5.5.8 - June 13, 2022 - Code Project’s SenseAI Version 1 - See V2 here https://ipcamtalk.com/threads/codeproject-ai-version-2-0.68030/

So I had this problem even with DS, but I guess I may as well ask it here since I'm now running SenseAI.

On all my cameras, whenever there's a lot of objects on the screen at once, the AI tends to just focus on one and log it as an alert. Is this normal? I would love it if it would detect multiple people in the same picture, even if it means creating separate alerts.

What it's doing now is that it's successfully creating an alert for the first person walking past the cam, but for as long as the cam remains in the 'triggered' state, it won't recognize any other objects (ie. a second person, or a car driving past), even if the first person is now out of the frame.

This is particularly frustrating with my 'overview' cam located at the front of the house, which is set up to pick up passing cars/pedestrians going past.
At night, no matter what movement triggers the cam, it always thinks that a light post with reflective tape on it is a 'person', and thus no vehicles or passing cats get tagged in the alert. I've tried unselecting this area in the zone setup, but it makes no difference.

I've even got zones set up so that objects only trigger when moving along the pavement/street from left-right and right-left. This makes no difference. It still only logs the first object it sees during the day, and that damn lightpost at night without fail.

This is really frustrating and I don't know what to do.

Can anybody offer any suggestions?

Your system is working as intended, in that the BI stops/alerts after the AI reports back on the first object it finds that match your criteria e.g person 75% - After this no more AI processing takes place during the same trigger – Not sure if this is desirable but I think if you change your break time setting so that End trigger time is really short (say 1 second) you’d pick up that second person.

1666709319091.png
 
  • Like
Reactions: Corvus85
if you change your break time setting so that End trigger time is really short (say 1 second) you’d pick up that second person.

Thanks, I think that might help with the problem during the day, however it might not help with the problem at night where any trigger at all will make it recognize the streetlight as a 'person' - usually at around 47-49% accuracy.

Any ideas for that?
 
Also, isn't there a setting to automatically extend the trigger time so that there aren't any breaks at all if it keep detecting triggers? This would be ideal since there will be a big, continuous chunk of mainstream video instead of back-to-back alerts punctuated by substream video.

Is that what this setting does?
1666709749220.png
 
Forgot to add, for multiple people (objects) in the same frame – The AI is detecting all of them and should give each one a highlight (box round them) along with a confidence score. However the BI alert will only show the object listed once with the highest confidence score. For example if you have two people in the same frame and the AI detects both of them person 1 with 85% and person 2 with 75% the BI alert will read person:85% but your alert images should show both of them and their confidence score. It’s worth having “Burn label mark-up onto alert images” enabled for this

1666709761670.png
 
  • Like
Reactions: Corvus85
Thanks, I think that might help with the problem during the day, however it might not help with the problem at night where any trigger at all will make it recognize the streetlight as a 'person' - usually at around 47-49% accuracy.

Any ideas for that?

you don't want to be using anything lower than about 75% so that shouldn't be an issue. If you're only getting people as 40% they must be far away or you need to increase the model size you use.
 
f you're only getting people as 40% they must be far away or you need to increase the model size you use.

Do you mean by selecting 'large' in the CodeprojectAI server?
Is this also applicable if I'm using the ipcam-combined and ipcam-dark models exclusively? Or only if I'm using the standard or YOLO models?
 
Also, isn't there a setting to automatically extend the trigger time so that there aren't any breaks at all if it keep detecting triggers? This would be ideal since there will be a big, continuous chunk of mainstream video instead of back-to-back alerts punctuated by substream video.

Is that what this setting does?
View attachment 143765

TBH I only record mainstreams 24/7 - I only use my sub streams for motion triggers and AI to generate alerts and call my attention things. I've always preferred a nice CBR main stream with direct to disk recording 24/7. To answer your question, I'm not sure there is a way to get what you want. Most people are happy for triggers to last 10-20seconds and when they get an alert just to watch the short clip and from that if there are extra people beyond the first thing the AI alerted for.

Object crossing is nothing to-do with what you’re asking, it’s merely a way of adding logic to the motion trigger (rather than just pixels changing in a zone, an object needs to move from one motion zone to another, pixel change in one zone and then another and in a certain order depending on how you configure it.) - You should really read the help from within BI, it's really good and explains most things well.
 
You should really read the help from within BI, it's really good and explains most things well.

Thanks. Yeah, I did, although it didn't really explain the 'not required for re-triggers' option that well.

Also, if I select 'large' in the CodeprojectAI server, will it have an effect on custom models like ipcam-combined and ipcam-dark? Or only the standard/YOLO models?
 
Thanks. Yeah, I did, although it didn't really explain the 'not required for re-triggers' option that well.

Also, if I select 'large' in the CodeprojectAI server, will it have an effect on custom models like ipcam-combined and ipcam-dark? Or only the standard/YOLO models?

No it won't as there is only one model size for Mike's custom models, from my testing I'd say they're meduim type models. are you getting low scores for people? if so are people far away or do you have poor images for AI to work with?
 
Won't solve all your issues but I clone the cam and have one clone only look for people and the other only look for cars. Then I only have alerts sent for people, but the cars are also recorded, or you could have both alert.

