Journal of Neuroscience Methods, 43 (1992) 77-82 © 1992 Elsevier Science Publishers B.V, All rights reserved 0165-0270/92/$05.00
77
NSM 01376
A non-contact system for 2-dimensional trajectory recording Christian U r q u i z a r a n d Denis P61isson Vision et Motricitd, INSERM U94, 69500 Bron (France) (Received 21 October 1991) (Revised version received 18 March 1992) (Accepted 3 April 1992)
Key words: 2-Dimensional recording; Trajectory recording; Position detector; Infra-red; Opto-electronic; Computer A new non-contact, real-time, 2-dimensional recording technique is described. This technique used in a cat training procedure, permits on-line monitoring of the position of a hand-held food target displaced at variable speeds, directions and amplitudes in front of the animal. Due to the optical constraints imposed by this training procedure, 2 orthogonally located l-dimensional position-sensitive detectors (PSDs) are used in the near infrared bandwidth in association with a pair of pulsing infrared emitting diodes (IREDs) orthogonally mounted on the monitored object. This geometrical configuration and the wide infrared emission angle insure the visibility of each IRED from its associated PSD, whatever the position of the IRED pair in the working area. A computer program controls the sequence of both IREDs pulses and sampling of the corresponding PSDs outputs and provides the x and y coordinates of the IREDs pair at a 80-Hz rate. This non-contact linear technique can be duplicated at a very low cost for use in a variety of contexts, even under extreme luminance conditions.
Introduction
Current knowledge in spatially oriented motor behavior mainly derives from studies of movements directed at a visual goal. In primate studies, the presentation of small light spots (lightemitting diodes, CRT spot or laser beam projected on a screen) which serve as targets, is usually automatically controlled (e.g., Fuchs and Luschei, 1970; Goldberg and Wurtz, 1972; Mountcastle et al., 1975). Training cats to orient toward visual targets of this kind has been attempted essentially in experiments of gaze ( = visual axis) control (Schlag et al., 1974; Evinger and Fuchs, 1978; Fuller et al., 1983; Fukushima et al., 1987). However, the achievement of a
Correspondance: Denis P61isson, Vision et Motricit6, INSERM U94, 16 av. du Doyen L6pine, 69500 Bron, France. Tel.: (33) 78-54-6578; Fax (33) 72-36-9760.
reliable and accurate goal-directed performance by cats requires a long training period (Evinger and Fuchs, 1978; Fukushima et al., 1987). To circumvent this problem, Guitton et al. (1984) have designed a training procedure in which cats behave in a more "ecological" situation. The hungry animal is placed in front of a vertical opaque screen and is required to look at a food target (spoon filled with moisture food) whenever it appears at one edge of the screen. A few days of training with this hide-and-seek game is sufficient to generate accurate gaze shifts either toward the visible target or in anticipation to its reappearance from behind the screen. A major difficulty with this procedure is to monitor the position of the target which is moved manually in a vertical plane facing the animal. In Guitton's set-up, a search-coil-in-magnetic-field technique (Robinson, 1963) is used to record eye and head movements, and an additional search coil is simply fitted on the hand-held target. How-
78
tween those 2 planes and would mask the target from the sensor's view. Second, the desired high speed of the movements applied by the experimenter to the target renders unpracticable mechanical recording methods, such as those using potentiometers and strings or links attached to the target. Third, a large spatial working range (70 × 70 cm) is necessary for our experiments. We have tested a technique using 4 hall effect sensors (Nienhuis and Siegel, 1989) which fulfilled the first 2 criteria but turned out to be unusable for a working range larger than 40 cm.
ever, under such conditions, the search-coil system provides highly non-linear information which can only serve as an event-detection signal about target motion. We have thus developed a more general non-contact technique, which provides accurate information about the position of an object in a vertical fronto-parallel plane.
Method
Although many recording methods are available, the following constraints impose to develop a new technique. First, the use of bi-dimensional sensors (video and ultrasonic systems of motion analysis) located in a plane parallel to and distant from target motion plane is impossible as the experimenter holding the target is situated be-
Principle Two 1-dimensional position sensitive detectors (PSD) have been chosen with a maximum sensitivity in the near infrared bandwith. PSDs are placed in a vertical plane located 40 cm in front
SENSOR#1
A
I
COMPUTER
i0 ,
/
~, i
' tx2 SENSOR#2
LED ,
i
X~ POSITION OFA
...........
