Seeking Assistance with Eliminating False Positives in Motion Detection

General discussion about Blue Iris
User avatar
Sardon
Posts: 114
Joined: Sun Nov 19, 2023 10:50 am

Re: Seeking Assistance with Eliminating False Positives in Motion Detection

Post by Sardon »

I wanted to provide you with an update on the ongoing refinement of the motion detection settings. Firstly I'm sorry I did not mention that the camera in question is a Foscam FI9831W.

The task of fine-tuning motion detection settings has proven to be quite tricky. It has become apparent that without establishing distinct profiles for day and night, the system's detection efficacy is significantly compromised, especially between the transitions of night and day. This issue is clearly demonstrated below.

thumbnails.jpg
thumbnails.jpg (69.53 KiB) Viewed 1120 times
To improve the detection accuracy, I have had to adjust the sensitivity levels for object and contrast detection, setting them to extremely sensitive. Additionally, incremental adjustments to the make time settings were necessary to ensure accurate capture upon re-entry into the frame. A notable challenge has been a shadow cast by a light source outside the frame's view, which impacts the detection area. As illustrated in the provided snippet above, there were multiple instances where the system failed to detect my re-entry. After testing and making adjustments, I have arrived at a configuration that, while not flawless, is acceptable for the time being. Interestingly, the system does not detect exits from the frame due to the adjusted make time, but this is less concerning during the night hours.

My primary aim is to ensure effective detection of any movement within the frame during the late evening and having a person captured in the thumbnail, a period of heightened importance for surveillance.

Regarding the cloning method, I have found it somewhat difficult to adapt to. My use with Blue Iris has me accustomed to managing alerts directly through the alerts tab that instantly plays footage upon double-clicking an alert.

In your last post at 10:34 pm, discussing the schedule setup, it seems it would deactivate alerts at night, limiting them to daytime only. I am somewhat unclear on this, I read in the manual that the inactive profile simply displays a live stream without recording...

As an experiment, I have created two separate profiles, one for day and the other for night, incorporating specific timings into the schedule.

schedule.jpg
schedule.jpg (24.04 KiB) Viewed 1120 times
This approach seems to be functioning as my camera now displays night on the profile tab in the specific camera settings

After several trials, as depicted above, I have identified a balance which may just work

I believe it would be wise to monitor the setup for a few days, making any minor adjustments as necessary as you suggested. Currently during the day after making the detection settings more sensitive it seems to be working great and I have not had any false positives. But those changes clearly affect the camera at night. Oh before I forget , the algorithm is set to Simple.

This is my first time using separate profiles, a need I had not encountered before. Given the fact I couldn't get to grips with cloning the camera in terms of managing alerts etc, and the way Blue Iris has been setup from the offset, just maybe the creation of distinct profiles may just do the trick for this purpose.

I extend my gratitude to you once again for your invaluable advice and to all those who have contributed their knowledge as well.

Thanks again
SN
User avatar
Pogo
Posts: 396
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 »

Sardon wrote: Sun Jan 21, 2024 2:18 am I wanted to provide you with an update on the ongoing refinement of the motion detection settings.
Blue Iris can certainly be a challenge. I hope you aren't losing sight of the tremendous progress you've made in just the past week. Well done.

Firstly I'm sorry I did not mention that the camera in question is a Foscam FI9831W.

The task of fine-tuning motion detection settings has proven to be quite tricky. It has become apparent that without establishing distinct profiles for day and night, the system's detection efficacy is significantly compromised, especially between the transitions of night and day. This issue is clearly demonstrated below.
With all due respect, that particular camera could by itself be introducing unnecessary complications into the mix. My initial impression from your earlier images led me to think it was a fairly hi-res camera. The specifications I just looked up indicate it is quite the opposite and only 1.3MP at 1080x960p with a mere 8 meter effective IR nightime range.

