Training Note Video Analytics Left Objects
article discusses 3 levels of intelligence for detecting Left Objects which
could be a threat to public safety or an aid to identifying the original owner
by jumping back to the video recording when the event was detected.
- This is
an entry level application and applies to empty space only and certainly
not for any busy environment. It
can detect an object left in an empty scene up to a period which is
shorter than the predefined Motion Retention time (MRT). In case the object is blocked from the
camera, the application will reset the timer and start counting
again. In case we set the period to
be longer than MRT, all left objects will be ignored. MRT pre-definition
is required to eliminate various transient or illumination changes.
- For false alarm reduction, it requires
the size of potential objects to be predefined. This helps to eliminate falling leaves
on the ground, or a big dog lying down.
- It can detect one object at a
time. Other objects that are left
at the same time would be ignored.
can detect objects as small as 8*8 pixels and IQ-180 can detect very small
objects of 4*4 pixels in busy scenes. The application needs to be told to expect the level of business of
the scene and to relearn the scene if it has changed beyond a certain
percentage. Default parameter is
50% for business of scene and an additional 10% change (that is 60%
business) for triggering relearning of the scene. These parameters are different for each
site and tuning is essential to improve accuracy. The point is that the application will
take time to relearn the scene. If
the business level hits 70% in the meantime, the application will miss
certain changes (including left objects). A message will be displayed on the screen in case someone is
monitoring in real time.
- Both applications are capable of
detecting multiple left objects.
- A left
object that has a similar colour to the background is the worst case
scenario for detection (such as a black bag in a shadowed background). The applications are able to detect an
object that is the same colour as the background with 50% difference in
luminance. That ratio comes down
when there is significant difference in colour. A white object against a black
background will require 0% difference in luminance.
- The false
alarm rate depends heavily on the amount of moving traffic and changing
environment. If we deploy this in an indoor environment with minimal
traffic (with 2 to 3 persons through traffic per minute) we would be
looking at close to 0 false alarms. If we deploy this in an outdoor
environment with foot traffic density similar to a busy city street, we
are probably looking at around 1 false alarm per camera per 3 to 4 hours.