Lecture 8 - Motion
Visually guided organisms can have issues with object perception:
● Colour blindness (8% of male population)
● Stereo blindness (7-10%, binocular depth vision)
● Form blindness (prosopagnosia)
But motion blindness is very rare because many areas with large receptive fields in both
hemispheres are sensitive to motion, so it is a robust system and more resistant to damage
(bilateral damage is needed for loss of motion).
Motion processing in the retina (convergence of two different points in time) → rabbits, frogs,
turtles, birds, squirrels but not all with the same mechanisms. In humans, motion is not
processed in different anatomical systems.
Motion is a signal from a change of an object’s position over time, but some objects can only be
defined when moving, so it is difficult to say that motion processing depends on object
processing. First order motion (motion based on luminance/change in position) can be
explained by two models:
1. Bilocal correlator/coincidence model → Simple model that only detects motion from
left to right. 2 inputs (photoreceptors) that differ in space, converge onto a summation
collector point. There is a delay mechanism of one of the inputs, which reflects change in
time and the direction of the luminescence motion.
2. Reichart detector model → Complicated model that only detects motion from left to
right, but 2 inputs go to 2 summation collector sites (one excitatory and one inhibitory).
The summation sites are only activated with two incoming signals. When the stimulus is
static or flickers, the whole system is activated and there is no response because the
excitatory and inhibitory outputs cancel each other out. When moving light reaches point
1 first and point 2 after that, the first summation site is activated before the second
summation site → excitation only.
Speed selectivity → changing the delay (δ) or the span between the receptors (s) increases or
decreases the speed of the response.
Detector failure → signals can’t be matched based on luminance/colour. The fact that detector
failure can occur is proof that we actually use a mechanism like the Reichart detector. Second
order motion (motion based on spatial frequencies) suggests that we use additional
mechanisms beyond the Reichart detector as well, as it depends more on interocular transfer
and the binocular system in later visual systems, instead of eye specific events in the early visual
system. There is double dissociation evidence for first and second order motion (
Local motion: small local changes in motion
● V1 (receives input from LGN, V2, MT)
○ 20% direction selective
○ In V2-V3, there are direction selective responses similar to V1, suggesting that it
inherits local motion processing from V1
● Captured by small receptive fields with low convergence, information on the exact
location is maintained → less sensitive when detecting
● Bilocal correlator model
Global motion: large global changes in motion
● MT (mediotemporal area)/hV5 (human V5) (receives input from superior colliculus,
pulvinar and V1-V4)
○ 90% direction selective
, ● Captured by large receptive fields with high convergence, loose information on
specific spatial location of inputs → more sensitive when detecting
● Associated with the perception of motion
Other forms of complex global motion (rotation, expansion, contraction) are processed in the
later MST (medial superior temporal area) with large receptive fields → integration of local and
global motion signals with different directions and speeds.
Aperture problem → motion signals can be ambiguous when viewed through a small aperture
with small receptive fields, the motion can come from different sources. The full stimulus is
required to perceive the correct direction of the motion.
Biological motion occurs spontaneously and aids object recognition by identifying the activity,
species, gender, emotional state, person based on point-light stimuli. In chickens it was shown
that there is an innate preference for biological motion over non-biological motion. In humans
objects need to be learned, so biological motion sensitivity may take years.
Bilateral MT damage affects motion perception. MT neurons respond to a specific direction of
motion. In a study rhesus monkeys looked at moving dots and indicated in a visual display
whether the motion was towards the left or right. This was done with different motion
coherence:
● 0% coherence → all dots moved randomly without direction (monkey guesses, 0.5 score)
● 100% coherence → all dots moved in the same direction
The MT neurons that have a preferred direction for upper right were microstimulated → the
monkey perceived the motion as upper right more often (even at 0% coherence). So MT neurons
do not only fire when you look at motion, but there is also causality (activity of individual MT
neurons contribute to perceptual judgments about visual motion).
Motion after effect → after viewing a moving stimulus, stationary objects appear to move in the
opposite direction. This can be explained by motion opponency models:
● Ratio model (MT) → the perception depends on the ratio between neurons for the
motion direction (adaptor) vs. neurons for the opposite direction. The neurons for the
opposite direction do not actually fire more, it is a relative effect.
1. You look at a waterfall (downward motion), so down-sensitive neurons are highly
active and up-sensitive neurons are at baseline level
2. Down-neurons are depleted and activity decreases (adaptation)
3. Stimulus stops → down-neurons go under baseline and up-neurons stay the
same, so up-neurons are now relatively stronger
4. You perceive a motion-after effect in the upward direction
● Disinhibition model (V1) → the perception depends on inhibition between neurons
sensitive for opposite directions, it is an absolute effect.
1. You look at a waterfall (downward motion), so down-sensitive neurons are highly
active and inhibit up-sensitive neurons at the same time
2. Down-neurons are depleted and activity decreases (adaptation)
3. Stimulus stops → down-neurons go under baseline (depolarization) and
up-neurons are disinhibited, so go above baseline
4. You perceive a motion-after effect in the upward direction
There is evidence for both models, and they do not contradict each other since the ratio-model
refers to MT-neurons, and the disinhibition-model refers to V1-neurons. When adapting to motion
signals to the upper left and upper right, these models would both predict a perceived motion
after effect to the lower left and lower right (but with different explanations). However, the actual
, motion after effect (straight downwards) does not match these predictions since these models
only focus on motion opponency.
● Distribution-shift model → before adaptation, all motion directions are equally
responded to (net direction = 0). After adaptation, there are weakened responses to
specific motion directions, and relatively stronger responses to all other directions. The
brain takes the average of all directions (not only opposite directions) to make its
prediction. This is in line with the ratio-model more, because there is no disinhibition
effect needed for the motion-after effect.