If I might make a suggestion that would be nearly perfect for your situation at only $25 US, take a look at the TP-Link Tapo C110. An incredible camera for the price and is even ONVIF 20.06 conformant. A slightly older ONVIF version, but still legitimately an ONVIF device which makes it completely independent from any proprietary cloud support for its functionality. In other words, no Internet required to operate exclusively on a private LAN/WAN, but still provides p2p connectivity for direct remote access via their very nice little app when/if required. Otherwise, it operates exclusively on your LAN/WAN with no strings attached.

Apologies for the diversion, but you really do need to give some consideration to getting even a marginally better camera for the purpose. This particular camera would more than suffice and bring some pretty cool additional features of its own to the table -- including a limited onboard AI component among many other features -- plus two way audio and a siren. Very discrete, but very capable. And don't let the baby and pet monitor sales pitch fool you. This little rascal rocks. 2K/HD 3MP @ 15fps

https://www.amazon.com/smart-indoor-sec ... 9hdGY&th=1

I bought one on a whim to monitor the flu temperature on my main wood burning furnace and was amazed when I got it set up and configured in the app alone. Drops right into Blue Iris without issue as well. Definitely worth serious a look.

To improve the detection accuracy, I have had to adjust the sensitivity levels for object and contrast detection, setting them to extremely sensitive. Additionally, incremental adjustments to the make time settings were necessary to ensure accurate capture upon re-entry into the frame. A notable challenge has been a shadow cast by a light source outside the frame's view, which impacts the detection area. As illustrated in the provided snippet above, there were multiple instances where the system failed to detect my re-entry. After testing and making adjustments, I have arrived at a configuration that, while not flawless, is acceptable for the time being. Interestingly, the system does not detect exits from the frame due to the adjusted make time, but this is less concerning during the night hours.
The trigger/re-trigger timing relative to the event duration is equally as important to capturing the desired activity as the make time and sensitivity settings are -- not to mention the pre-trigger record considerations -- perhaps for each direction of movement origination, coming and going. This can be adjusted to allow x-amount of activity to continue being recorded between the pre-trigger record value, the actual trigger and the onset of the next rigger or a series of triggers within the time frame selected. It can be adjusted here:

Motion A.jpg
Motion A.jpg (100.3 KiB) Viewed 1106 times

By also adjusting your clips storage parameters you can then compile several short duration 'alert clips' into 5, 10, 15 minute summaries for fast and convenient review just using the clip viewer.

Motion B.jpg
Motion B.jpg (95.93 KiB) Viewed 1106 times
In your last post at 10:34 pm, discussing the schedule setup, it seems it would deactivate alerts at night, limiting them to daytime only. I am somewhat unclear on this, I read in the manual that the inactive profile simply displays a live stream without recording...

As an experiment, I have created two separate profiles, one for day and the other for night, incorporating specific timings into the schedule.

This approach seems to be functioning as my camera now displays night on the profile tab in the specific camera settings.

The approach I suggested would involve two clones (or your main and a clone) with one configured for daytime conditions and the other for nightime conditions and scheduled accordingly by the method described. One simply becomes inactive and the other becomes active. No additional profiles to create or manage. No unnecessary activity during inactivity unless specified otherwise.

The profiles should also work well for you once the bugs get ironed out. They are another important and powerful tool to have in your hip pocket. That said...,

I would highly recommend upgrading the camera as probably the most important consideration for stabilizing the settings and performance inconsistencies you are currently wrestling with. It will make a difference on multiple levels.
Last edited by Pogo on Sun Jan 21, 2024 2:20 pm, edited 1 time in total.
User avatar
Sardon
Posts: 114
Joined: Sun Nov 19, 2023 10:50 am

Re: Seeking Assistance with Eliminating False Positives in Motion Detection

Post by Sardon »

I did indeed contemplate changing it and, in early January 2023, I opted for an Annke 4K PoE camera to replace the Foscam overseeing my balcony. I chose a PoE model due to the lack of a nearby power outlet, necessitating the use of a PoE splitter and some soldering to make the current Foscam work.