..........working l area
ANALOG X-Y DISPLAY Fig. l. Lay-out of the recording system. Two infrared sensors are placed along the x and y axes of an orthogonal coordinate system (0,x,y,z). The position of a pair of I R E D s (point A) inside the working area (dashed line) is determined in the (O,x,y) plane. The I R E D s are sequentially driven by a PC-like computer which subsequently samples the 2 output signals from each sensor. The computed x and y I R E D position signals are converted back for external use. An "out-of-range" signal drives an indicator L E D when I R E D s are not visible from either sensor (accidental masking or position outside the spatial range in either x, y or z dimension).
79
of the animal, one ( S E N S O R # I ) above and the other ( S E N S O R # 2 ) to the right of the point of origin (point 0 in Fig. 1). The optical axes of the 2 sensors are orthogonal to each other and intercept at point 0. Two infrared emitting diodes ( I R E D 1 and I R E D 2 ) are fitted orthogonal to each other on the monitored object (point A), such that each sensor receives a pulsed infrared b e a m from its associated I R E D ( S E N S O R # 1 and S E N S O R # 2 view I R E D # 1 and I R E D # 2 , respectively). In the optima! condition, the directions of infrared emission are parallel to the 2 axes of the (0,x,y) coordinate system. (We will see below that rotation of the I R E D s pair around z axis do not degrade significatively the measurements due to the wide emission angle of the IREDs). Each sensor provides 2 output voltages (V~ and V 2) which vary with the location of the infrared spot on the PSD-sensitive surface. The angle (alpha) sustended by the infrared b e a m with the normal direction of the sensor can be computed by the following equation t g ( a ) = K * (g= - g l ) / ( g 2 "1- g l ) .
(1)
where K = proportionnality coefficient Knowing the angles ( a 1 and a 2) that the 2 infrared beams sustend with the normal of each sensor, respectively, and the distance (d 1 and d 2) of each sensor to the center of the working area (point 0 in Fig. 1) and, provided the 2 sensors are orthogonal, the horizontal (x) and vertical ( y ) coordinates of the target calculated with respect to the origin 0 are: x = - d I * t g ( a 2 ) * (1 + t g ( a l ) ) / ( t g ( a , ) *
A Sensitive Surface
IRED
O~
-
-
~ VI
Vmf 24x36 camera V2 Current to voltage converter
B kJD Samples
IRED Command Vla v,o
E~ I V2
~
I
T
I delay 100 psec I
tg(c~)=K(V2-V1)/(V2 +V 1)
with Vt = V t a - V l b V2-V2a-V2b
Fig. 2. A: schematic representation of a sensor. An infrared beam is converging through a 28-ram camera lens onto the sensitive surface of the PSD, which produces currents 11 and 12 converted to voltages V, and V 2. B: time sequence of infrared pulse emission and of A / D sampling for an I R E D sensor pair. From top to bottom are represented the computer-generated c o m m a n d to the IRED, and the 2 output voltages from the corresponding sensor. On each channel, baseline values (Vlb and Vzb) are sampled before the I R E D is turned on, and active values (Vxa and gza) are sampled about 75 /~s after the onset of the c o m m a n d pulse, when sensor outputs are stabilized. The equation relating the angle of the incident beam ( a ) to the sensor outputs is shown at the bottom.
tg(a2) )
(2) y = d 2 * tg(a,)*(1
+ tg(ctz))/(tg(a,)*
t g ( a 2 ) ).
(3) Note that, since tg(a 1) and t g ( a z ) are replaced by their actual values obtained from Eqn. 1, no trigonometric computation is required. Construction Each sensor consists of a PSD mounted at the back of a classic 24 X 36 camera and of a wiring providing 2 output tensions (Fig. 2a). The sensor is a 1-dimensional PSD from H a m a m a t s u (type
$1352) with a maximum sensitivity at 900 nm. The rectangular sensitive surface is parallel to the lens ( f = 28 mm, aperture = 2.5) and both the lens and the PSD are parallel to either x ( S E N S O R # 1 ) or y axis ( S E N S O R # 2 ) . The size of the sensitive surface (33 x 2.5 mm) allows an angular spatial range, as viewed through the 28 m m lens, of _+30.5 ° along the x and y axes and of +_2.5° along the z axis. In our configuration (d 1 = d 2 = 124 cm), measurements are restricted to a working area of 90 x 90 cm in the (0,x,y) plane. Translations of the I R E D s along the z axis have no
80
effect on the PSD outputs as long as they remain visible by the PSDs. According to the PSD width, the spatial range along the z axis measured at point 0 is +5.4 cm. This tolerance in the z dimension allows x-y coordinates of an object to be recorded when its movements are not strictly restricted to the (O,x,y) plane. The wiring converts PSD output currents into voltages. The IREDs have been chosen both for their high-pulsed output (400 mW at a 10 -2 duty cycle) and their large emission angle (50% of maximum power was obtained at an angle of 65 ° with respect to the direction of maximum emission). This last point is critical in our application, in which information about object position must be preserved despite small rotations around the z axis. IREDs are fed by a train of 160-/zs pulses, at a rate of 80 Hz, under full computer control. IREDs current intensity (2.6 A) are adjusted to the maximum under the duty cycle of 10 --2. Input signals to the IREDs drivers and output signals from the sensors are interfaced to a PC
