Yet Another Free Extension for Blue Iris Adding AI Object Dectection/Reduction in False Alarms/Enhanced Notification of Activity/On Guard

[Ken98045]

Especially since Sentry has seemed to drop the ball...when was the last time someone posted here that it works right or they get a return email from them?

I tried signing up for Sentry a year or more ago and couldn't even get through the sign-up -- I'll give you money if you want it --- part. I emailed them and got a response about 3 weeks later that said "it works for us". I tried it again, and it didn't work still so I gave up on them.

 

[wittaj]

Yeah, I tried it once as well and just didn't work effectively. I was getting better results with IVS and drawing boxes. And my cams that have AI are even better!

But I still have many without AI, which is why I plan to try your solution when I get a block of time to give it a go!

 

[Holbs]

things have slowed so back to doing Blue Iris related stuff. Still on OnGuard 1.5.1 so will upgrade to latest & greatest

 

[Sireone]

Man, I would love to join in on the On Guard fun, but I'm running Windows Server and gets no love.... LOL

Sent from my Pixel 3 using Tapatalk

 

[Holbs]

Man, I would love to join in on the On Guard fun, but I'm running Windows Server and gets no love.... LOL

Sent from my Pixel 3 using Tapatalk

I'm no windows server dude...but couldn't you just to a VMware of windows10 instance just for OnGuard?

 

[Sireone]

I'm no windows server dude...but couldn't you just to a VMware of windows10 instance just for OnGuard?

Hmm... probably, but I thought it played nicer on the same BI machine? AI Tools detects BI and Deepstack and makes setup a breeze.

Sent from my Pixel 3 using Tapatalk

 

[Holbs]

Hmm... probably, but I thought it played nicer on the same BI machine? AI Tools detects BI and Deepstack and makes setup a breeze.

Sent from my Pixel 3 using Tapatalk

I would imagine you just simply change from localhost pointing to deepstack, and put in the IP address of deepstack (wherever it is). This is something I may go down the road to remove some of the CPU spiking that occurs (such as: in a snowstorm blizzard where motion or IVS triggers often).

 

[Holbs]

now running 1.7.3.0
nice & easy install
all areas of interest migrated over without any hitch.
first thing I notice, I no longer have to use the scroll bar on the image as it now fits! Dunno what changed with OnGuard but it was a minor irritating issue

 

[jz3082]

This is probably all related to the parked car problem (I think). You can turn on "Log Detailed Information" and maybe you can get a better idea of what is going on. There is a lot of spew when this is on, and not everything captured in the log will necessarily make sense to you, but if you have the patience to work through it it may help.

Since I am trying to actively work on the parked vs. moving car problem I don't have the time to work through it for you. Once I feel that the parked car stuff is under control I'll work with you on it if it is still happening.

I posted the detailed log information on post 359. Those logs were from 3 different cameras that do not have any visibility to parked cars. The moving vehicle was visible by 3 different cameras. The log showed the vehicle was removed because a parked car was found. From the log I have determined that every moving vehicle is being canceled out by a parked vehicle so no alert is sent. The only time it currently alerts for a vehicle correctly is if you close OnGuard and reopen it. It alerts on the first vehicle correctly and then every other after that is canceled out by a parked car. All other objects are being identified correctly.

What you are working on should correct my issue.

 

[jz3082]

Hmm... probably, but I thought it played nicer on the same BI machine? AI Tools detects BI and Deepstack and makes setup a breeze.

Sent from my Pixel 3 using Tapatalk

My production environment is a Windows 10 Pro PC running BI, OnGuard and DeepStack. I moved DeepStack around to several environments including a QNAP NAS. The setting change in OnGuard is simply entering the IP of the DeepStack box and the port number you have assigned to DeepStack when you start it. I have a laptop that I test new versions of OnGuard on. You create a mapped drive to the BI drive where the images are stored and navigate to that when setting up the cameras in OnGuard. I am using the Beta CPU version of DeepStack. I installed WSL2 on the Windows box for the Docker Desktop backend. I found the best performance running on DeepStack in Docker versus Windows. I get the same processing speed in Docker in High Mode that I got in Windows with DeepStack in Medium Mode (250ms). BI, DeepStack, OnGuard and HomeSeer are on the same box which is an I-7 6700 with 32G of ram. I have 3-4MP cams and 1-2MP with 24/7 recording. The CPU utilization is normally between 6% and 8%. My cameras are set to trigger every 2 seconds for 6 seconds so in a snow storm the CPU usage gets in the 30% range. Memory usage is 30%

As far as the OnGuard to BI setup, OnGuard defaults to the default BI port of 80 and populates the default notification urls for triggering BI. You need to know your short camera names in BI so configuration is pretty easy.

