Seeking Assistance with Eliminating False Positives in Motion Detection

General discussion about Blue Iris
User avatar
Sardon
Posts: 111
Joined: Sun Nov 19, 2023 10:50 am

Re: Seeking Assistance with Eliminating False Positives in Motion Detection

Post by Sardon »

I tried Pogo's sensitivity settings 220/44 and that prevented detection by me walking past, which is strange, I presume it's because high res is enabled and not enough pixels. So I reverted back to 100/20 and then the condensation from downstairs started to cause alerts again..

So I've disabled the clone, for now... I've also removed my old zone on main balcony cam (after taking a backup of the camera to .reg) that encompassed the entirely balcony and railings and have created this.
main-mod.jpg
main-mod.jpg (70.32 KiB) Viewed 14701 times
I've walked up and down and it's captured both ways no problems.

I'll leave it like this for a while as it's easier for me to manage alerts etc and see what happens.. convenient, double click on the alert and video plays from that point.

P.S. I can't see lighten conditions in this configuration causing any issues at this stage... I'll keep you guys posted nonetheless...

Thanks again
User avatar
Pogo
Posts: 389
Joined: Tue Jul 18, 2023 7:21 pm
Location: Reportedly in the Area

Re: Seeking Assistance with Eliminating False Positives in Motion Detection

Post by Pogo »

You're already getting way ahead of yourself! LOL

Take a deep breath. Break it all back down to the main elements, the original purpose and necessity of/for having just gone through all of this and proceed accordingly, but methodically. As mentioned, and as simple as it all seems now, there is still a lot to unpack here..., one item at a time.

I would leave the main camera alone for starters -- especially if relying on it for recording other activity beyond the new area of interest -- which it can also be set up to do above and beyond its normal recording when the clone trigger is activated. Or it can just be set up on an 'event basis' now -- including, but not limited to, the actual alert activity itself and can be tuned just for that.

You could then generate multiple clips instead of a continuous recording and eliminate the need for specific main camera alerts which basically just act as index points for scrubbing the time line for events anyway. Eliminate the time line for scrubbing and just use the Clip Viewer instead. Lots more to unpack there, but a very efficient way to view a day's worth of activity quickly and conveniently..., especially with only a few cameras, and is a whole other thread! LOL

But for the main purpose as initially understood to be the objective, and as I've also mentioned previously, the only place the alert images need to be accessed is on your phone when they are generated for that purpose. They essentially have no other use and there is no reason to view them. If something extremely unique is worth saving, it will be in the 'sent' folder of the email account used for sending the alert -- and/or on your main camera as a recording if so configured.

The clone is accessible from the Status page even when hidden. Right click on its entry and you're there. No need to solo it.

A careful review of the thread may yield overlooked information that will make more sense and be useful to you now.

But however you choose to go, you did it!

Again, congratulations!
User avatar
TimG
Posts: 2084
Joined: Tue Jun 18, 2019 10:45 am
Location: Nottinghamshire, UK.

Re: Seeking Assistance with Eliminating False Positives in Motion Detection

Post by TimG »

Good work guys. We are all learning 8-)
Forum Moderator.
Problem ? Ask and we will try to assist, but please check the Help file.
User avatar
Pogo
Posts: 389
Joined: Tue Jul 18, 2023 7:21 pm
Location: Reportedly in the Area

Re: Seeking Assistance with Eliminating False Positives in Motion Detection

Post by Pogo »

The thread has certainly generated plenty of interest. Probably because the title literally applies (or has applied) to anyone who has ever used Blue Iris for anything beyond a platform to just view a few cameras..., which is how many of us started.

Then most of us got curious. "Hmmm..., I wonder how this works? WOW! That's awesome!" And here we are.

Hopefully this has been a productive resource for folks. It has certainly been an interesting exercise for me just reviewing and relating my simple little method as a possible solution for someone beating their head against a wall trying to use AI as a shortcut to performing a basic motion detection task that simply required a better understanding of the basics of BI. After all, basic motion detection is the fundamental function of Blue Iris. Everything else branches out from there one way or another..., simply, or as complex as one chooses to make it. But if one doesn't understand the basics or is willing to learn them, they're much further ahead out of the gate with an integrated NVR package, avoiding the headache and sparing the wall.

