Psychophysics of Human visual Motion Detection 1. Basic Processes George Mather
Psychophysics of Human visual Motion Detection
1. Basic Processes
George Mather
Questions for today
• Why does visual movement matter?
• Where in the brain is motion processed?
• How is motion detected?
• How can we measure and model motion detection?
What is motion perception?
• The perceptual impression that an object in your field of view is moving or has moved.
What use is motion perception?
Figure-ground segregation
• Relative motion reveals objects
Navigation
Three-dimensional structure
• The kinetic depth effect
Biological form
• Biological motion
• Given the versatility of motion information, it is not surprising that fMRI studies have highlighted a number of cortical regions involved in motion processing (Figure from Culham, He, Dukelow, & Verstraten, 2001).
• There are clinical cases of deficits in active motion perception (attentive tracking) following damage in the parietal lobe (Batelli et al., 2001).
• Deficits in passive (low-level) motion detection following damage in occipital and temporal lobes (eg. V1 and MT; Zihl et al., 1983; Vaina et al., 1998).
Damage to active motion perception
• Active motion perception is based on identifying features in the image and tracking their change in position over time. • Batelli et al. (2001) assessed motion perception in patients with damage in right parietal cortex.• Performance was normal for a low-level detection task:
Active motion perception
• Performance was poor for a high-level apparent motion task.
• We know relatively little about how this attentive tracking process works.
Damage to passive motion perception:A case of motion-blindness
She had difficulty, for example, in pouring tea or coffee into a cup because the fluid appeared to be frozen, like a glacier. She found face-to-face conversations difficult because she could not see the movements of the speaker’s face and mouth. Crowded rooms or streets made her feel unwell, because “people were suddenly here or there but I have not seen them moving”. This problem was particularly acute when attempting to cross a road with moving traffic, although she had no difficulty in actually identifying the cars.
(Zihl et al., 1983)
• We shall concentrate on low-level detection – the blue route.
How to detect retinal movement
• Responses in single receptive fields are ambiguous
How to detect retinal movement
• Responses in pairs of receptive fields must be compared to detect movement.
Motion stimuli in xt space
• A convenient way to depict motion along one spatial dimension.
Image of a
moving car
x-y-t plots of the moving car
Space-time or x-t plots of the moving car
x
y
t
x
t
1 sec
8.3m 5.6m
30 km/h 20 km/h
Receptive fields
• The receptive field of direction-selective neurons can be described as elongated in space-time.
t
x
+ --
+ --
Time
Space
Motion sensors in xt space
• A motion sensor samples motion along a specific orientation (velocity) in xt space to create a spatiotemporal receptive field (Figure from Adelson & Bergen, 1985)
Evidence for spatiotemporal receptive fields: single-cell recordings
1.0 deg.
20
0m
se
c.
• Feline data (left) from McLean & Palmer (1989)
• Primate data (right) from De Valois et al. (2000)
Evidence for spatiotemporal receptive fields: psychophysics
1.0 deg.
20
0m
se
c.
• Masking data from Burr et al. (1986), left.
• Contrast sensitivity data from Watson & Turano(1995), right.
• RFs extend over space and time.
• Two tilted RFs detect leftwardsmotion, two detect rightwards motion.
B
A
F
S
Computational modelling of motion sensors
• RFs extend over space and time.• Two tilted RFs detect leftwards motion, two detect
rightwards motion.• Outputs are squared, normalised and combined.• Normalisation is an elaboration consistent with cortical
physiology and psychophysics (Georgeson & Scott-Samuel, 1999).
• Matlab code is available from my web pages at Sussex University:
• http://www.lifesci.sussex.ac.uk/home/George_Mather/
S
O
SE
TF
TS
Psychophysical evidence for motion detectors
• The motion after-effect.
• See: Mather, Pavan, Campana, Casco (2008)
• Explaining the motion after-effect.
Random Block Kinematograms
RBKs are a
standard
psychophysical
tool for studying
early motion
detecting
processes.
Frame 1
Frame 2
Psychophysics
• Direction discrimination in random dot kinematograms is limited to short displacements and inter-stimulus intervals (ISI). Data from Baker & Braddick (1985).
• Why are there limits on discrimination?
Apparent Motion1.0 deg.
20
0m
se
c.
• Spatiotemporal receptive fields can explain the limits of motion direction discrimination.
Model output and psychophysics
• Model output to RDK was computed in Matlab using standard filter parameters (one row, only one sensor).
• Sensor output is very similar to human psychophysical performance.
• Note the reversal in predicted direction at longer ISIs.
