[tool] [tutorial] Free AI Person Detection for Blue Iris

seth-feinberg said:
I set them all to 3 secs, but now that you mention that I do remember that break time has to play nice with another time setting, the cool down?

There is also a Pre-trigger buffer of 5secs and i take a jpeg on a cloned camera every 4 secs. Though couldn't remember where cool down was right away (or what it did)....maybe it's in AI Tool?
Click to expand...

My understanding of your problem is that you have too many short clips. My setup is the clone camera creates the jpegs on motion 1 time every 3 seconds. Below are the trigger settings for that camera. If the camera is re-triggered between the first and second jpeg the clone will generate at least 2 more jpegs.

1616551245797.png

I use the clone Master to record the clips. I have the break time set to 20 seconds after the trigger ends. If within the 20 seconds the Clone Master recieves another trigger it will record for another 20 seconds.

1616551554133.png

The "also re-triggers" checkbox is on the alerts tab. Mine is unchecked. The checkbox will allow BI to send more alerts if you have BI configured to send alerts.

AiTool has several Cooldown settings on the "Actions" tab. These have to deal with how often the various actions are taken. I have my cooldown settings at 0 because AiTool sends MQTT topics and payloads to HomeSeer where I control how often BI is triggered and how often texts and push notifications are sent. I want AiTool to notify HomeSeer of every object detected.
 
  • Like
Reactions: seth-feinberg
Okay, so I am having trouble getting this to work. I will try to make this as detailed as possible without making it too long. I am running my deepstack server on the same Windows machine as BI. I am running DS on port 82 and BI on port 81. I start and stop the DS server through the AI tool. It shows as running when I go to localhost:82 so I’m good there.

I have my motion up and running and they are successfully sending jpg images to the correct folder. The only thing is the images aren't even showing up in the AI tool history. I’ve checked and double checked everything. It’s set up to monitor the correct folder, the beginning of the file name is correct. I’m at a total loss.

The two trigger URLs I have setup are the following incase they are relevant are

)

[summary]

I’m sure there is more relevant information you might need to help me but I don’t want to make the post too long. Let me know if you have any questions. Any help is appreciated.
 
BudworthTDog said:
Okay, so I am having trouble getting this to work. I will try to make this as detailed as possible without making it too long. I am running my deepstack server on the same Windows machine as BI. I am running DS on port 82 and BI on port 81. I start and stop the DS server through the AI tool. It shows as running when I go to localhost:82 so I’m good there.

I have my motion up and running and they are successfully sending jpg images to the correct folder. The only thing is the images aren't even showing up in the AI tool history. I’ve checked and double checked everything. It’s set up to monitor the correct folder, the beginning of the file name is correct. I’m at a total loss.

The two trigger URLs I have setup are the following incase they are relevant are

)

[summary]

I’m sure there is more relevant information you might need to help me but I don’t want to make the post too long. Let me know if you have any questions. Any help is appreciated.
Click to expand...

In the log are you seeing "adding new image to queue"? My version of AiTool has a "Edit AI Server" button. You can browse to a jpeg and send it to DeepStack to help eliminate the possible problems.

1616554732882.png
 
BudworthTDog said:
I am not getting that message in my log. I searched "image" in log and no results. My version of AITool (version 2.0.37.7608) does not show the Edit AI Server option either.
Click to expand...

I suggest you upgrade to a newer version. VorlonCD posts the zipped package in the Open Issues section on GitHub. You have to search through the open issues. This is the most recent version that I have found.


You need to verify that DeepStack is processing the images and that AiTool is detecting that a new image is added to the queue. In the camera settings under the masking section there is a custom button. Click scan and it will bring up your current image for that camera. If AiTool does not bring up an image it may be a camera name mismatch or a folder permission issue.
 
jz3082 said:
My understanding of your problem is that you have too many short clips. My setup is the clone camera creates the jpegs on motion 1 time every 3 seconds. Below are the trigger settings for that camera. If the camera is re-triggered between the first and second jpeg the clone will generate at least 2 more jpegs.

View attachment 85315

I use the clone Master to record the clips. I have the break time set to 20 seconds after the trigger ends. If within the 20 seconds the Clone Master recieves another trigger it will record for another 20 seconds.

View attachment 85316

The "also re-triggers" checkbox is on the alerts tab. Mine is unchecked. The checkbox will allow BI to send more alerts if you have BI configured to send alerts.

AiTool has several Cooldown settings on the "Actions" tab. These have to deal with how often the various actions are taken. I have my cooldown settings at 0 because AiTool sends MQTT topics and payloads to HomeSeer where I control how often BI is triggered and how often texts and push notifications are sent. I want AiTool to notify HomeSeer of every object detected.
Click to expand...

Man thank you so much, this is so helpful! I think you probably figured it out but to be clear my AI Tool-triggered clips are not especially short but seem to jump around (as in it's at least continuous. I suppose they are comprised of combined shorter clips but I just wanted to make that distinction in case it changed the answer. I will inact these changes to BreakTime for both cameras (I have 3 secs for both clones currently, sounds like at the very least that's wrong). Can't wait to tie AI Tool into a mostly unset up MQTT/Home Assistant install, just can't seem to find the time. Thanks again, i'll report back:)
 
seth-feinberg said:
Man thank you so much, this is so helpful! I think you probably figured it out but to be clear my AI Tool-triggered clips are not especially short but seem to jump around (as in it's at least continuous. I suppose they are comprised of combined shorter clips but I just wanted to make that distinction in case it changed the answer. I will inact these changes to BreakTime for both cameras (I have 3 secs for both clones currently, sounds like at the very least that's wrong). Can't wait to tie AI Tool into a mostly unset up MQTT/Home Assistant install, just can't seem to find the time. Thanks again, i'll report back:)
Click to expand...
You want to increase the break time on the camera that is recording the clips. This will keep you from getting a bunch of short clips. I use the clones to create the jpegs. When creating jpegs the jpeg each X seconds and the break time determine how often and how many jpegs are created.
 