A
SENSOR#1
like computer via a T E C M A R Labmaster board (12 bit A / D converter and digital outputs); the calculated x and y position signals are converted back to analog signals (12 bit D / A converters of a M E T R A B Y T E DDA-06 board) for external use (e.g., x-y display, sampling by a general purpose acquisition software). An "out-of-range" LED is lit whenever IREDs are accidentally masked from the view of any sensor. The computer program, developed in C language on a 386-SX based system at 16 MHz, controls the following sequence for each S E N S O R - I R E D pair (Fig. 2b): (1) sample the 2 voltages from sensor ("baseline values" Vjb and V2b); (2) turn on IRED; (3) wait for about 75 ~s ("delay") to compensate for the combined rise-times of the I R E D and PSD outputs; (4) sample the 2 voltages from sensor ("active values" V~a and V2a); (5) turn off IRED; and (6) compute tg(a) from Eqn. 1 with V= = Via - V l b , V 2 = V 2 a - V 2 b . After completion of this sequence for both S E N S O R - I R E D pairs, Eqns. 2 and 3 are corn-
@ 0
e
0
0
0
•
19
o rr
X
EE O
X -91--
SENSOR#1
B
k.L)
cO
m
O
Z uJ o9
Z LU
®
O
Y 15era 15cm
® Y
~
® 100 SOy
r 100SDx Fig. 3. Performance of the system in the central 60 × 60 cm portion of the working area (see Fig. 1 for the position of the 2 sensors with respect to the center of this area). A: linearity of the system shown by the almost perfect match of the calibration grid to the matrix of IREDs physical locations (open circles). B: noise of the system. The size of each ellipse corresponds to a 100-fold magnification of the standard deviation obtained from 10 consecutive measurements of the coordinates of an IRED pair, which physical location corresponds to the center of the ellipse (cross).
81 puted and the results output on 2 D / A channels. This program sequence is controlled by a clock running at 80 Hz, thus providing the object coordinates every 12.5 ms. Sensor output values measured before the I R E D pulse (baseline values) are used as a reference in our calculation in order to preclude any artefact due to variations of background luminance. Constants appearing in the different equations are given the following values: K (Eqn. 1 ) = 1.65 and 1.67 for SENSOR1 and SENSOR2, respectively, as defined by a calibration procedure for each sensor (theoritical value obtained from geometry = 1.7); d~ and d 2 (Eqns. 2 and 3 ) = 124 cm (actual distance from each sensor to origin 0).
Results The performance of the recording system was checked by monitoring the position of an I R E D pair in a 60 × 60 cm central subset of the working area. The I R E D s pair was positioned sequentially to the nearest millimeter at each of the 25 equidistant positions of a 5 x 5 matrix. Ten consecutive measures were taken for each of these positions. A complete set of measures was achieved under normal luminance condition (tungsten bulbs). For the central position of the matrix, 10 other measures were made with the room lights off. Finally, as the system is used conjointly with a search-coil system in our experimental set-up and to assess the effect of magnetic interferences on the output signal, 10 measures were also made with this coil system powered on. Fig. 3a shows the actual (open circles) and mean ( N = 10) measured positions (intersecting points of the grid) of the I R E D s in the central 60 × 60 cm square of the working area. The offsets measured when the I R E D s were at the center (x = - 0 . 5 5 and y = 1.54 cm) have been substracted from all mean values before plotting. There is no consistent non-linearity, and the maximum error is small (0.3 cm). The 10 consecutive measurements made under different experimental conditions when the I R E D pair was at the center provided the following results: room lights and coil system off, x =
- 0.56 _+ 0.04 cm and y = 1.61 +_ 0.05 cm; room lights on and coil system off, x = - 0 . 5 6 _+ 0.03 cm and y = 1.58 _+ 0.08 cm; room lights off and coil system on, x = - 0 . 5 _+ 0.27 cm and y = 1.65 + 0.25 cm. From these figures, one can note the absence of any effect of experimental conditions on the mean values but a 5-fold increase of variability when the coil system is powered on. As well, large rotations of the I R E D pair around the z axis ( + 40°) do not affect the mean estimate of its position but lead to a 2-fold increase in variability. Fig. 3b summarizes the variability in the calculation of I R E D s position in the same 60 x 60 cm area. For each position of the I R E D pair, 10 measures were obtained ( N = 10) with the lights on and the coil system off. The average standard deviation is rather low (x = 0.05 cm and y = 0.06 cm). The standard deviation for each position in the matrix is represented by an ellipse with a magnification factor of 100. This figure shows that the variability is not uniformly distributed over the working area; it is minimum (x = 0.03 cm, y = 0.04 cm) in the upper right quadrant, when I R E D s are closer to both sensors, and progressively increases when the I R E D s are farther away from the sensors (bottom and left areas) to reach a maximum (x = 0.09 cm, y = 0.12 cm) in the lower left position.
