False alarms + missed real ones

Support and queries relating to all previous versions of ZoneMinder
jamescollings
Posts: 59
Joined: Wed Nov 22, 2006 11:26 am
Location: Bucks, UK

Post by jamescollings » Wed May 09, 2007 2:40 pm

robi wrote:The events I'd be interested would be typical: anyone approaching the car you see on the picture. I wouldn't mind falsies like motion around the other car on the right, or headlights reflection from other cars passing by or things like that.
Looking at the image then,
anyone approaching the car :
people are not going to disrupt the bottom section of the image, their feet are likely to be around 10% from the bottom of the image with their head around half-way up. This is basically the region covering the main tarmac'd area up to the number plate of the car. I would set up 3 regions (in line with the parking bays) at this range. If you feel that someone might approach the car by sneaking along the back wall, then you will need an extra (3) region that covers that as well. Since the posts are likely to reflect headlights of other cars, give each of these their own region. If your car is moved, then its primary motion will be FORWARD, and as such it won't alter the image too much, give the car it's own region and be prepared NEVER to get the settings for it right!

As you will have guessed, *my* approach would be to divide your image into Loads of regions (at least 9), and set them ALL to active with the highest sensitivity setting that a zone has by default. This will generate LOADS of events (90-99% false), but then you'll be able to see what the TRUE events do to each of your regions. You'll also be able to identify zones that only trigger when there is a FALSE event (and you can therefore set to preclusive).

User avatar
robi
Posts: 477
Joined: Sat Mar 17, 2007 10:48 am

Post by robi » Thu May 10, 2007 7:50 am

jamescollings wrote:I assume that you have the STATS turned on? If so, you could try triggering an alarm manually, and then acting the part of the "target" yourself. i.e. Trigger an alarm, then walk/drive in front on the camera in the manner that you want to cause a "real" event. Then you can see from the stats what a "REAL" event looks like (in terms of the change in pixel values, % of the zone that is matched etc). It is conceivable that you will be able to filter out the FALSE events by using the "MAX" value (for the alarmed/filtered/blob area).
I did this experiment, however it did not save any stats for the frames!
I have Stats on in the Options, but all my frames are labelled "There are no statistics recorded for this event/frame" :(

Perhaps for manually triggered events stats are not recorded at all. :?
Last edited by robi on Thu May 10, 2007 8:06 am, edited 1 time in total.
v1.25.0 + Ubuntu Linux 12.04 Server

User avatar
robi
Posts: 477
Joined: Sat Mar 17, 2007 10:48 am

Post by robi » Thu May 10, 2007 7:59 am

jamescollings wrote:Looking at the image then...
Thanks for the tips, these are good.
However, what really annoys me is not the big amout of false alarms, but the lack of real ones. I first need to make sure all movement is detected and then start filtering out the cause of the falses.

The problem is that real events get missed, like these:
Image
Image

The images above are the first, triggering frames of the events in cause. They occured too late (movement around the car started way earlier, the guy came into picture from left) while the system only detected them at this frame in time. Too late. (Details on the settings while these pictures were taken in my first post in the topic)

I often miss complete events, watching the live stream I can see things happening and the box doesn't recognize motion. (All these while false alams occured by i-don't-know-what I have hundreds)
v1.25.0 + Ubuntu Linux 12.04 Server

jameswilson
Posts: 5111
Joined: Wed Jun 08, 2005 8:07 pm
Location: Midlands UK
Contact:

Post by jameswilson » Thu May 10, 2007 1:32 pm

is your box public, wanna pm me a login so i can setup a new monitor and have a go myslef?
James Wilson

Disclaimer: The above is pure theory and may work on a good day with the wind behind it. etc etc.
http://www.securitywarehouse.co.uk

User avatar
robi
Posts: 477
Joined: Sat Mar 17, 2007 10:48 am

Post by robi » Thu May 24, 2007 12:54 pm

:?: :roll:
v1.25.0 + Ubuntu Linux 12.04 Server

User avatar
ammaross
Posts: 61
Joined: Mon Mar 12, 2007 8:34 pm
Location: Utah, USA

Post by ammaross » Thu May 24, 2007 8:49 pm

Hopefully I can help you out Robi. I was auto-banned when I tried to respond to your posting about 3 posts into it... :?

Anyway, I noticed your zones are set too sensitive.
Image
This outlines an event will trigger if the pixel threshold exceeds 16 for 9 or more pixels in a blob for one or more blobs. Easily managed by even slight lighting changes. This is probably why you are experiencing false alarms.