However, replacing the camera proved more challenging than anticipated. When I attempted to swap the Foscam for the Annke, I encountered difficulties in achieving the same focal view. Despite its age, I've been using the Foscam for nearly a decade, and it has PTZ capabilities, it allowed me to easily adjust and obtain a clear view of the balcony. After the correct position was established I simply disabled the PTZ feature. This is something I struggled to replicate with the Annke camera as it's limited in terms of position , I think it was a poor choice to be fair however, it was last year.

Additionally, had I been aware of the trip wire zone feature (poor mans) which you have now brought to my attention, I might have been more inclined to accept the positioning limitations of the new camera. However, due to the camera's position constraints, I was not keen on capturing the interior of the room where it was installed. My previous attempts to set up motion detection zones, like drawing over the entire railings like a lunatic, also influenced my decision to retain the Foscam as I was none the wiser and I presumed it may affect detection overall.

In the end, I allocated the Annke to a different location. This, however, was not entirely in vain as it allowed me to experiment with 4k live view and sub-streams for playback so more insight gained at the time. While I would like to upgrade to a camera it has to support PoE, given that the necessary cabling is already in place, lack of power outlet and I also face challenges with wifi due to 'dead zones' in my apartment, possibly due to the construction materials of the walls. The room where the Foscam is installed is one such area.

I had initially hoped to upgrade, but ultimately, the primary purpose of the camera was to simply monitor people approaching my door and I thought BI would take care of the rest, which has proven that it actually can.

Well pointed out Pogo and food for thought nonetheless :)

I've reviewed the BI settings this morning and day time profile has been activated so I think in terms of during the day all should be well for now. I'll continue to monitor the situation, particularly at night, to see if further adjustments are needed.

If it successfully captures anyone approaching my door at night using a night profile, and sends me a thumbnail of the person, that will be a satisfactory outcome.

I'm now just waiting for the overcast days to see how it pans out with changing lighting conditions and will continue to monitor detection activity.

So far so good, thank you once again.
User avatar
Pogo
Posts: 396
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 »

Good morning (afternoon?) and congratulations on the profiles seemingly working out..., and a well-desrved pat on the back as well.

After just reviewing last evening's exchange I realized the wrong image was used to describe my short clips storage concept and was probably just slightly confusing. LOL

This was the intended screenshot. Not that it is particularly relevant to the subject at hand, but is an associated method to use for managing a bunch of shorth length clip compilations as would be created by using a duration setting greater than just 60 seconds. In my example, and given the necessary trigger activity, a single clip or combination of smaller clips up to a maximum of 6 minutes in length will be created and compiled into a single file before being saved on a 15 minute interval schedule. This results in each 'clip' having the potential to be a single clip 5 minutes long or many shorter clips (depending on actual motion activity) that are combined into the compilation to be saved. Could also just be one 20 second clip within entire 15 minute interval if no other activity occurs. No activity at all would obviously not create anything. OK. So much for that!


Porch Record.jpg
Porch Record.jpg (119.71 KiB) Viewed 1073 times

POE to the camera location is huge and opens a whole new range of camera possibilities, many of which are still quite reature rich and reasonably priced as you probably know. The key point here is establishing more pixels for Blue Iris to use in most effectively implementing all of the settings involved in such a relatively compact area/zone.

You sort of hit the nail on the head with the following observation:

It appears the current sensitivity settings might be too coarse, particularly when considering the varied lighting conditions between day and night.
I would suggest the camera capability is what is 'too coarse' at 1080x960 which is 1MP '1080N'. This technically is not even HD as it uses obsolete 'native' vs. current 'progressive' technology to scan and create the image. A 3MP camera is 2048 x 1536 pixels 2K HD. Even a 720p camera is superior simply by the progressive scanning with a 16:9 aspect ratio. Here is a link with a very clear explanation of 1080N. https://www.cctvoutlet.co.uk/blogs/news ... resolution

And if you can solder, you can get power to a cheapie little 3MP camera using the extra pair in your POE cable ethernet run! LOL

Seriously though, there are many economical camera options to vastly improve the image you are expecting Blue Iris to effectively process for you.

