Dahua N3wbie , can’t figure out why IVS events won’t record

Specific concerns: Does IVS have issues when the apparent size of the person can be drastically different within a scene / boundary ? If so, should I create multiple boundaries /tripwires within each of which the person size should not be drastically different ?
Should I be using the "inside" function of the boundary instead of cross/appear ? I assume that will keep triggering as long as the target is inside the boundary ?
 
Yiou're walking "Under" the IVS tripwires on the porch

The endpoint should be at the base of the wall

Turn off Vehicle

Do NOT set a min/max size.
Max should be entire screen, min should be 0

Same issues with the Intrusion box. Its drawn too high
 
So, i've been having some issues setting up IVS... and I can't tell if i'm just misusing it or what. I DO get detection, but it seems like it's very iffy on finding humans and when they cross a tripwire or perimeter ... and generally much worse than SMD is at picking up humans... am I using it wrong here , or is the range of possible human sizes in this perspective of a ceiling-porch mount just not going to work well for IVS?...

The way that you have your rules drawn in the yard and drive are exactly what I was talking about in my example above re needing to think in a 3D space. It's likely that someone walking though a good portion of your lines in the yard won't be picked up until they're well within. You'll probably need to extend the lines up so they cover the road. Yep, I understand that's going to get you alerts that you don't want but that's likely how it's going to work with that view. SMD works because it uses the entire view (and also will get unwanted alerts outside of your rule area).

Do you need to distinguish all of the different zones? If so, that's fine. If not and as a first step even if so, I'd simplify things and get it working well one at a time and work
 
  • Like
Reactions: JDreaming
Ok, i'm still struggling to wrap my head around this.. Maybe these questions will help illuminate the confusion i'm having..

For IVS:
  1. Do tripwires / perimeters trigger when any part of the identified human/car crosses the line ?
  2. I assume the system works by #1 Identifying a target (no matter where on screen) and then determining if the target is triggering the rules? My understanding is that the blue boxes that appear on live view is the target track, and the red flashes are when it triggers a rule, correct ? These target track boxes seem to show up for IVS rules but not for SMD for some reason.
  3. How long does an IVS rule keep triggering for after it triggers and finds a valid target .. for example.. say I have a simple perimeter rule, and that it's set to cross/appear.. does the rule keep firing for as long as the target track is kept and until they leave the perimeter, or only on the initial appearance/crossing and IVS doesn't re-arm until it's lost the blue box target track? For me, in synology surveillance station this corresponds to how long the "event" ends up being , for example.
  4. Are there any good tutorial videos that teach IVS ? I feel like I need to see worked interactive/live examples with do/do-nots to get this to click in my head.. and I can't seem to find a good one.. :(
 
1- the IVS line needs to see enough of the human to determine its a human. So no
2- Yes
3- Dont know the reset time. But it does reset after a target is ID'd even if the target is still within the box
4- Not sure, but you might want to refer to mine and others experience and start there
 
  • Like
Reactions: JDreaming
Yes IVS needs to be thought through a bit. I like zig zag tripwires, sometimes encompassed within a bounding Intrusion box. If direction is critical, then zig zag isnt the right way to go, instead use individual tripwire lines, multiples with the same direction filter

There is a 3d effect to IVS lines. Be cautious of drawing lines that targets can simply "walk under".
Example:
Lady being shorter and on the higher side of the Intrusion line has not yet set it off. Taller man, on the shorter segment of the line has tripped the alert
View attachment 198837 View attachment 198838


So, this is a point of confusion for me here.. in the first picture it's locked onto both of them as people (blue boxes), and both people are crossing the intrusion line.. why is it not triggering on the woman too ? Is it because too much of her body has already crossed the intrusion line by the time she's ID'd as a person.. and the algorithm is triggering on something like when 75% of the box crosses ? [I made up 75% as a number just to convey the idea of a weighted "center of mass" crossing being the trigger rather than any part of the box crossing]
 
1. No. It needs to be enough of whatever that the AI recognizes that it meets whatever criteria. i.e., If someone just steps into your intrusion box/across a line, then it may not trigger. It needs to see more of a person's body/vehicle crossing the line/appearing within/whatever.

2. Generally right. As I said, SMD uses the entire view so there are no boxes/lines to light up for it.

3. I believe the former but triggering/alerting is different vs detecting. It only triggers and sends an event alert once for whatever event occurs within whatever time frame. It continues to be in an alerted/red state and detecting that event as long as it sees whatever meeting the detection criteria. It then falls to a clear status. If you download a program called ONVIF Device Manager, and look at the events there, you'll see better how that works. It will detect an event, set the flag for an event, continue in that state, then fall to a non-event status when it recognizes that event is no longer active.