Your preclusive zones are set up oddly too. You'll have to show us your settings for them, but putting them at the tops of the pillars like that will cause any 'car passing' events to be prevented, simply because of the light on the pillars. This could potentially cause 'real' events to be skipped. Try deleting the preclusive zones and you'll get your real events back.

I have a 640x480 feed that is set up for blob detect with a pixel threshold of 40 and filter height/width of 5/5. Min/Max alarmed area is 1800 & 0 (you'll probably do fine with 500-800 & 0). Zero means no cap. I used the same numbers for filtered and blob area with a blob of 1 (since it is ment to capture one car or more). Give these blob settings a try and see what happens (without preclusive zones!) We can add them back later if needed.

User avatar
cordel
Posts: 5210
Joined: Fri Mar 05, 2004 4:47 pm
Location: /USA/Washington/Seattle
Contact:

Post by cordel » Fri May 25, 2007 1:12 am

He means can it be accessed over the internet?
If so, you can create a login for him and PM him the details and he will take a stab at setting up a monitor.

User avatar
robi
Posts: 477
Joined: Sat Mar 17, 2007 10:48 am

Post by robi » Fri May 25, 2007 6:56 am

I sent login info to James, he logged in a few times making some changes but I think he must have forgotten me.... :(
v1.25.0 + Ubuntu Linux 12.04 Server

jameswilson
Posts: 5111
Joined: Wed Jun 08, 2005 8:07 pm
Location: Midlands UK
Contact:

Post by jameswilson » Fri May 25, 2007 7:21 pm

whoops
i did forget you, how is it going anyway?

Ill try and log in tomorrow and see whats going on
James Wilson

Disclaimer: The above is pure theory and may work on a good day with the wind behind it. etc etc.
http://www.securitywarehouse.co.uk

caseystone
Posts: 98
Joined: Fri Feb 25, 2005 3:41 am
Location: England

Post by caseystone » Sat May 26, 2007 6:01 am

I'm not much of an expert here either, but I'd try making those inactive zones and not preclusive zones!

-Casey

User avatar
robi
Posts: 477
Joined: Sat Mar 17, 2007 10:48 am

Post by robi » Sat May 26, 2007 1:30 pm

jameswilson wrote:whoops
i did forget you, how is it going anyway?

Ill try and log in tomorrow and see whats going on
Thanks!
v1.25.0 + Ubuntu Linux 12.04 Server

User avatar
robi
Posts: 477
Joined: Sat Mar 17, 2007 10:48 am

Post by robi » Sat May 26, 2007 1:31 pm

caseystone wrote:I'm not much of an expert here either, but I'd try making those inactive zones and not preclusive zones!

-Casey
Currently those zones don't exist anymore, James is helping me in setting up things in a completely new way.
v1.25.0 + Ubuntu Linux 12.04 Server

electron
Posts: 36
Joined: Fri Aug 19, 2005 6:43 am

Post by electron » Sat May 26, 2007 8:15 pm

Yea, I agree the pillars would be best as inactive, simply because they reflect so much light from far away, but I'm not the resident expert here, just want to offer my advice, maybe it will help. I don't think you can ever get reliable detection with that camera you are using at nite, in that garage without additional lighting. If you could get it set up to detect good for dark at nite like that, I'd bet you'd need to setup daytime states & nite time states & have ZM switch between the two, to have optimum settings for day/nite. I do think you can get reliable detection when it's not so dark, with that camera. I'd look at possibly an IR light source you could add, or maybe one of those cheaper security cams sold everywhere with build in IR lights. The other reason is, even if someone steals or causes damage to that car, if they do it at nite, you'll have very poor video because the scene is so dark. I do as others say & set it to sensitive at first, then back off a little. I think of it as your not really detecting motion, but lighting changes, so no it's not perfect, but very very good, with a little tweaking. I'd also reccomend you make videos & figure out how to make good videos now, & export frames etc... so if you really have a need to hand a video to the police, you already have it tested.
Also consider a cam with real auto iris lens. I know that may not be what you want to hear, but if you goal is to protect that car as best as possible, that would help a lot, it will provide a much greater dynamic range (contrast) of the video signal at nite, making motion detection way easier to setup in zm.

Post Reply

Who is online

Users browsing this forum: No registered users and 7 guests