I would seriously consider going shopping soon.
Last edited by Pogo on Tue Jan 23, 2024 2:11 am, edited 1 time in total.
User avatar
Sardon
Posts: 114
Joined: Sun Nov 19, 2023 10:50 am

Re: Seeking Assistance with Eliminating False Positives in Motion Detection

Post by Sardon »

Well, I've been grappling with the camera setup today, the sunlight caused such a disruption that I had to disable the motion detection just before midday. It's dark now, and I've just remembered to switch it back on.

Firstly, I want to say how much I appreciate your suggestion to upgrade the camera. It's certainly on my radar, and I'm open to it. However, I've been reflecting on some previous experiences with AI and how it might complement even my existing 960p camera, despite its limitations.

Here's a link to the vast number of alerts I've been dealing with this morning:

https://drive.google.com/file/d/1-SmRUw ... _Gba6/view

It's quite a pickle, as you can see. I had to tweak the sensitivity settings to such a degree as it completely missed capturing a delivery person in a high-visibility vest. Whilst making the settings more sensitive in general, it's clearly making it more prone to the light changes now.

Interestingly, when I experimented with AI in the past, I didn't realise it could be used in conjunction with a designated zone. Despite the camera's somewhat archaic specs, the AI was surprisingly effective, even detecting human presence downstairs and they were very small and i was overtly impressed. However I did not want to capture anything outside the scope of my boundary.

So, I'm considering backing up the current camera, and giving CPAI another try. I cannot see any harm at this stage... It had its quirks before, like mistaking teddy bears for intruders, but it did a decent job of reducing false alerts with light changes.

Your guidance has been invaluable, and I'm not dismissing the idea of a new camera and sticking with this simple approach. I just think exploring all avenues, including AI, might be worthwhile at this stage given the current challenges. I really appreciate your understanding and patience as I navigate this situation, and any further advice you have would be greatly appreciated.

Thanks again for all your help, Pogo. Looking forward to your thoughts on this.

P.S. Those alerts killed me this morning LOL

Best regards,
User avatar
Sardon
Posts: 114
Joined: Sun Nov 19, 2023 10:50 am

Re: Seeking Assistance with Eliminating False Positives in Motion Detection

Post by Sardon »

CodeProject appears to be rife with issues and does not support the integration of custom models, stating "Restart AI to refresh the list" for model use. These problems are likely due to its current Beta phase. Furthermore, this has started to negatively impact the BI interface, creating glitches in basic navigation and tasks such as double-clicking on a camera, right-clicking to access properties, or checking the status of logs, which now appear empty. These issues seem to have arisen after the installation of CodeProject. I intend to continue experimenting with it for a bit longer for testing purposes, but it seems increasingly probable that I will revert to using Trip Wire and remove CPAI.

Well, it was worth a try...

Could anyone recommend a reasonably priced, semi-cheap PoE camera? At this point, it's worth considering this option. However, I have concerns about light changes, which I'm glad I shared the video in this thread yesterday, so you can understand the issue clearly. Will a higher resolution camera actually circumvent this issue for sure? Light changes are still light changes, after all. I would have thought that a higher resolution camera might be more susceptible. I'm not sure.
User avatar
Pogo
Posts: 396
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 »

Sardon wrote: Mon Jan 22, 2024 7:19 pm
Looking forward to your thoughts on this...

Is this compatible?
It is and It would certainly be an upgrade, though the link indicates it is essentially no longer available. There are many other reasonably priced options available with equal or superior features.

As I mentioned early on, your particular situation presents several significant challenges. They are proving to be formidible obstacles to implementing what would otherwise be the simplest and most effective solution possible for detecting activity on your balcony -- and that was before the shortcomings of the camera itself were part of the equation.

Revisting the AI aspect may or may not prove to be an effective workaround for the lighting issue with more fiddling around. You couldn't pay me to jump into the whole CPAI debacle..., especially for such simple tasks.

I'm still not quite understanding the lighting issue as being such a problem unless there is a bar next door with a flashing neon sign at night or a rapidly changing contrast environment during daylight hours. Some elaboration on those conditions or a real time video may be helpful in more affectively assisting with a possible solution. I guess I'm missing what the actual motion aspect could possibly be to require such extreme consideration. I saw nothing insurmountable in your recent video.