Seems to help only finding one object, I used to miss people because it saw the car
 
Although I love a development thread like this and it's interesting, unless you follow the thread from the very start, it canbecome very confusing with info split over dozens of pages. There's sometimes a lot to be said for separate thread which is trimmed down to just give the basics - what it is, how to install, how to use, latest changes (maybe incorporated into updated usage instructions or feature list).

Ask and you shall receive: A Guide to CodeProject.AI Server
 

Thanks. I'm sure the thread will help many. Very easy to get lost in discussions otherwise which are great but detract from the essential parts of "how to".

It’s another app on my server connected to the internet. I believe you in its intention. But no trusty.

Just block it at your firewall. That's what firewalls are there for. Just to get the information out there for all readers, past, present and future:

This is my Samsung TV. but only because I don't have my BI server running curently. The item is irrelevent It could easily be a BI PC.

Routers vary, but in my Asus it's a simple as 1 click. In others you may have to block the IP but it's not hard. At the worst if you can do none of those you could alter the sub net mask of the BI PC Network adapter to a sub net the router can't talk to as per the instructions for isolating cameras with a 2nd NIC card (located in the help section of this forum). Ultimate and the same as altering the network mask, is don't plug it into the network and have it standalone.

However, for most, it can be easily done in router.

All I've done here is in the network map view (LHS), clicked on the item BI Server (- here I'm using a Samsung TV standing in for it as it's not connected) located in the list to the right h/s, and I get the pop up box seen in the centre. From here in Asus it's a simple as clicking the "block internet access" button.

Block Internet.jpg

As for receiving alerts without, beyond my knowledge, but you probably need to set up some kind of sms texting system but then you're giving that access as it will probably need to send sms over the internet unless you happen to have your own SMS station hardware! Can you trust a random downloaded sms messaging app more than BI with internet access? BI is a reputable business.

For most, a VPN is just fine. If my bank uses a VPN, I'm sure I can trust them.

Mine is double protected for the really paranoid as my router is a router not router modem so I'm also stood behind a modem firewall as well. I did it because I needed a modem simply to connect, not for additional firewall, but it's there as a consequence albeit the other firewall isn't the one you're going to block through as the only itme it sees is the router. It does give extra antihack protection though through being layered. If anyone hacks me though they will be disappointed. No financials kept on pc and no state secrets. A few scenic photos and some internet discussion bookmarks. If China ever wants a few pics of the countryside, they only need to ask!
 
Last edited:
I updated two computers over 2 days.

My test computer, CPU, updated like a breeze. No problems whatsoever. It was up and running after the time it took to install CP after updating BI.

My production computer, GPU, totally choked. BI was unstable, freezing up and totally not working. I reinstalled Windows, clean install of BI, clean install of CP.

Added the cameras by exporting each camera from test and then importing them into production. This was a few hours of work. All seems to be working.

Whew!
 
I'm going through this thread page by page, changing my settings, testing, and re-testing, but for the life of me I can't seem to get object detection working with Blue Iris 5.6.2.10 x64 and CodeProject.AI Server 1.6.7.0.

I want to be able to detect a dog, or a person. I am using a Wyze Cam v3, with the firmare for RTSP and the cam stream seems to be running on Blue Iris just fine. My computer is pretty old. I am running an Intel Core i7 Quad-core 3.40 GHz, with an AMD Radeon R9 270X (so I don't need the GPU or CUDA support). I ran the hardware for CodeProject.AI Server and it seems my computer is OK?

I am using Mike Lud's custom model ipcam-general because I hear it's the cats pajamas.

If someone could please share their settings or tell me what's wrong with mine I would greatly appreciate it.

blue-iris-settings.png
blue-iris-camera-general.png

blue-iris-camera-video.png
blue-iris-camera-AI.png
blue-iris-camera-alerts.png
blue-iris-camera-motion.png
blue-iris-camera-trigger.png
blue-iris-status.png
codeprojectai-settings-png.144113


Actually looking at my log it seems like it is working. Is it working? How are you all showing those pictures with detection boxes around objects that indicate what the object is? I tried to get my dog to walk into the frame, but so far he doesn't seem to be detected.

blue-iris-status-log-person.png
 

Attachments

  • blue-iris-status-log-person.png
    blue-iris-status-log-person.png
    29.2 KB · Views: 5
  • 202210.txt
    202210.txt
    5.1 KB · Views: 2
  • codeprojectai-settings.png
    codeprojectai-settings.png
    71.9 KB · Views: 312
I'm going through this thread page by page, changing my settings, testing, and re-testing, but for the life of me I can't seem to get object detection working with Blue Iris 5.6.2.10 x64 and CodeProject.AI Server 1.6.7.0.

I want to be able to detect a dog, or a person. I am using a Wyze Cam v3, with the firmare for RTSP and the cam stream seems to be running on Blue Iris just fine. My computer is pretty old. I am running an Intel Core i7 Quad-core 3.40 GHz, with an AMD Radeon R9 270X (so I don't need the GPU or CUDA support). I ran the hardware for CodeProject.AI Server and it seems my computer is OK?

@Pharoahs2 I see a space between "person, dog". Remove that space to see if that helps.