Encountered problems The most serious problem encountered has been to find I R E D S which are at the same time powerful enough (to get a sufficient signal/noise ratio at the sensor output level) and not too directive. Regarding possible reflection problems, all reflective surfaces within the field of the sensors have been covered with a dark material. Finally, as the system is to be used in an alternating magnetic field generated by the coil system, electronic circuits have been shielded by foils of permalloy.
Possible improvements Although the performance of the system is highly satisfying, with regard to its cost and to our experimental requirements, better performances can be reached by 2 types of improvements. The
82 first is specific of our application and consists of increasing the protection against alternating magnetic waves from the coil system. This can be achieved by using thicker foils of permalloy a r o u n d wiring and cables. Second, in a m o r e general way, the accuracy o f the sensor o u t p u t signals can still be improved by clearer optics, m o r e powerful I R E D s a n d / o r m o r e resistant to high-duty cycles, and by assemblies of I R E D s to provide a powerful infrared beam. N o t e that, due to the trade-off between spatial and temporal resolution which relates to an inverse relationship between m a x i m u m I R E D s driving current and duty cycle, all these actions can be u n d e r t a k e n in situations requiring either a lower noise level or a higher sampling rate, or both. Total cost
T h e cost of the current, routinely used, configuration (consisting of 2 PSDs, 2 lenses, electronic chips and wiring) is only 5000 FF.
Conclusions T h e main positive features of this system are its linearity and stability, making its use fairly easy without the n e e d for repetitive calibrations, its accuracy, the absence o f any rigid mechanical contact (a single wire to p o w e r - u p the I R E D s ) , and its low cost. In our experimental studies on gaze control in the cat, this m e t h o d provides major improvements in both the training procedure and the off-line quantitative analysis o f behavioral data; it also permits an on-line control of experimental stimulations according to the feedback information about target position. Since this infrared system is i m m u n e to changes of ambient illumination (50-cycle interferences, brisk illumination changes by an electro-mechanical shutter in front o f a light source), it can also be used in its current configuration in m a n y o t h e r contexts
and experimental conditions. For applications requiring a higher spatial a n d / o r temporal resolution, substantial improvements can be m a d e to this configuration, as suggested above.
Acknowledgments W e are greatful to C. Prablanc for helpful c o m m e n t s on the manuscript. T h e technical contribution of P.-M. Chorier and J.-L. Borach is also acknowledged.
References Evinger, C. and Fuchs, A.F. (1978) Saccadic, smooth pursuit, and optokinetic eye movements of the trained cat, J. Physiol., 285: 209-229. Fuchs, A.F. and Luschei, E.S. (1970) Firing patterns of abducens neurons of alert monkeys in relationship to horizontal eye movement, J. Neurophysiol., 33: 382-392. Fukushima, K., Takahashi, K., Fukushima, J. and Kato, M. (1987) Lack of suppression of the short-latency vestibulocollic reflex during active head movements in cats, Brain Behav. Evol., 30: 200-209. Fuller, J.H., Maldonado, H. and Schlag, J. (1983) Vestibularoculomotor interaction in cat eye-head movements, Brain Res., 271: 241-250. Goldberg, M.E. and Wurtz, R.H. (1972) Visual receptive fields of striate cortex neurons in awake monkeys, J. Neurophysiol., 35: 560-574. Guitton, D., Douglas, R.M. and Voile, M. (1984) Eye-head coordination in the cat, J. Neurophysiol., 52: 1030-1050. Mountcastle, V,B., Lynch, J.C., Georgopoulos, A., Sakata, H. and Acuna, C. (1975) Posterior parietal association cortex of the monkey: command functions for operations within extrapersonal space, J. Neurophysiol., 38: 871-908. Nienhuis, R. and Siegel, J.M. (1989) Analysis of head movement and position using hall effect devices, Physiol. Behav., 45: 199-203. Robinson, D.A. (1963) A method of measuring eye movement using a scleral search coil in a magnetic field, IEEE Trans. Bio-med. Electron., 10: 137-145. Schlag, J., Lehtinen, I. and Schlag-Rey, M. (1974) Neuronal activity before and during eye movements in thalamic internal medullary laminae of the cat, J. Neurophysiol., 37: 982-995.