 

[Ken98045]

I have released another version that fixes the problem where cars in motion were determined to be parked and were therefore ignored. That release is here: Release On Guard Security Assistant Version 1.7.3.2 · Ken98045/On-Guard.

In addition, ignoring cars that are in fact parked should be somewhat better. There is still work to be done there.

If you still have a problem where moving cars are ignored please let me know. There may be some situations where this is the case, but my tests show that it happens very rarely. However, your camera angle and the location of vehicles you care about relative to the camera may differ substantially from mine. In the next release (1.7.4 probably depending on the extent of the changes) I will give you the option to turn off the check for parked vehicles and/or give you the ability to fine tune the parked vehicle parameters. Please remember that if you have an area where parked vehicles are commonly found you can define that area as an "ignore" area for vehicles.

 

[OldBobcat]

I have released another version that fixes the problem where cars in motion were determined to be parked and were therefore ignored. That release is here: Release On Guard Security Assistant Version 1.7.3.2 · Ken98045/On-Guard.

In addition, ignoring cars that are in fact parked should be somewhat better. There is still work to be done there.

If you still have a problem where moving cars are ignored please let me know. There may be some situations where this is the case, but my tests show that it happens very rarely. However, your camera angle and the location of vehicles you care about relative to the camera may differ substantially from mine. In the next release (1.7.4 probably depending on the extent of the changes) I will give you the option to turn off the check for parked vehicles and/or give you the ability to fine tune the parked vehicle parameters. Please remember that if you have an area where parked vehicles are commonly found you can define that area as an "ignore" area for vehicles.

Thanks for this update! I look forward to trying it out later tonight.

 

[Ken98045]

I posted the detailed log information on post 359. Those logs were from 3 different cameras that do not have any visibility to parked cars. The moving vehicle was visible by 3 different cameras. The log showed the vehicle was removed because a parked car was found. From the log I have determined that every moving vehicle is being canceled out by a parked vehicle so no alert is sent. The only time it currently alerts for a vehicle correctly is if you close OnGuard and reopen it. It alerts on the first vehicle correctly and then every other after that is canceled out by a parked car. All other objects are being identified correctly.

What you are working on should correct my issue.

The 1.7.3 version had several problems in the area of ignoring moving cars and not identifying parked cars. Please try 1.7.3.2 (see the other post about this). I'm sorry I didn't spot this earlier. My test cases worked for the 1.7.3 version, but that was apparently a fluke. The determination that a vehicles is parked was intended to cut down on vehicle related false alarms. However, it also cut out moving vehicles in many situations due to a bug. I think that 1.7.3.2 should be much, much better. There is more work I'm doing on this, but I don't think I'll ever get to the case where parked vehicles are filtered out 100% of the time with the AI in its current state. One case that I now of where it will not filter correctly now is when multiple people are waking in front of a car at both the left and right corners. A lot depends on camera angles with respect to the vehicle.

 

[jz3082]

The 1.7.3 version had several problems in the area of ignoring moving cars and not identifying parked cars. Please try 1.7.3.2 (see the other post about this). I'm sorry I didn't spot this earlier. My test cases worked for the 1.7.3 version, but that was apparently a fluke. The determination that a vehicles is parked was intended to cut down on vehicle related false alarms. However, it also cut out moving vehicles in many situations due to a bug. I think that 1.7.3.2 should be much, much better. There is more work I'm doing on this, but I don't think I'll ever get to the case where parked vehicles are filtered out 100% of the time with the AI in its current state. One case that I now of where it will not filter correctly now is when multiple people are waking in front of a car at both the left and right corners. A lot depends on camera angles with respect to the vehicle.

