  pulse  voltage.  (b)  Measured \noutput for the complete smooth pursuit system. \n\n\f712 \n\nR.  ETIENNE-CUMMINGS, J.  VAN DER SPIEGEL, P.  MUELLER \n\nbright  (250  JlW/cm 2)  and  dim  (2.5  JlW/cm 2)  lighting,  respectively.  However,  for  any \nmaximum  speed chosen, the minimum  speed  is  a  factor  of 0.03  slower.  The  minimum \nspeed  is  limited  by  the  discharge  time  of  the  temporal  differentiators  in  the  motion \ndetection circuit to 0.004 cmls on chip.  The contrast sensitivity of this system  proved  to \nbe the stumbling block,  and it can not track objects in  normal  indoor lighting.  However, \nall  circuits  components  tested  successfully  when  a  light  source  is  used  as  the  target. \nAdditional measured data can be found in (Etienne-Cummings,  1995).  Further  work  will \nimprove  the  contrast  sensitivity,  combat  noise  and  also  consider  two  dimensional \nimplementations with target acquisition (saccades) capabilities. \n\n4  CONCLUSION \nA model for  biological  and  silicon  smooth  pursuit  has  been  presented.  It  combines  the \nnegative feed  back  and  positive  feedback  models  of Robinson  and  Wyatt and  Pola.  The \nsmooth  pursuit  system  is  stable  if  the  gain  product  of the  retinal  velocity  detection \nsystem  and  the  eye  movement  system  is  less  than  2.  VLSI  implementation  of  this \nsystem has been performed and tested.  The performance of the  system  suggests  that  wide \nrange (92.9 - 0.004 cmls retinal speed) target tracking is possible with a  single  chip  focal \nplane  system.  To improve  this  chip's  performance,  care must  be  taken  to  limit  noise, \nimprove matching and increase  contrast  sensitivity.  Future  design  should  also  include  a \nsaccadic component to re-capture escaped targets, similar to biological systems. \n\nReferences \n\nC.  Bandera,  \"Foveal  Machine  Vision  Systems\",  Ph.D.  Thesis,  SUNY  Buffalo,  New \nYork,  ]990 \n\nH. Barlow  and  W. Levick,  'The Mechanism  for  Directional  Selective  Units  in  Rabbit' s \nRetina\", Journal of Physiology, Vol.  178, pp. 477-504,  ]965 \n\nT.  Delbruck,  \"Silicon  Retina  with  Correlation-Based,  Velocity-Tuned  Pixels \",  IEEE \nTransactions on Neural Networks,  Vol. 4:3, pp.  529-41,  1993 \n\nM.  Eckert and G. Buchsbaum, \"Effect of Tracking Strategies on the  Velocity  Structure  of \nTwo-Dimensional Image Sequences\", J. Opt.  Soc.  Am.,  Vol.  AIO:7, pp.  1582-85,  1993 \nR.  Etienne-Cummings  et  at.,  \"A  New  Temporal  Domain  Optical  Flow  Measurement \nTechnique  for  Focal  Plane  VLSI  Implementation\",  Proceedings  of  CAMP  93,  M. \nBayoumi, L. Davis and K. Valavanis (Eds.),  pp.  24]-25] ,  1993 \n\nR.  Etienne-Cummings,  R.  Hathaway  and  J.  Van  der Spiegel,  \"An  Accurate  and  Simple \nCMOS 'One-Over' Circuit\", Electronic Letters, Vol.  29-18, pp.  ]618-]620,  1993b \n\nR.  Etienne-Cummings et aI.,  \"Real-Time Visual Target Tracking:  Two Implementations \nof Velocity Based Smooth Pursuit\", Visual Information Processing  IV,  SPIE  Vol.  2488, \nOrlando, 17-18 April  1995 \n\nW. Reichardt, \"Autocorrelation, A Principle for the Evaluation of Sensory Information by \nthe Central Nervous System\", Sensory  Communication, Wiley, New York,  1961 \n\nD. Robinson,  \"The Mechanism of Human  Smooth  Pursuit  Eye  Movement\", Journal of \nPhysiology ( London)  Vol.  180,  pp. 569-591 ,  1965 \n\nM.  Steinbach,  \"Pursuing  the  Perceptual  Rather  than  the  Retinal  Stimuli\",  Vision \nResearch, Vol.  16, pp.  1371-1376,1976 \n\nH.  Wyatt  and  J.  Pola,  \"The  Role  of  Perceived  Motion  in  Smooth  Pursuit  Eye \nMovements\",  Vision  Research, Vol.  19, pp. 613-618, 1979 \n\n\f", "award": [], "sourceid": 1032, "authors": [{"given_name": "Ralph", "family_name": "Etienne-Cummings", "institution": null}, {"given_name": "Jan", "family_name": "Van der Spiegel", "institution": null}, {"given_name": "Paul", "family_name": "Mueller", "institution": null}]}