77
NSM 01376
A non-contact system for 2-dimensional trajectory recording Christian U r q u i z a r a n d Denis P61isson Vision et Motricitd, INSERM U94, 69500 Bron (France) (Received 21 October 1991) (Revised version received 18 March 1992) (Accepted 3 April 1992)
Key words: 2-Dimensional recording; Trajectory recording; Position detector; Infra-red; Opto-electronic; Computer A new non-contact, real-time, 2-dimensional recording technique is described. This technique used in a cat training procedure, permits on-line monitoring of the position of a hand-held food target displaced at variable speeds, directions and amplitudes in front of the animal. Due to the optical constraints imposed by this training procedure, 2 orthogonally located l-dimensional position-sensitive detectors (PSDs) are used in the near infrared bandwidth in association with a pair of pulsing infrared emitting diodes (IREDs) orthogonally mounted on the monitored object. This geometrical configuration and the wide infrared emission angle insure the visibility of each IRED from its associated PSD, whatever the position of the IRED pair in the working area. A computer program controls the sequence of both IREDs pulses and sampling of the corresponding PSDs outputs and provides the x and y coordinates of the IREDs pair at a 80-Hz rate. This non-contact linear technique can be duplicated at a very low cost for use in a variety of contexts, even under extreme luminance conditions.
Introduction
Current knowledge in spatially oriented motor behavior mainly derives from studies of movements directed at a visual goal. In primate studies, the presentation of small light spots (lightemitting diodes, CRT spot or laser beam projected on a screen) which serve as targets, is usually automatically controlled (e.g., Fuchs and Luschei, 1970; Goldberg and Wurtz, 1972; Mountcastle et al., 1975). Training cats to orient toward visual targets of this kind has been attempted essentially in experiments of gaze ( = visual axis) control (Schlag et al., 1974; Evinger and Fuchs, 1978; Fuller et al., 1983; Fukushima et al., 1987). However, the achievement of a
Correspondance: Denis P61isson, Vision et Motricit6, INSERM U94, 16 av. du Doyen L6pine, 69500 Bron, France. Tel.: (33) 78-54-6578; Fax (33) 72-36-9760.
reliable and accurate goal-directed performance by cats requires a long training period (Evinger and Fuchs, 1978; Fukushima et al., 1987). To circumvent this problem, Guitton et al. (1984) have designed a training procedure in which cats behave in a more "ecological" situation. The hungry animal is placed in front of a vertical opaque screen and is required to look at a food target (spoon filled with moisture food) whenever it appears at one edge of the screen. A few days of training with this hide-and-seek game is sufficient to generate accurate gaze shifts either toward the visible target or in anticipation to its reappearance from behind the screen. A major difficulty with this procedure is to monitor the position of the target which is moved manually in a vertical plane facing the animal. In Guitton's set-up, a search-coil-in-magnetic-field technique (Robinson, 1963) is used to record eye and head movements, and an additional search coil is simply fitted on the hand-held target. How-
78
tween those 2 planes and would mask the target from the sensor's view. Second, the desired high speed of the movements applied by the experimenter to the target renders unpracticable mechanical recording methods, such as those using potentiometers and strings or links attached to the target. Third, a large spatial working range (70 × 70 cm) is necessary for our experiments. We have tested a technique using 4 hall effect sensors (Nienhuis and Siegel, 1989) which fulfilled the first 2 criteria but turned out to be unusable for a working range larger than 40 cm.
ever, under such conditions, the search-coil system provides highly non-linear information which can only serve as an event-detection signal about target motion. We have thus developed a more general non-contact technique, which provides accurate information about the position of an object in a vertical fronto-parallel plane.
Method
Although many recording methods are available, the following constraints impose to develop a new technique. First, the use of bi-dimensional sensors (video and ultrasonic systems of motion analysis) located in a plane parallel to and distant from target motion plane is impossible as the experimenter holding the target is situated be-
Principle Two 1-dimensional position sensitive detectors (PSD) have been chosen with a maximum sensitivity in the near infrared bandwith. PSDs are placed in a vertical plane located 40 cm in front
SENSOR#1
A
I
COMPUTER
i0 ,
/
~, i
' tx2 SENSOR#2
LED ,
i
X~ POSITION OFA
...........
..........working l area
ANALOG X-Y DISPLAY Fig. l. Lay-out of the recording system. Two infrared sensors are placed along the x and y axes of an orthogonal coordinate system (0,x,y,z). The position of a pair of I R E D s (point A) inside the working area (dashed line) is determined in the (O,x,y) plane. The I R E D s are sequentially driven by a PC-like computer which subsequently samples the 2 output signals from each sensor. The computed x and y I R E D position signals are converted back for external use. An "out-of-range" signal drives an indicator L E D when I R E D s are not visible from either sensor (accidental masking or position outside the spatial range in either x, y or z dimension).