Visually guided organisms can have issues with object perception:
● Colour blindness (8% of male population)
● Stereo blindness (7-10%, binocular depth vision)
● Form blindness (prosopagnosia)
But motion blindness is very rare because many areas with large receptive fields in both
hemispheres are sensitive to motion, so it is a robust system and more resistant to damage
(bilateral damage is needed for loss of motion).
Motion processing in the retina (convergence of two different points in time) → rabbits, frogs,
turtles, birds, squirrels but not all with the same mechanisms. In humans, motion is not
processed in different anatomical systems.
Motion is a signal from a change of an object’s position over time, but some objects can only be
defined when moving, so it is difficult to say that motion processing depends on object
processing. First order motion (motion based on luminance/change in position) can be
explained by two models:
1. Bilocal correlator/coincidence model → Simple model that only detects motion from
left to right. 2 inputs (photoreceptors) that differ in space, converge onto a summation
collector point. There is a delay mechanism of one of the inputs, which reflects change in
time and the direction of the luminescence motion.
2. Reichart detector model → Complicated model that only detects motion from left to
right, but 2 inputs go to 2 summation collector sites (one excitatory and one inhibitory).
The summation sites are only activated with two incoming signals. When the stimulus is
static or flickers, the whole system is activated and there is no response because the
excitatory and inhibitory outputs cancel each other out. When moving light reaches point
1 first and point 2 after that, the first summation site is activated before the second
summation site → excitation only.
Speed selectivity → changing the delay (δ) or the span between the receptors (s) increases or
decreases the speed of the response.
Detector failure → signals can’t be matched based on luminance/colour. The fact that detector
failure can occur is proof that we actually use a mechanism like the Reichart detector. Second
order motion (motion based on spatial frequencies) suggests that we use additional
mechanisms beyond the Reichart detector as well, as it depends more on interocular transfer
and the binocular system in later visual systems, instead of eye specific events in the early visual
system. There is double dissociation evidence for first and second order motion (
Local motion: small local changes in motion
● V1 (receives input from LGN, V2, MT)
○ 20% direction selective
○ In V2-V3, there are direction selective responses similar to V1, suggesting that it
inherits local motion processing from V1
● Captured by small receptive fields with low convergence, information on the exact
location is maintained → less sensitive when detecting
● Bilocal correlator model
Global motion: large global changes in motion
● MT (mediotemporal area)/hV5 (human V5) (receives input from superior colliculus,
pulvinar and V1-V4)
○ 90% direction selective
, ● Captured by large receptive fields with high convergence, loose information on
specific spatial location of inputs → more sensitive when detecting
● Associated with the perception of motion
Other forms of complex global motion (rotation, expansion, contraction) are processed in the
later MST (medial superior temporal area) with large receptive fields → integration of local and
global motion signals with different directions and speeds.
Aperture problem → motion signals can be ambiguous when viewed through a small aperture
with small receptive fields, the motion can come from different sources. The full stimulus is
required to perceive the correct direction of the motion.
Biological motion occurs spontaneously and aids object recognition by identifying the activity,
species, gender, emotional state, person based on point-light stimuli. In chickens it was shown
that there is an innate preference for biological motion over non-biological motion. In humans
objects need to be learned, so biological motion sensitivity may take years.
Bilateral MT damage affects motion perception. MT neurons respond to a specific direction of
motion. In a study rhesus monkeys looked at moving dots and indicated in a visual display
whether the motion was towards the left or right. This was done with different motion
coherence:
● 0% coherence → all dots moved randomly without direction (monkey guesses, 0.5 score)
● 100% coherence → all dots moved in the same direction
The MT neurons that have a preferred direction for upper right were microstimulated → the
monkey perceived the motion as upper right more often (even at 0% coherence). So MT neurons
do not only fire when you look at motion, but there is also causality (activity of individual MT
neurons contribute to perceptual judgments about visual motion).
Motion after effect → after viewing a moving stimulus, stationary objects appear to move in the
opposite direction. This can be explained by motion opponency models:
● Ratio model (MT) → the perception depends on the ratio between neurons for the
motion direction (adaptor) vs. neurons for the opposite direction. The neurons for the
opposite direction do not actually fire more, it is a relative effect.
1. You look at a waterfall (downward motion), so down-sensitive neurons are highly
active and up-sensitive neurons are at baseline level
2. Down-neurons are depleted and activity decreases (adaptation)
3. Stimulus stops → down-neurons go under baseline and up-neurons stay the
same, so up-neurons are now relatively stronger
4. You perceive a motion-after effect in the upward direction
● Disinhibition model (V1) → the perception depends on inhibition between neurons
sensitive for opposite directions, it is an absolute effect.
1. You look at a waterfall (downward motion), so down-sensitive neurons are highly
active and inhibit up-sensitive neurons at the same time
2. Down-neurons are depleted and activity decreases (adaptation)
3. Stimulus stops → down-neurons go under baseline (depolarization) and
up-neurons are disinhibited, so go above baseline
4. You perceive a motion-after effect in the upward direction
There is evidence for both models, and they do not contradict each other since the ratio-model
refers to MT-neurons, and the disinhibition-model refers to V1-neurons. When adapting to motion
signals to the upper left and upper right, these models would both predict a perceived motion
after effect to the lower left and lower right (but with different explanations). However, the actual
, motion after effect (straight downwards) does not match these predictions since these models
only focus on motion opponency.
● Distribution-shift model → before adaptation, all motion directions are equally
responded to (net direction = 0). After adaptation, there are weakened responses to
specific motion directions, and relatively stronger responses to all other directions. The
brain takes the average of all directions (not only opposite directions) to make its
prediction. This is in line with the ratio-model more, because there is no disinhibition
effect needed for the motion-after effect.