I personally like the simple approach. It happens to suit my purposes quite well. But BI in even 'simple' terms requires a learning curve. There are no corners to cut if one really wants to learn how it can work best for them. The fundamentals are everything. They're also the beauty and genius of the platform. Embrace the learning. Don't avoid it.

By no means am I an authority on any of this stuff. I've only been using BI for a couple of years and have still only scratched the surface. But I took the time to learn what works for me. It may not work for others or even apply.

And in the interest and spirit of full disclosure after all my bragging about no nuisance alerts with my particular 'poor man's trip wire' setup discussed here, I must confess that recent weather conditions have proven it can't tell the difference between a blizzard and a white car. Oh, well. I'm not so sure AI could either.
User avatar
Sardon
Posts: 111
Joined: Sun Nov 19, 2023 10:50 am

Re: Seeking Assistance with Eliminating False Positives in Motion Detection

Post by Sardon »

Thank you to the wonderful individuals who have contributed to this discussion. I have been diligently adjusting the motion settings and zone configurations.

Last night, while refining these settings, I encountered some issues, which I think are worth sharing given everyone's involvement in this thread and the amount of interest this topic has generated so far.

Previously, I set up the tripwire zones of two pixels in width which you can see at the top of this page...

The trip wire zones were located towards my front door on the right and the entrance to the camera on the left. However, following Pogo's previous advice, I found this true to be unnecessary and removed it, keeping only one on the left side of the frame.

With the current zone placement toward the left, when someone enters the frame with a pre-trigger record time set to three seconds, it's impossible to identify the cause of the alert as the alert thumbnail itself is blank. Additionally, the shadows entering the zone at night seem to trigger the zone prematurely, possibly because the shadows reach the zone before the person.

To address this, I adjusted the sensitivity settings to Min Obj Size: 150 - Min Contrast: 40, which improved the situation during further tests and left these settings as is.

I spent the evening fine-tuning these settings, walking back and forth to achieve the best configuration. I eventually placed a single vertical line of pixels to
and drew very carefully inside the railing pixel by pixel to avoid capturing any background condensation from my neighbour's flu.

new-zone.jpg
new-zone.jpg (34.32 KiB) Viewed 11033 times
This seemed to work as condensation has been firing up all morning and not a single alert so I think it was being captured before because maybe a single pixel overlaps on the railing itself may have been sufficient for it to flag up as a detection.

I also adjusted the pre-trigger timers back to their normal values of 3 which seemed to work very well. If I had not deleted the old zone I had no choice but to not use a pre-trigger timer, just to be able to capture who was entering the thumbnail on the far left and I just couldn't deal with it with it's abrupt playback, hence I create a new zone only a single pixel this time, on the X axis and a single pixel Y axis all the way down. I did have concerns this wouldn't work but it did.

This morning, an Amazon delivery arrived, and everything worked as expected, boosting my confidence in the system and thinking this is it, mission accomplished.

morning.jpg
morning.jpg (18.3 KiB) Viewed 11033 times
However, when the doorbell rang later at 15:24 and there was no alert from Blue Iris, I was puzzled. Reviewing the footage, I couldn't understand why it failed to capture another Amazon delivery, this time by a person wearing a high-visibility vest.

afternoon-no-alert.jpg
afternoon-no-alert.jpg (47.17 KiB) Viewed 11033 times
I'm considering increasing the sensitivity of the settings, although I am concerned... Living in an area with a high risk of burglaries, I need the system to capture all activity. It's frustrating that it captured one person in the morning but failed to detect someone in a high-visibility vest later.

For now, the only solution seems to be to make the settings more sensitive and try again. I'm concerned about the system's reliability, especially in critical situations like a break-in attempt. It's baffling why it captured one person clearly but missed another, and this inconsistency is worrying.

Regards
SN
User avatar
Sardon
Posts: 111
Joined: Sun Nov 19, 2023 10:50 am

Re: Seeking Assistance with Eliminating False Positives in Motion Detection

Post by Sardon »

Upon further reflection and analysis of the recent inconsistencies in motion detection, particularly the missed detection of a person in a high-visibility vest, I believe I've pinpointed a potential root cause. It appears the current sensitivity settings might be too coarse, particularly when considering the varied lighting conditions between day and night.