ISI (msec)
Dis
plac
emen
t (d
eg)
0 20 40 60 80 100 120
0.1
0.15
0.2
0.25
0.3
0.35
0.4
0.45
0.5
-0.1
0
0.1
0.2
0.3
0.4
0.5
Model output and psychophysics
• Psychophysical data are taken from Shioiri & Cavanagh(1990, Vision Res., 30, 757-768); again, RDKs.
• Sensor output to the RDK is computed in Matlab using standard A&B filter parameters, using a grey ISI.
-0.15
-0.1
-0.05
0
0.05
0.1
0.15
0.2
0.25
0.3
0.35
0 50 100 150
Mo
tio
n E
ner
gy
ISI (msec)
Two-stroke apparent motion
• Quarter-cycle to-and-fro oscillation in an annular grating appears unidirectional, due to the inter-stimulus interval at one frame transition.
• Generates a MAE.• We can use the adaptation to probe the
filter properties of the motion sensor.
Time
Temporal properties of two-stroke
• Mean MAE duration as a function of two-stroke ISI for 5 subjects (Challinor & Mather, 2010).
• Either photopic luminance (blue) or scotopic luminance (red).
• Peak MAE occurs at longer ISIs at lower luminance.
0.00
2.00
4.00
6.00
8.00
10.00
12.00
14.00
0 50 100 150 200 250 300 350
Mea
n M
AE
Du
rati
on
(se
c)
ISI (msec)
Photopic
Scotopic
Effects of mean luminance
• Dawson & Di Lollo (1990) found that direction discrimination in RDKs extended to longer ISIs at low luminance.
• Billino et al. (2008) found that motion detection was more impaired at high velocity at low mean luminance.
• These results indicate that motion sensors integrate over a longer temporal window at lower luminance.
• How does the receptive field change at low luminance?
Effects of mean luminance
-0.04
-0.02
0.00
0.02
0.04
0.06
0.08
0.0 0.1 0.2 0.3 0.4 0.5
Res
po
nse
Time (sec)
40
50
60
70
80
90
100
110
120
• Varying parameter k in the temporal filter stretches out their response over time.
• In the example plots n and are fixed at 9 and 0.9 respectively. k varies as the parameter.
• We found the k value that best-fitted our MAE data.
Temporal properties of two-stroke
0.00
0.20
0.40
0.60
0.80
1.00
1.20
0.00
1.00
2.00
3.00
4.00
5.00
6.00
7.00
8.00
9.00
10.00
0 50 100 150 200 250 300 350
Mo
tio
n E
ner
gy
Mea
n M
AE
Du
rati
on
(se
c)
ISI (msec)
Photopic
k=90
0.00
0.20
0.40
0.60
0.80
1.00
0.00
1.00
2.00
3.00
4.00
5.00
6.00
7.00
8.00
9.00
10.00
0 50 100 150 200 250 300 350
Mo
tio
n E
ner
gy
Mea
n M
AE
Du
rati
on
(se
c)
ISI (msec)
Scotopic
k=55
• For the best-fitting values at high luminance, the temporal filters have a centre frequency of 4 - 6Hz (2 - 4 Hz at low luminance).
Coding velocity
• Hammett et al. (2005) propose that perceived velocity is given by the ratio of activities in fast (dotted line) and slow (solid line) temporal channels.
• Slow movement excites the slow channel more, fast movement excites the fast channel more:
VELOCITY = FAST/SLOW
Second-order motion: a problem?• Some displays cannot be explained by spatio-temporal receptive fields.
• These are called second-order motion displays.
Frame 1 Frame 2
Contrast Reversal Contrast Reversal Contrast Reversal
SPACE
TIME
Solution: Second-order motion
detectors
• 2nd order detectors respond to spatio-temporal orientation, as do 1st order detectors.
• In 2nd order, detection is preceded by some form of non-linearity that converts texture variation into ‘neural intensity’ variation.
• As a result, 2nd order detectors respond to texture edges rather than luminance edges.
FRF texture segmentation
Image
Filter (orientation-selective)
Rectify
Filter
Psychophysical evidence for second-order detectors
• Data from motion detection in RDKs (Mather & West, 1993)
• See also data from cell responses in monkey cortex (Albright, 1992)
20
30
40
50
60
70
80
90
100
110
0 0.2 0.4 0.6 0.8 1 1.2 1.4
Co
rrec
t (%
)
Displacement (deg)
First-First
Second-Second
First-Second
Levels of processing
• Three levels of processing in motion perception:
Local Motion Detection
Regional Rigid Integration
Global non-rigid structure
Questions for tomorrow
• Why do matrix displays appear tilted?
• Why does sport become more difficult at sunset?
• Why do people drive too fast in fog?
• Why do tennis players challenge line calls?