I will install it in my test environment and let you know if I find any issues. I will try to install the updates updates as quickly as I can.

 

[Vettester]

There is still work to be done there.

1.7.3.2 is still missing some motion events.

Edit: I’ve tested this version with people and animals and it works well so the only issue is with moving vehicles.

 

[Ken98045]

1.7.3.2 is still missing some motion events.

View attachment 80970

If you can turn on the Log Detailed Information, repeat a test and search for the file(s) in question. You should see a lot of detail there. If you like you can copy the section relative to one of the files. Either attach that to a post or send it to me. I don't need the whole thing, just the area starting with that file being read and ending after it goes through the end of the FrameAnalyzer. At this point I don't know if there is a logic problem or if some parameters need to be tightened up.

Also, please be aware of what the original intent of the "parked" movement exclusion was. It was not necessarily intended to apply to busy streets. Why you'd want to start recording video every time a car goes down a street is a little beyond me, but maybe you do. It is possible that in the future I may apply the parking exclusion only to specified areas (i.e. your driveway, or maybe just cars on the side of your street). Let me know if you think that would be valuable. That would be relatively easy, but would require some additional setup information. I hesitate to add anything that would complicate initial setup, but.....

Thanks for your help!

 

[Vettester]

Thanks for your help!

PM sent...

 

[Vettester]

Why you'd want to start recording video every time a car goes down a street is a little beyond me, but maybe you do. It is possible that in the future I may apply the parking exclusion only to specified areas (i.e. your driveway, or maybe just cars on the side of your street). Let me know if you think that would be valuable.

I have multiple cameras facing the street, but only 1 PTZ that I use to record vehicles. Ultimately I would only like to use this to record vehicles at night because I've had cars pull up at night and then someone will jump out and check the vehicles in my driveway to see if they are unlocked. It has been helpful to know what kind of vehicle they are driving. Right now I am testing this 24/7 until I can get it to work right. I only have one AOI setup for this camera that covers everything this camera sees.

 

[jz3082]

I am have an issue with 1.7.3.2 that I saw in 1.7.2. It was corrected in 1.7.3. On motion the initial MQTT alert is sent successfully. Subsequent alerts fail with this error message.



OnGuard then bogs down the MQTT broker with traffic until it pegs the CPU of the broker.

 

[Ken98045]

I am have an issue with 1.7.3.2 that I saw in 1.7.2. It was corrected in 1.7.3. On motion the initial MQTT alert is sent successfully. Subsequent alerts fail with this error message.

View attachment 80982

OnGuard then bogs down the MQTT broker with traffic until it pegs the CPU

I have multiple cameras facing the street, but only 1 PTZ that I use to record vehicles. Ultimately I would only like to use this to record vehicles at night because I've had cars pull up at night and then someone will jump out and check the vehicles in my driveway to see if they are unlocked. It has been helpful to know what kind of vehicle they are driving. Right now I am testing this 24/7 until I can get it to work right. I only have one AOI setup for this camera that covers everything this camera sees.

OK, my suggestion is that you setup an area for people. Setup BI with a pre-trigger buffer of at least 10 - 20 seconds. In your area for people put in an estimate width/height for the people that would include anyone close enough to your property. You could also setup your phone MMS set for only your nighttime hours. So, in theory you should get a video including people driving up when only when someone gets close enough to the camera. Also, your phone alert may wake you in time to call the police in time for them to get someone in the general area. Not a lot of fun in case of false alarms, but it should be very reliable unless you have a lot of nighttime dog walkers. MMS isn't always fast, but at night it should be fast enough. Also, use a smart light switch with MQTT or IFTTT to turn lights on if you aren't already. IMO DeepStack "people" recognition even with IR cameras is good for the purpose. I do have a (relatively) cheap camera with very, very good IR range if you need a recommendation.

I can definitely put in something that will turn off the parked vehicles check. I just wouldn't want you to get notifications of a car in an area every time someone walked down the street. This does seem like a case where you would want parked exclusion if it were 100% accurate.