79
of the animal, one ( S E N S O R # I ) above and the other ( S E N S O R # 2 ) to the right of the point of origin (point 0 in Fig. 1). The optical axes of the 2 sensors are orthogonal to each other and intercept at point 0. Two infrared emitting diodes ( I R E D 1 and I R E D 2 ) are fitted orthogonal to each other on the monitored object (point A), such that each sensor receives a pulsed infrared b e a m from its associated I R E D ( S E N S O R # 1 and S E N S O R # 2 view I R E D # 1 and I R E D # 2 , respectively). In the optima! condition, the directions of infrared emission are parallel to the 2 axes of the (0,x,y) coordinate system. (We will see below that rotation of the I R E D s pair around z axis do not degrade significatively the measurements due to the wide emission angle of the IREDs). Each sensor provides 2 output voltages (V~ and V 2) which vary with the location of the infrared spot on the PSD-sensitive surface. The angle (alpha) sustended by the infrared b e a m with the normal direction of the sensor can be computed by the following equation t g ( a ) = K * (g= - g l ) / ( g 2 "1- g l ) .
(1)
where K = proportionnality coefficient Knowing the angles ( a 1 and a 2) that the 2 infrared beams sustend with the normal of each sensor, respectively, and the distance (d 1 and d 2) of each sensor to the center of the working area (point 0 in Fig. 1) and, provided the 2 sensors are orthogonal, the horizontal (x) and vertical ( y ) coordinates of the target calculated with respect to the origin 0 are: x = - d I * t g ( a 2 ) * (1 + t g ( a l ) ) / ( t g ( a , ) *
A Sensitive Surface
IRED
O~
-
-
~ VI
Vmf 24x36 camera V2 Current to voltage converter
B kJD Samples
IRED Command Vla v,o
E~ I V2
~
I
T
I delay 100 psec I
tg(c~)=K(V2-V1)/(V2 +V 1)
with Vt = V t a - V l b V2-V2a-V2b
Fig. 2. A: schematic representation of a sensor. An infrared beam is converging through a 28-ram camera lens onto the sensitive surface of the PSD, which produces currents 11 and 12 converted to voltages V, and V 2. B: time sequence of infrared pulse emission and of A / D sampling for an I R E D sensor pair. From top to bottom are represented the computer-generated c o m m a n d to the IRED, and the 2 output voltages from the corresponding sensor. On each channel, baseline values (Vlb and Vzb) are sampled before the I R E D is turned on, and active values (Vxa and gza) are sampled about 75 /~s after the onset of the c o m m a n d pulse, when sensor outputs are stabilized. The equation relating the angle of the incident beam ( a ) to the sensor outputs is shown at the bottom.
tg(a2) )
(2) y = d 2 * tg(a,)*(1
+ tg(ctz))/(tg(a,)*
t g ( a 2 ) ).
(3) Note that, since tg(a 1) and t g ( a z ) are replaced by their actual values obtained from Eqn. 1, no trigonometric computation is required. Construction Each sensor consists of a PSD mounted at the back of a classic 24 X 36 camera and of a wiring providing 2 output tensions (Fig. 2a). The sensor is a 1-dimensional PSD from H a m a m a t s u (type
$1352) with a maximum sensitivity at 900 nm. The rectangular sensitive surface is parallel to the lens ( f = 28 mm, aperture = 2.5) and both the lens and the PSD are parallel to either x ( S E N S O R # 1 ) or y axis ( S E N S O R # 2 ) . The size of the sensitive surface (33 x 2.5 mm) allows an angular spatial range, as viewed through the 28 m m lens, of _+30.5 ° along the x and y axes and of +_2.5° along the z axis. In our configuration (d 1 = d 2 = 124 cm), measurements are restricted to a working area of 90 x 90 cm in the (0,x,y) plane. Translations of the I R E D s along the z axis have no
80
effect on the PSD outputs as long as they remain visible by the PSDs. According to the PSD width, the spatial range along the z axis measured at point 0 is +5.4 cm. This tolerance in the z dimension allows x-y coordinates of an object to be recorded when its movements are not strictly restricted to the (O,x,y) plane. The wiring converts PSD output currents into voltages. The IREDs have been chosen both for their high-pulsed output (400 mW at a 10 -2 duty cycle) and their large emission angle (50% of maximum power was obtained at an angle of 65 ° with respect to the direction of maximum emission). This last point is critical in our application, in which information about object position must be preserved despite small rotations around the z axis. IREDs are fed by a train of 160-/zs pulses, at a rate of 80 Hz, under full computer control. IREDs current intensity (2.6 A) are adjusted to the maximum under the duty cycle of 10 --2. Input signals to the IREDs drivers and output signals from the sensors are interfaced to a PC
A
SENSOR#1
