Download Motion Sensitivity

Survey
yes no Was this document useful for you?
   Thank you for your participation!

* Your assessment is very important for improving the workof artificial intelligence, which forms the content of this project

Document related concepts

Perception wikipedia , lookup

Psychophysics wikipedia , lookup

Artificial intelligence for video surveillance wikipedia , lookup

Perception of infrasound wikipedia , lookup

Neuroesthetics wikipedia , lookup

Neuroscience in space wikipedia , lookup

Visual servoing wikipedia , lookup

Feature detection (nervous system) wikipedia , lookup

Time perception wikipedia , lookup

Transcript
Motion Sensitivity
Visual Motion:
Visual motion, at the most basic level, is a change in the position of a luminance pattern
over time. The velocity is expressed in degrees of visual angle/second.
Humans are very sensitive at detecting motion in visual images. The smallest motion that
humans have been reported to detect is about 10 arc seconds, which is similar to the best
threshold for vernier acuity. Motion detection is therefore like other hyperacuities, the
threshold for the smallest detectable shift is much less than the size of a single cone.
Motion is a correlation across space and time:
In order for us to see visual motion, our brains have to compare the activity of individual
receptors across space and time. One cone cannot see motion! One cone can only see
light increasing or decreasing at it’s location. Motion comes from a comparison of two or
more detectors at different locations, with activity that is a little delayed from one to the
next. See the demo GratingThroughHoles.avi to illustrate what individual cones see
when a grating drifts across the retina.
Apparent Motion:
Although objects in the world move continuously across our visual field, continuous real
motion is not required for the perception of motion. When two stimuli appear with the
correct separation in time and space, our brains link them together into “Apparent
Motion” and we see them as a single moving object. The phenomenon is variously called
Apparent Motion, Sampled Motion, Phi-phenomenon and/or Beta-phenomenon. The last
two terms were coined by Max Wertheimer in his 1912 studies of motion perception. See
demo of Phi phenomenon at http://www1.psych.purdue.edu/Magniphi/SimpliPhi.html
SpatioTemporal Contrast Sensitivity and Motion
When the moving object is a sine wave grating, we can make a meaningful link between
spatial frequency, temporal frequency, and motion. For example, if a 1 cycle per degree
grating drifts at 10 degrees per second, then the luminance at each point in the display is
oscillating sinusoidally at 10 Hz. If a 10 cycle per degree grating drifts at 1 degree per
second, that will also produce 10 Hz flicker at each point. See the demo movies
driftgrat4Hz32cpi.avi driftgrat1Hz8cpi.avi and driftgrat1Hz2cpi.avi. The first two
have the same speed (8 seconds for a bar to cross the picture), but the last two have the
same temporal frequency (1 Hz flicker at each point in the image).
Detection of drifting gratings follows the same pattern as we have seen for flickering
gratings, once converted to temporal frequency. Current thinking is that flicker is actually
detected by motion mechanisms in most cases.
The figure at right shows
Contrast Sensitivity for detecting
drifting gratings of various spatial
frequencies and drift speeds.
Each motion stimulus is plotted
as a function of its equivalent RESEARCH NOTE
temporal frequency. For example,
SPATIAL
a 1 cpd grating moving
at 10 PHASE SENSITIVITY FOR
SINUSOIDAL
GRATING TARGETS
deg/sec makes a 10 Hz
flicker at
each point. A grating that is 0.1 GERALD WESTHEIMER
cpd that moves
at 100
degrees per University of California. Berkeley, CA 94770, U.S.A.
Department
of Physiology-Anatomy.
second also makes a 10 Hz
(Rrceiced 30 Nouember 19771
flicker.
Except
for the slowest
motion (1
Abstract-Because
the full specification of a sinusoidal grating target must include phase as well as
deg/sec)
the curves
all was performed to determine the phase sensitivity for sinusoidal gratings
amplitude,
an experiment
in central vision for normal human observers. Phase displacement thresholds for sinusoidal gratings
superimpose,
showing
that
ranging in spatial frequency from 3 to 25 cycles/degree are a linear function of spatial frequency
(1’ of frequency
phase angle per
c’deg). The just-detectable lateral distances are constant and equal to those
temporal
is the
for a single line.
important
variable for detecting
the moving gratings at contrast
threshold. (Data from Burr &
modulation thresholds for sinusoidal grating targets
This problem was approached by studying the senRoss 1982)
have been investigated thoroughly. but the full specilisitivity of two normal human observers to suddenly
cation of sinusoids of given frequencies requires information
as to both
amplitude
and phase. The nature
Minimum
Motion
Threshold:
of phase sensitivity has important consequences for
thresholds
are a mechanmeasure
theMinimum
identificationmotion
of the underlying
detection
imposed phase changes in suprathreshold sinusoidal
grating patterns. The subject3 task was to identify
the direction of the phase change. Seven stimuli were
of
the smallest
displacement
of aand
pattern
presented
in random
order: no shift;
shifts that
of
one, in
twothe
or range
three times
a basic
moduleand
either
the
typically
of arc
seconds
dotonot
a subject can just identify. Values are
depend much on spatial frequency above 5 cycles per degree.
This figure from Westheimer (1978,
Vision Research) shows minimum
detectable motion for a high contrast
grating pattern that jumps suddenly from
one position to another. Subjects had to
say if the grating jumped left or right,
and the smallest displacement that
produced 75% correct responses was
taken as threshold. All gratings from 5
to 30 cycles per degree showed the same
threshold of about 10 arc seconds.
Thresholds for a single thin line gave
similar results. The upper figure plots
these results as “phase angle” changes,
