Zm Zones
From:     https://wiki.zoneminder.com/Understanding_ZoneMinder%27s_Zoning_system_for_Dummies 



Understanding ZoneMinder's Zoning system for Dummies





Draw A Zone
Re: Ubuntu - How to draw the zone box???

Post by GivingItMyBest ยป Tue Jun 23, 2020 3:57 pm
When you click add new zone, it should have the entire box highlighted in RED, 
with a green dot in each corner. You can move these dots to reshape the box; 
and at the bottom of the page you will see where each dot has it's x and y 
coordinates, and next to each co-ordinate is a plus and a minus sign. If you 
need a shape other than square, you can add points in between existing points 
and then move those. If you hover over the green dots, they will tell you what 
point/vertex number they are so you can see which two points you need to add 
between.




Contents




Background
ZoneMinder has a flexible (albeit hard to easily configure) zone detection
system using which you can modify how sensitive, precise, accurate your motion
alarms are. The official ZM documentation does a good job of describing all the
concepts here. However, you will see gobs of posts in the forum about people 
complaining that ZM logs all sorts of events (ahem, as did I), ZM's detection is
rubbish and in-camera is better (ahem, as did I) and what not. But once you get
the concept, its incredibly powerful.

So instead of giving you a theoretical explanation, lets walk through a live
use-case. (Credit: user kkrofft helped a lot in me getting a hang of things
here. You should also read his earlier explanation here)




Real object detection
People often ask if ZM supports "object detection". The core ZM engine only
detects change in pixels (motion), which is what this article is about. That
being said, incase you did not know, ZoneMinder now has support for person,
object, face recognition. So while you can and should optimize your zones,
if your real interest is person detection, you can do that using my event
server. Note that object detection works after ZM detects motion change, so
this article applies either way.




Some concepts
Let's take a look at this area below. Lets suppose you want to trigger
motion if someone tries to break into your basement. Does it make sense you
monitor the full area (pillars/walls/floor)? Probably not. If someone were
to break in, they'd break in from some door, some window, or maybe break in
from upstairs and climb down the stairs. So doesn't it make more sense to
monitor these areas specifically? I think so. So the first 'common sense'
logic is delete the default zone that ZM creates for each monitor (which is
called All). Monitoring every part of your image may make sense if you are
monitoring and outdoor lawn, for example. Not here.
	
	Nph-zms.jpeg




Defining the zone areas
So given the explanation above, how about we define zones where motion
matters? Any zone you define as "active" is what ZoneMinder will analyze for
motion. Ignore the 'preclusive for now'. So let's look at the image below.
I've defined polygons around places that are the "entry points"
	
	With zones.jpg




Okay, now how do I specify the sensitivity
of the zones?
ZoneMinder has pre-sets. We live in a world of pre-sets. I bet you want to
select "Best and highly sensitive" don't you? DON'T. Not because that
setting is nonsense, but because you should understand some concepts first.




Core Concepts
The ZM documentation I pointed to earlier does a great job of explaining
different methods. At the cost of repeating what has already been said, it's
important to note:

The first image is a 20x20 grid. Let's assume this is a zone. And the black
circle is some object in this grid. The second image shows the next frame of
that image, where new 'objects' have appeared, or in ZM's view 'new sets of
pixel patterns'

Now let's talk about Alarmed Pixels, Filtered Pixels and Blobs
	

	

	
	

	
	


Reference.jpg Reference next frame.jpg




Alarmed Pixels
Alarmed pixels only deals with pixels changes. If we use the alarmed pixel
method and specify a minimum of "5 pixel" changes (lets forget max for now),
then all the new pixels of set A + B + C + D will count as alarmed pixels
and the total alarmed pixel count will be A+B+C+D




Filtered Pixels
Now let's assume we used Filtered pixels and set it to 2x2 pixels. Then in
addition to computing the alarmed pixels (A+B+C+D), it will also count how
many of these sets have at least 2 pixels around them that are also alarmed
pixels. This will result in B+C+D (set A will be discarded as they don't
have any pixels surrounded by at least 2 pixels that have changed color from
the prev. frame)




Blob Pixels
Now let's assume we used Blob and said a blob needs to be at least 10
pixels. Then what it will do is based on the set computed by Filtered
pixels, which is B+C+D it will look for contiguous blobs of 10 pixels and
that only means D

So, in Alarmed pixels any of A, B, C or D would raise an alarm In filtered
pixel mode, only B, C or D would raise an alarm In blob mode only D would
raise an alarm

Okay, that was a simple explanation. And I did not cover more details on
min/max. But I hope you get the core idea.




Got the theory. Let's get back to your 
basement image
Okay, back to my basement and my 3 zones.




Which detection type should I use?
I personally feel to detect "humans", blob is the best. As I described
above, it combines Alarm + Filtered + ensures that the pixel differences are
contiguous and then does an algorithmic analysis to see if it forms 'blobs'




Pixels or percents?
What makes more sense to you? "Raise an alarm if 178 pixels are changed" or
"Raise an alarm if more than 20% of my zone has changed?". To some, the
latter makes much more sense. However, if you really want more fine grained
control, you should use pixels. I used percent when I first started off, but
then realized that pixels was more powerful when you are trying to eliminate
false alarms.

(Pixels are especially useful if the difference in sizes is small. Example:
its not possible to visualize the difference between 10% and 15%)




Selecting the right values




Using percents

(As of 2019, I recommend pixels, but I've kept this here for those who
prefer percentages)

It helps to think visually here. Let's go back to the zones I drew of my
basement and try and visually place how a person and a pet would look in
each zone. Here is a take: Of men and animals.jpg
	


	
	Stairs foyer.png




Using pixels
(This is what I ended up using once I got comfortable with zones)


Tip: Don't try to get super precise in your first round: Use less aggressive
values to begin with - that is, lesser area values. Keep increasing and testing
till you reach a good threshold between bogus detections and real detections. 
Pixel based detections have a limit. You'll never get it perfect. For real 
object detection as an 'add-on' to motion, look at zmeventnotification, the 
machine learning based alarm detection extension to ZM
	
	Pixel area.jpg


ZM's motion detection algorithm for this zone:





Some other optimizations

There are other optimizations you can do: