Sensitivity vs Threshold
In
video surveillance, sensitivity determines the smallest motion a camera
can detect, while threshold defines the magnitude of that motion needed to
trigger an event. A higher sensitivity setting allows the camera to
register even minor movements, whereas a higher threshold makes it harder to
trigger an event, requiring a more significant amount of motion to activate the
motion detection. You often adjust these together to reduce false alarms
from things like light changes while still capturing real events like a person
or vehicle.
Sensitivity measures how responsive a system is to a stimulus, while the threshold is the minimum stimulus energy or change required to trigger a response, with sensitivity being the reciprocal of the threshold. In simpler terms, a high sensitivity means a low threshold and the system responds to small or subtle changes, whereas a low sensitivity implies a high threshold, requiring a significant change to elicit a reaction.
Sensitivity
·
Definition: The ability of a system or person to detect a
stimulus or respond to a change.
·
What it is: The ease with which a camera detects motion.
·
Measurement: It is often defined as the inverse
of the threshold.
·
Function: A high sensitivity means a system is more
responsive and can detect very weak or small signals/changes.
·
How it works: A higher sensitivity value means the camera's motion detection
algorithm is more "tuned in" to changes in pixels, allowing it to
pick up on smaller or fainter movements.
·
Effect: A high sensitivity can lead to more alerts, as it's more likely to
trigger from small, irrelevant movements.
· Example: A person with high sensitivity to noise might hear a faint sound that others don't.
Threshold
·
Definition: The minimum level of stimulus energy, intensity,
or change that is needed to activate a system or produce a detectable response.
·
What it is: The minimum amount of movement required to
trigger a motion event.
·
Measurement: The specific level of input that is just enough
to cause a reaction.
·
Function: A low threshold means the system is easily
triggered, while a high threshold requires a greater stimulus for a response.
·
How it works: It sets the bar for how much of a motion pixel change must occur
before an alert is generated.
·
Effect: A higher threshold makes the system less likely to trigger, while
a lower threshold will trigger the event more easily.
· Example: A security camera's motion threshold determines the minimum amount of pixel change required to trigger an alarm.
Relationship in Practice
·
Sensitivity and Threshold are Inversely Related:
When sensitivity is high, the threshold is low, meaning less stimulus is
needed for a response. Conversely, when sensitivity is low, the threshold
is high, requiring a stronger stimulus to get a reaction.
·
Adjusting for Performance:
In tasks like motion detection, you adjust these settings together.
ü
Low sensitivity (high threshold): Reduces
false alarms but may miss actual events.
ü High sensitivity (low threshold): Detects more subtle changes but increases the risk of false alerts.
Balancing Sensitivity and Threshold
·
Reduce False Alarms: You can use a combination
of a higher threshold and moderate sensitivity to avoid triggering events from
non-threatening movements (like swaying branches).
·
Capture Key Events: You might use high
sensitivity with a higher threshold to ensure you don't miss real events while
still filtering out minor disturbances.
Threshold value
As
you mentioned the threshold value is a number in % and is the amount of pixels
that is covered by the object relative the total amount of pixels in the
picture. Let’s say that a person cover 15% of the total amount of pixels. Then
the threshold value is 15%. For the system to detect that person the threshold
level must be set to a lower value than 15%, let´s say 10%. Now, if something
covers more than 10% of the pixels in the field of view the system will
acknowledge that as an object of interest. But, that information will solely
not trigger the system for motion detection… Therefore, we need a second
parameter called Sensibility value.
Sensibility value
Sensibility
is a parameter connected to if the object is moving slowly or fast. If the
object is moving at all or if it is moving slowly or fast is determined from
more mathematical calculations during a certain time interval. Let´s say that
at time=1 the calculation gives the first threshold value. A specific time
later at time=2 the second calculation gives the next threshold value and after
time=3 it gives the third threshold value. If all these threshold values are
the same the object is not moving and the system will not generate a motion
trigg to start a recording of a video. But, if it is a difference between the
threshold values something is happening with the object and it gets
interesting. If the differences between the threshold values are small the
object is probably moving slow. If the difference is big the object is probably
moving fast. Let´s say that the system calculates a sensibility value that is
between 0 and 100. A value of 0 means no change in motion, completely still. A
value of 100 means t that the object is moving very fast. In Netcam system the
Sensibility can be set to; very high – normal – very low. If you want to detect
an object that moves very slowly you need to set the Sensibility to very high.
Small changes in the sensibility value should be recognized as important and
tell the system that here is something important going on. If the object is
moving fast the sensibility value is high and it will also be recognized by the
system as important. Well, why not always set the Sensibility to very high
because then we will never miss a moving object? That is the million dollar
question. If you set the Sensibility to very high it will detect everything and
you will have a lot of false alarms! If you set the Sensibility to very low you
will most likely never get a false alarm, but you will never get the moving
object of interest either. Setting the Sensibility correct needs quite often
some testing since it is very dependent of the situation.
‘Sensitivity’ is specified by most camera manufacturers using the ISO 12232 methodology. When using this saturation-based method, a higher ISO value means that it takes less light for the image to reach saturation but does not define how sensitive a sensor is to light (i.e., how many electrons are generated per incident photon) relative to the sensor noise.
ISO 12232 was established to give people an idea how a digital sensor compared to film speed, however it was never intended to give a full range of sensor performance. High-speed cameras are used in applications that demand good performance in low-light environments, which can not be determined by the ISO spec. To achieve a higher ISO rating the display settings of an image can be manipulated, such as by reducing the bit depth or full-well capacity. These specific changes make the image appear brighter but have an adverse effect on image quality and performance.
Consider
the following limitations with the ISO 12232 specification when using it to
compare high-speed cameras:
·
ISO 12232 does not account for noise performance.
Instead, meaningful sensor qualities like Temporal Dark Noise (aka Read Noise)
and Absolute Sensitivity Threshold, which indicates how well the sensor can
identify detail from noise, are key attributes to discerning low light
performance of the camera.
·
ISO is easily manipulated with added gain, which
lowers Signal-to-Noise Ratio (SNR) and lowers Dynamic Range (DR). This
trade-off gets masked because SNR and DR are not always reported, particularly
not in relation to ISO.
·
The rounding factor – Because ISO is based on
film speed, manufacturers are instructed to round up to the closest defined ISO
value above what was measured. This can inflate the value by up to 1/3 F-stop
and is one reason it is not possible to get accurate photon level measurements
with ISO as the basis for sensitivity.
· The light source used for measuring ISO can be Tungsten or Daylight, and a monochrome camera using a Tungsten source will have a much higher spec when an IR filter is not used. Many applications do not have scattered or reflected light matching the spectrum of tungsten or daylight sources. Instead, it is best practice to use Spectral Responsivity plots (or QE curves) to determine how many electrons are generated relative to the number of incident photons across the visible and NIR spectrum. Spectral response curves are provided for all Phantom cameras.
Ultimately, there are too many unknowns to rely on the ISO 12232
specification when comparing sensitivity or any aspect of image quality. Vision
Research has moved away from ISO as the way to spec sensitivity in favor of the
EMVA 1288 standard, providing customers a full set of sensor parameters to
evaluate the camera’s imaging performance.