4. I'd have to search too. Sure some are out there.
 
  • Like
Reactions: JDreaming
So, this is a point of confusion for me here.. in the first picture it's locked onto both of them as people (blue boxes), and both people are crossing the intrusion line.. why is it not triggering on the woman too ? Is it because too much of her body has already crossed the intrusion line by the time she's ID'd as a person.. and the algorithm is triggering on something like when 75% of the box crosses ? [I made up 75% as a number just to convey the idea of a weighted "center of mass" crossing being the trigger rather than any part of the box crossing]

It eventually did trigger on her when she crossed the next line, on that picture, she is walking "underneath" the line. Their is a 3d effect as mentioned. The AI recognized her as a human, thus the blue box. She did not at that point "Trigger" because she walked underneath the line. Anyone other than a Dahua engineer that gives you a percentage is guessing
 
  • Like
Reactions: JDreaming
So, this is a point of confusion for me here.. in the first picture it's locked onto both of them as people (blue boxes), and both people are crossing the intrusion line.. why is it not triggering on the woman too ? Is it because too much of her body has already crossed the intrusion line by the time she's ID'd as a person.. and the algorithm is triggering on something like when 75% of the box crosses ? [I made up 75% as a number just to convey the idea of a weighted "center of mass" crossing being the trigger rather than any part of the box crossing]

It probably did at some point. Those lines flash back and forth sometimes. Also, she likely turned the outside intrusion box red.

Yes, I don't think that it's a hard percentage like that but it uses some weighted criteria to determine enough of whatever to trigger.
 
  • Like
Reactions: JDreaming
See Post #22 above
 
  • Like
Reactions: JDreaming
It eventually did trigger on her when she crossed the next line, on that picture, she is walking "underneath" the line. Their is a 3d effect as mentioned. The AI recognized her as a human, thus the blue box. She did not at that point "Trigger" because she walked underneath the line. Anyone other than a Dahua engineer that gives you a percentage is guessing
Just to be clear.. you're saying "walking underneath the line" as a way to describe the AI's behavior of not triggering because she was already too far past the line at the point it ID'd her as human, right ? I think we're saying the same thing here.
 
Im saying there is a 3D effect. Ive given you 2 examples
 
  • Like
Reactions: JDreaming
Im saying there is a 3D effect. Ive given you 2 examples
I've seen the examples, and looked at them.. i'm just having trouble "seeing" the effect... and yes.. I feel stupid here. I'm just trying to internalize what you're communicating... but apparently i'm still not quite getting it. I'm not quite sure what's 3D about any of this ... should I be thinking of the tripwires as needing to be placed on the "ground" of the image in each place rather than in the air? The other advice i'm being given here is to make sure it's tall enough and that my tripwires are too small.. so i'm trying to understand if the feet and legs need to pass the tripwire or if the whole body does
 
Should have said too, the "percentage," for lack of any better way of referencing it, may change for the same rule depending on a variety of factors. e.g., One person/car moving though may trigger it very quickly, another may take a little longer. Contrast, speed and direction of movement, day vs night, etc., all factor into the detection.
 
  • Like
Reactions: JDreaming
I've seen the examples, and looked at them.. i'm just having trouble "seeing" the effect... and yes.. I feel stupid here. I'm just trying to internalize what you're communicating... but apparently i'm still not quite getting it. I'm not quite sure what's 3D about any of this ... should I be thinking of the tripwires as needing to be placed on the "ground" of the image in each place rather than in the air? The other advice i'm being given here is to make sure it's tall enough and that my tripwires are too small.. so i'm trying to understand if the feet and legs need to pass the tripwire or if the whole body does

You're getting closer, but making it too hard

Walk through the lines,. If it hits you higher than waist high, your line is too high

The endpoint should be at the base of the wall

Turn off Vehicle

Do NOT set a min/max size.
Max should be entire screen, min should be 0

Same issues with the Intrusion box. Its drawn too high


There is a difference between extending lines beyond (horizontally) the target area and extending them vertically
 
...so i'm trying to understand if the feet and legs need to pass the tripwire or if the whole body does

Not one foot. Not the whole body. But enough of whatever to be detected. Nobody can really tell you some specific "percentage" since it's dynamic as evaluated by the AI and varies by view. We're just giving you general guidelines that you'll need to apply and test for the specifics of your case.
 
  • Like
Reactions: JDreaming
You're getting closer, but making it too hard

Walk through the lines,. If it hits you higher than waist high, your line is too high

The endpoint should be at the base of the wall

Turn off Vehicle

Do NOT set a min/max size.
Max should be entire screen, min should be 0

Same issues with the Intrusion box. Its drawn too high


There is a difference between extending lines beyond (horizontally) the target area and extending them vertically


So I think you're saying do it something like this (with just the porch as an example) ? I did just try this, but the results were pretty bad with it failing to blue-box me for most of my transit of the porch toward and away from the camera.. and really only caught me some of the time going from stairs towards the door and back.

Is this because of the failure to ID me as human, or because the IVS rules are not great , or both ? I've got the sensitivity turned all the way up to 10 here and it won't see me reliably as shown in this video:

View attachment Test.mov
 
In that image I would guess that its simply too dark and not enough contrast, but yes, right idea

Blue box has little or nothing to do with the actual lines. My cameras blue box possible targets well far away from any rule line.

Lets see your image settings, exposure, etc

AND lets see your recorded timeline and recording rules. I dont care about blue boxes, I care about triggers
 
  • Like
Reactions: JDreaming
The IVS line DID trigger once you hit it. Red Box = trigger.

Yellow lines- extend your line horizontally
Add WDR or otherwise more light on area to provide better contrast

trigger.jpg
 
I would also recommend that the camera isnt in a great place for detection in that corner. I would have placed it to the other side of the front door.

You probably placed it where you did to try and accomplish too much with one camera. A common thing for newbies, I did it too in 2016
You need two cameras, one for the door (placed opposite side of door facing OUT from the dark corner), and one possibly where you have it focused on the steps and yard.