This leads me to a crucial question: Is it feasible within Blue Iris to establish distinct profiles for day and night, each with its customised sensitivity settings? My hypothesis is that a less sensitive setting might be more effective during the night to mitigate premature triggers (like shadows), while a more sensitive setting could be beneficial during the day to ensure all activity is captured.

If such a configuration is possible, it could be the key to striking the right balance between avoiding false positives and ensuring no significant activity goes undetected. I'm keen to hear your thoughts on this and whether anyone has experience setting up similar profiles.

Any guidance on how to configure these settings effectively would be greatly appreciated.
-SN
louyo
Posts: 161
Joined: Sat Apr 18, 2020 1:16 am

Re: Seeking Assistance with Eliminating False Positives in Motion Detection

Post by louyo »

I use 4 schedules in BI settings, then select the ones I want to configure in each camera's record and trigger pages. One of the choices for each camera is something like "use Active settings"
Active
Working hours
Night
Weekends
(Active is the short time between night and working hours)
louyo
Posts: 161
Joined: Sat Apr 18, 2020 1:16 am

Re: Seeking Assistance with Eliminating False Positives in Motion Detection

Post by louyo »

From old screen shots, roughly like this.
Attachments
trigger.PNG
trigger.PNG (101.86 KiB) Viewed 8464 times
schedule.PNG
schedule.PNG (20.87 KiB) Viewed 8464 times
User avatar
Pogo
Posts: 389
Joined: Tue Jul 18, 2023 7:21 pm
Location: Reportedly in the Area

Re: Seeking Assistance with Eliminating False Positives in Motion Detection

Post by Pogo »

Ahhh..., refinement. LOL

Such setups are not at all uncommon and in many cases are achievable with the onboard firmware of the camera itself. (You never did say what camera we've been discussing.)

In the case of exclusively using Blue Iris capabilities for such a task, as usual, there are several approaches one can take. The first is to continue learning and tweaking what you already have in place a little at a time and gauge the results as you go.

The existing setup should be able to accomodate the subtleties you are describing. It should have also easily detected the individual you indicated was entirely undetected. That in and of itself suggests additional adjustments are necessary on a fairly basic level before undertaking profiles, scheduling, another clone perhaps..., all of which would be involved to one extent or another as additrional requirements to implement a complete A/B - Day/Night solution for the situation.

Two specific items come to mind for posible improvement in your current trigger settings. The first would be to switch the Motion algorithm from 'Edge Vector' to 'Simple' if currently set to 'Edge Vector'. Some claim this to provide a better result for less complicated detection schemes with 'Edge Vector' offering better results with AI configurations. i've used both, but not extensively evaluated the differences, if any, in my particular configurations. I currently use 'Simple' based on having fairly 'simple' setups and no AI.

The other main area would be continuing to experiment with Object Size and Minimum Contrast along with the Make Time and Break Time settings. Object size can be quite small and still activate a trigger based on the size/shape of the zone. It can also enter and exit the zone quite rapidly. Both of these conditions establish a very fast and precise trigger activation. Combine that with trigger onset sensitivity, reset time, duration of triggered activity, etc., and there is some very capable flexibility there which can present significant oportunities for addition tuning. But it will take time and additional committment.

As mentioned, schedules and profiles are entirely different areas requiring a whole new level of persistance and patience to expore, but also provide another layer of flexibility for managing complicated situations.

Another clone would be the way to go for initial experimentation. I personally run mine on a simple schedule vs. setting up an additional profile for what is essentially a daily on/off timer task. The same could be done in your situation either with your main and clone camera or with a second clone if your main is configured for more general recording, etc. You could start by just cloning the clone, figure out day settings for one and night settings for the other and set timers for both to shut off and fire up at the same time with alerts being modified to reflect each schedule as well.
User avatar
Pogo
Posts: 389
Joined: Tue Jul 18, 2023 7:21 pm
Location: Reportedly in the Area

Re: Seeking Assistance with Eliminating False Positives in Motion Detection

Post by Pogo »

So I just missed that before posting. Here's the super easy on/off approach without needing to set up an additional profile -- providing a simple on/off solution is all that's needed for a given camera (aside from any other global profile considerations). Monday is the only day indicated in the example but it's actually a daily schedule. The black area is simply 'Inactive' Zone 0 and is what's there if nothing else is. Stupid simple.

DrivewaySchedule.jpg
DrivewaySchedule.jpg (106.27 KiB) Viewed 8456 times
Post Reply