where one grating cycle = 360° of phase.
The lower figure plots the same data as
displacement in visual angle.
Threshold for seeing the smallest
motion is the same for all spatial
0
IO
20
SPATIAL
FREQUENCY
frequencies
tested(CYCLES/
here.DEGREE)
SPATIAL
FREQUENCY
KYCLh’DEGREEl
7
Fig. 1. (Upper) ~~~~i~~~~~poinrs are the thresholds for detection of sudden change in spatia1 phase
of sinusoidal gratings, 0.5’ high. contrast 50”:. centrally fixated. for two subjects with binocular observation. Results expressed in degrees of phase angle for each grating. Vertical brackets: f 1 standard
error of the mean. Srruighr lines show the mean (solid line) kl standard error (interrupted lines)
for the threshold of detection of direction of a lateral displacement of a single line target. graphed
as the phase angle of its Fourier components. (Lower) Same data, but displacement thresholds are
graphed in the measure of linear distance rather than phase angle.
1073
Motion Aftereffect
Our motion perception shows adaptation effects similar to color, orientation, brightness
and even faces. Stare at a strong stimulus for a period of time, it will seem less strong.
Look away at something neutral, and it will appear to have the opposite sign. For motion,
this means you adapt to one motion for a long time, and when you look away you will see
stationary things appear to move.
The existence of specialized motion sensitive units is dramatically illustrated by the spiral
motion aftereffect. Steady objects will appear to contract after watching an expanding
spiral. Adaptation to the moving stimulus causes an aftereffect of perceived motion
where there is actually only stationary images.
A rotating spiral provides a strong stimulus for
adapting motion detectors. Stare at the center of a
rotating spiral for about one minute to fatigue
motion mechanisms that respond to the expanding
pattern. Then shift gaze to some stationary object
and watch it appear to contract without getting
smaller! See the demo DriftingSpiral.avi
Motion Can Break Camouflage
Our sensitivity to motion can allow us to see things that otherwise would blend in with
the surroundings. Many creatures have colors and patterns that make them blend in with
their background, but when they move our visual system is able to see the shape clearly.
In the laboratory, we can make stimuli that are perfectly camouflaged except for motion
by using completely random colors in the object and the background. These stimuli are
called Random Dot Kinematograms. An example is in the demo movie
RandomDotUHCO.avi
Global Processing in Motion Perception
Our perception of motion in the world is built up from individual detectors at each
location in the visual field. A single local detector may often have ambiguous motion
information, but higher-level mechanisms are usually able to combine all the local
motion signals into a unified percept. There are long range interactions across the visual
field that “bind” local motions together into a single large object moving in one direction.
See the demos DiaMovieG.mov and DiaMovieW.mov for an example of local and
global motion perception being very different.
In the demo movies, the diamond moves left and right behind some occluding bars. If the
bars are visible, the brain connects the four parts of the diamond and sees one object. If
the bars are less obvious, the brain sees four independent lines moving in different
directions. Motion perception requires the brain to put together local motion signals into
one sensible global motion pattern, and sometimes this is ambiguous.
The “Aperture Problem”:
The individual parts of the moving diamond are seen by neurons in the cerebral cortex as
isolated line segments moving through the cell’s “receptive field”: that part of the visual
field that activates the particular cell. Seen through this small aperture, the motion always
appears to be going perpendicular to the line. Only by combing the activity of many such
cells can the brain extract the whole pattern motion.
The white circle represents the receptive field of a hypothetical neuron in primary visual
cortex. This neuron is not sensitive to areas outside the circle (in gray) but only that part
of the visual field that falls inside the white circle. When a straight line or edge moves
through the visual field, the cell sees the direction of motion as perpendicular to the
orientation of the line, regardless of the true direction of the pattern. On the left, the line
moved vertically, on the right it moved horizontally, but the one cell cannot tell the
difference.
Motion processing in cortex:
Area V1 (also called “Area 17”) in cortex is
the primary visual area, and has cells with
motion selectivity, but they cannot solve the
aperture problem. The Middle Temporal
area (“area MT”) in the Temporal lobe
collects the responses of all those V1 cells
and has cells that respond to the whole
pattern, rather than small pieces.
Experiments suggest that this area has the
closest association to our perception of
motion in complex displays.
Damage to this part of the brain leads to a
condition of “motion blindness” or
Akinetopsia. Patients with this condition
will see objects clearly, and be able to
indicate position, but are unable to smoothly follow moving objects or judge their speed.
A nearby, related area called the Middle Superior Temporal area (MST) has similar
sensitivity to global motion, but also includes information about the rotation of the eyes
to help us track moving objects.
Motion on the retina produces a flickering signal in individual cones
(grating through holes movie).
Motion cannot be detected if the local flicker rate is too high or if contrast
is too low.
Neurons in the cortex compare flicker signals to find the local motion.
Local motion is ambiguous (shifting diamond movie).
Local motion is adaptable (spiral aftereffect movie).
Neurons in specialized regions of the cortex compare local motion
signals
to find the global motion.
Global motion lets us identify objects from their motions (UHCO in
random dots).
Damage to these brain areas can make you motion blind.