And If I may make yet another overly-simplistic suggestion, try reducing the vertical size of the zone to further minimize errant triggers. Place it in the middle of the scene. Use a pre-trigger time of at least how long it would take a person to enter the scene from the left until entering the zone. Observe how long a person occupies the zone as they walk by and set your make time just below that. Set re-triggering for at least 10 seconds with an active trigger duration of at least several minutes. This should provide a very sound foundation for your desired result -- which after all is supposed to only be for triggering an alert, not generating clips -- which it will still do but would involve additional considerations.

Again, this is the clone zone at the end of my driveway approximately 150' from the camera. Nuisance alerts from rain, daytime sun/shadow changes, autumn leaves falling, nightime traffic headlight bleed, bugs, etc., are very rare and nearly non-existent with the exception of extreme weather or complete obstruction of the lens itself. Yet it will trigger an alert from me walking into one pixel of the zone of a 1080p camera. It is immediately sent to my Android phone. It does not include video as that is not the purpose of the alert. If viewing the associated video is desired, I view the associated clip recorded by the main driveway camera. The example below could not be more clear in exhibiting how simple this all should be.

Driveway Detection.jpg
Driveway Detection.jpg (221.57 KiB) Viewed 901 times
Drive Trigger.jpg
Drive Trigger.jpg (237.45 KiB) Viewed 901 times

So back to camera considerations. A better camera for your situation would be the next logical step toward removing one obstacle which I personally consider an unnecessary challenge. Mounting it outdoors to eliminate the reflections presented by the window would be the next logical obstacle to remove, but I understand the reasons for this not being a viable option.

So what are the location and mounting options? The Foscam appears to be a basic table top style unit. The angle of view, however, would suggest it may somehow be mounted or positioned toward the upper right area of the window frame. If that's the most practical location for positioning the camera, the camera mounting options will obviously be an equally important consideration to all others for starters.

Otherwise, higher resolution/more pixels. 2K would be adequate. 5MP would be pushing it. 4K would be overkill and possibly defeat the purpose entirely depending on the lens. A modest MP camera with 3.3 or 4.0mm lens for a tighter/closer view of the actual trigger area, better IR capability, modern onboard features such as its own motion/AI capability, large capacity SD card recording, two-way audio, ONVIF compliant to provide its own triggering to Blue Iris...., and h264 encoding ability regardless of it also possibly touting all the h265 magic hype..., all need to be considered as bare minimum must features. (Color night vision will rely on white light from the camera which will create totally unmanageable reflections.) POE is obvious, but again, if you can solder..., low voltage DC can go a long way should wi-fi become an option, particularly at 5GHz.

Let's go shopping!
User avatar
Pogo
Posts: 396
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 »

And at the risk of suffering the 'Reolink Wrath', this is probably as good a place to start as any. Finding a suitable POE solution may be a challenge, but several of the Wi-Fi options will also offer ethernet connectivity as does the E1. Getting power to a suitable camera should be trivial compared to everything else you're encountering.

https://www.amazon.com/dp/B08RS4C67L/re ... RhaWw&th=1
User avatar
TimG
Posts: 2098
Joined: Tue Jun 18, 2019 10:45 am
Location: Nottinghamshire, UK.

Re: Seeking Assistance with Eliminating False Positives in Motion Detection

Post by TimG »

The Reolink wrath ? All I keep saying is you will never be bored if you buy Reolink :mrgreen:

I bought one, but I think I got away with it :?
Forum Moderator.
Problem ? Ask and we will try to assist, but please check the Help file.
User avatar
Pogo
Posts: 396
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 »

I hear nothing but great things about the doorbell. Problem is, my wife loves her Ring and isn't giving it up anytime soon..., and I'm not arguing, nor would I win that fight. LOL

The E1 also seems to get some love once folks figure it out. It's the hi-res h265 units that still don't get along with Blue Iris without some diligence (and the latest unpublished firmware upgrade attempts of varying levels of effectiveness).
Post Reply