like computer via a T E C M A R Labmaster board (12 bit A / D converter and digital outputs); the calculated x and y position signals are converted back to analog signals (12 bit D / A converters of a M E T R A B Y T E DDA-06 board) for external use (e.g., x-y display, sampling by a general purpose acquisition software). An "out-of-range" LED is lit whenever IREDs are accidentally masked from the view of any sensor. The computer program, developed in C language on a 386-SX based system at 16 MHz, controls the following sequence for each S E N S O R - I R E D pair (Fig. 2b): (1) sample the 2 voltages from sensor ("baseline values" Vjb and V2b); (2) turn on IRED; (3) wait for about 75 ~s ("delay") to compensate for the combined rise-times of the I R E D and PSD outputs; (4) sample the 2 voltages from sensor ("active values" V~a and V2a); (5) turn off IRED; and (6) compute tg(a) from Eqn. 1 with V= = Via - V l b , V 2 = V 2 a - V 2 b . After completion of this sequence for both S E N S O R - I R E D pairs, Eqns. 2 and 3 are corn-
@ 0
e
0
0
0
•
19
o rr
X
EE O
X -91--
SENSOR#1
B
k.L)
cO
m
O
Z uJ o9
Z LU
®
O
Y 15era 15cm
® Y
~
® 100 SOy
r 100SDx Fig. 3. Performance of the system in the central 60 × 60 cm portion of the working area (see Fig. 1 for the position of the 2 sensors with respect to the center of this area). A: linearity of the system shown by the almost perfect match of the calibration grid to the matrix of IREDs physical locations (open circles). B: noise of the system. The size of each ellipse corresponds to a 100-fold magnification of the standard deviation obtained from 10 consecutive measurements of the coordinates of an IRED pair, which physical location corresponds to the center of the ellipse (cross).
81 puted and the results output on 2 D / A channels. This program sequence is controlled by a clock running at 80 Hz, thus providing the object coordinates every 12.5 ms. Sensor output values measured before the I R E D pulse (baseline values) are used as a reference in our calculation in order to preclude any artefact due to variations of background luminance. Constants appearing in the different equations are given the following values: K (Eqn. 1 ) = 1.65 and 1.67 for SENSOR1 and SENSOR2, respectively, as defined by a calibration procedure for each sensor (theoritical value obtained from geometry = 1.7); d~ and d 2 (Eqns. 2 and 3 ) = 124 cm (actual distance from each sensor to origin 0).
Results The performance of the recording system was checked by monitoring the position of an I R E D pair in a 60 × 60 cm central subset of the working area. The I R E D s pair was positioned sequentially to the nearest millimeter at each of the 25 equidistant positions of a 5 x 5 matrix. Ten consecutive measures were taken for each of these positions. A complete set of measures was achieved under normal luminance condition (tungsten bulbs). For the central position of the matrix, 10 other measures were made with the room lights off. Finally, as the system is used conjointly with a search-coil system in our experimental set-up and to assess the effect of magnetic interferences on the output signal, 10 measures were also made with this coil system powered on. Fig. 3a shows the actual (open circles) and mean ( N = 10) measured positions (intersecting points of the grid) of the I R E D s in the central 60 × 60 cm square of the working area. The offsets measured when the I R E D s were at the center (x = - 0 . 5 5 and y = 1.54 cm) have been substracted from all mean values before plotting. There is no consistent non-linearity, and the maximum error is small (0.3 cm). The 10 consecutive measurements made under different experimental conditions when the I R E D pair was at the center provided the following results: room lights and coil system off, x =
- 0.56 _+ 0.04 cm and y = 1.61 +_ 0.05 cm; room lights on and coil system off, x = - 0 . 5 6 _+ 0.03 cm and y = 1.58 _+ 0.08 cm; room lights off and coil system on, x = - 0 . 5 _+ 0.27 cm and y = 1.65 + 0.25 cm. From these figures, one can note the absence of any effect of experimental conditions on the mean values but a 5-fold increase of variability when the coil system is powered on. As well, large rotations of the I R E D pair around the z axis ( + 40°) do not affect the mean estimate of its position but lead to a 2-fold increase in variability. Fig. 3b summarizes the variability in the calculation of I R E D s position in the same 60 x 60 cm area. For each position of the I R E D pair, 10 measures were obtained ( N = 10) with the lights on and the coil system off. The average standard deviation is rather low (x = 0.05 cm and y = 0.06 cm). The standard deviation for each position in the matrix is represented by an ellipse with a magnification factor of 100. This figure shows that the variability is not uniformly distributed over the working area; it is minimum (x = 0.03 cm, y = 0.04 cm) in the upper right quadrant, when I R E D s are closer to both sensors, and progressively increases when the I R E D s are farther away from the sensors (bottom and left areas) to reach a maximum (x = 0.09 cm, y = 0.12 cm) in the lower left position.
Encountered problems The most serious problem encountered has been to find I R E D S which are at the same time powerful enough (to get a sufficient signal/noise ratio at the sensor output level) and not too directive. Regarding possible reflection problems, all reflective surfaces within the field of the sensors have been covered with a dark material. Finally, as the system is to be used in an alternating magnetic field generated by the coil system, electronic circuits have been shielded by foils of permalloy.
Possible improvements Although the performance of the system is highly satisfying, with regard to its cost and to our experimental requirements, better performances can be reached by 2 types of improvements. The
82 first is specific of our application and consists of increasing the protection against alternating magnetic waves from the coil system. This can be achieved by using thicker foils of permalloy a r o u n d wiring and cables. Second, in a m o r e general way, the accuracy o f the sensor o u t p u t signals can still be improved by clearer optics, m o r e powerful I R E D s a n d / o r m o r e resistant to high-duty cycles, and by assemblies of I R E D s to provide a powerful infrared beam. N o t e that, due to the trade-off between spatial and temporal resolution which relates to an inverse relationship between m a x i m u m I R E D s driving current and duty cycle, all these actions can be u n d e r t a k e n in situations requiring either a lower noise level or a higher sampling rate, or both. Total cost
T h e cost of the current, routinely used, configuration (consisting of 2 PSDs, 2 lenses, electronic chips and wiring) is only 5000 FF.
Conclusions T h e main positive features of this system are its linearity and stability, making its use fairly easy without the n e e d for repetitive calibrations, its accuracy, the absence o f any rigid mechanical contact (a single wire to p o w e r - u p the I R E D s ) , and its low cost. In our experimental studies on gaze control in the cat, this m e t h o d provides major improvements in both the training procedure and the off-line quantitative analysis o f behavioral data; it also permits an on-line control of experimental stimulations according to the feedback information about target position. Since this infrared system is i m m u n e to changes of ambient illumination (50-cycle interferences, brisk illumination changes by an electro-mechanical shutter in front o f a light source), it can also be used in its current configuration in m a n y o t h e r contexts
and experimental conditions. For applications requiring a higher spatial a n d / o r temporal resolution, substantial improvements can be m a d e to this configuration, as suggested above.
Acknowledgments W e are greatful to C. Prablanc for helpful c o m m e n t s on the manuscript. T h e technical contribution of P.-M. Chorier and J.-L. Borach is also acknowledged.
References Evinger, C. and Fuchs, A.F. (1978) Saccadic, smooth pursuit, and optokinetic eye movements of the trained cat, J. Physiol., 285: 209-229. Fuchs, A.F. and Luschei, E.S. (1970) Firing patterns of abducens neurons of alert monkeys in relationship to horizontal eye movement, J. Neurophysiol., 33: 382-392. Fukushima, K., Takahashi, K., Fukushima, J. and Kato, M. (1987) Lack of suppression of the short-latency vestibulocollic reflex during active head movements in cats, Brain Behav. Evol., 30: 200-209. Fuller, J.H., Maldonado, H. and Schlag, J. (1983) Vestibularoculomotor interaction in cat eye-head movements, Brain Res., 271: 241-250. Goldberg, M.E. and Wurtz, R.H. (1972) Visual receptive fields of striate cortex neurons in awake monkeys, J. Neurophysiol., 35: 560-574. Guitton, D., Douglas, R.M. and Voile, M. (1984) Eye-head coordination in the cat, J. Neurophysiol., 52: 1030-1050. Mountcastle, V,B., Lynch, J.C., Georgopoulos, A., Sakata, H. and Acuna, C. (1975) Posterior parietal association cortex of the monkey: command functions for operations within extrapersonal space, J. Neurophysiol., 38: 871-908. Nienhuis, R. and Siegel, J.M. (1989) Analysis of head movement and position using hall effect devices, Physiol. Behav., 45: 199-203. Robinson, D.A. (1963) A method of measuring eye movement using a scleral search coil in a magnetic field, IEEE Trans. Bio-med. Electron., 10: 137-145. Schlag, J., Lehtinen, I. and Schlag-Rey, M. (1974) Neuronal activity before and during eye movements in thalamic internal medullary laminae of the cat, J. Neurophysiol., 37: 982-995.