  • Like
Reactions: seth-feinberg
jz3082 said:
You want to increase the break time on the camera that is recording the clips. This will keep you from getting a bunch of short clips. I use the clones to create the jpegs. When creating jpegs the jpeg each X seconds and the break time determine how often and how many jpegs are created.
Click to expand...
I understand thank you! I was just clarifying that I'm not seeing Short clips. They are are, say, 30 seconds in length but seem to be a combination of a number of 7 second clips...In any event I think i need to enact your suggested times, mine don't seem to make a ton of sense for my understanding fo how this works.

Edit: Enacted 20sec break time on AITool triggered cameras and 6.5sec break time on clone that takes jpegs (and changed Jpeg time to every 3 seconds). Will report back with results, thanks!
 
Last edited:
seth-feinberg said:
I understand thank you! I was just clarifying that I'm not seeing Short clips. They are are, say, 30 seconds in length but seem to be a combination of a number of 7 second clips...In any event I think i need to enact your suggested times, mine don't seem to make a ton of sense for my understanding fo how this works.

Edit: Enacted 20sec break time on AITool triggered cameras and 6.5sec break time on clone that takes jpegs (and changed Jpeg time to every 3 seconds). Will report back with results, thanks!
Click to expand...

I misunderstood your issue. It may be your camera settings i.e. frames per second or I-Frame interval. Trigger the camera by right-clicking and select trigger. View the video and see if it is still choppy. Settings that work for me are 15 fps and an I-Frame every 15 frames. You set those in the camera.
 
Could use some help from the experts, finally got everything setup and working on tweaking the system. I've got the driveway masked off so it'll only trigger for things that actually enter the driveway. The AI detected the car as it was entering the driveway, but once it hit the part that isn't masked, it identified it as a train so it wouldn't record an alert. Is there a way to add train as relevant object or improve identification? Can't believe I waited so long to finally take the dive and get this setup. Such a night and day improvement from my previous alerts
 

Attachments

  • Screenshot (10).png
    Screenshot (10).png
    171.3 KB · Views: 30
  • Screenshot (11).png
    Screenshot (11).png
    167.5 KB · Views: 29
blargrx said:
Could use some help from the experts, finally got everything setup and working on tweaking the system. I've got the driveway masked off so it'll only trigger for things that actually enter the driveway. The AI detected the car as it was entering the driveway, but once it hit the part that isn't masked, it identified it as a train so it wouldn't record an alert. Is there a way to add train as relevant object or improve identification? Can't believe I waited so long to finally take the dive and get this setup. Such a night and day improvement from my previous alerts
Click to expand...

Be sure you are on a fairly recent version. Objects are automatically added to the list as they are detected plus you can add an object.


1616755137541.png

1616755230749.png
 
sunnyhaibin said:
After running VorlonCD AITool for about a week, RAM usage stays super high and does not go down, I will have to close out AITool every so often to release the RAM. It has been doing that on 2 different machines after a short period of usage.

Does anyone have this issue with the latest version of VorlonCD AITool? What can I do to allow it releasing the RAM? Thanks in advance!

View attachment 84865
Click to expand...

Hit the > next to AITool in Task Manager - If running the deepstack windows version, it may be what is taking up that much memory.

Also, if BlueIris is not set to delete its alert files after a few days, the AITOOL window could be keeping many of those images in memory in the HISTORY tab. By default I think the history list stores a max of 5000 history items in memory so it could be a factor if it has been running for a while. I've never personally seen memory usage that high though. The number of cameras you have configured could also be a factor.
 
anhead said:
Hi have problem with ai tools and deepstack when ai tools tries to process an image it give me this error message "ERROR: Processing the image C:\AI-input/altansd.20210321_001024504.jpg failed. Failure in AI Tool processing the image." What can be the problem its not specified in the log why it fails so I need help someone know what to do?
Click to expand...
Hi, I'm just getting everything setup and am running into the same error. Everytime a new image gets saved to my aiinput folder, I get that same error message in AI Tool log.

anhead, how did you end up fixing it?

Anyone else have any suggestions on how to start troubleshooting?

Thanks in advance!!
 
jz3082 said:
Be sure you are on a fairly recent version. Objects are automatically added to the list as they are detected plus you can add an object.

Click to expand...

awesome, just updated. will start to play around with the new version.

Sorry for another dumb question, how do I find the updated version of the AI in the future. I'm still new to github, i clicked on releases from the download section of the readme but the most recent version was just 2.0
 
blargrx said:
awesome, just updated. will start to play around with the new version.

Sorry for another dumb question, how do I find the updated version of the AI in the future. I'm still new to github, i clicked on releases from the download section of the readme but the most recent version was just 2.0
Click to expand...

To use the code you are talking about I believe you have to download it into Visual Studio and the compile it. I follow the open issues section on GitHub. VorlonCD posts the new releases as zip files as an open issue is addressed so you don't have to compile it yourself. There is a member of this board that also compiles and posts the zip in this forum. Below is the link to the open issues section of AiTool.

VorlonCD/bi-aidetection
 
Hi,

I am running 1 docker container with deepstack. It had been working for months. I updated deepstack and updated to a new version of the AI tool and it has a new setting AI Server URL(s) I tried adding the default deepstack options but none of them seem to work. Can someone tell me what I need to do to setup AI Tool to use deepstack again?
 
Hi @warbux I have the same setup with a single docker container for deepstack, the address I use is (deepstack running on port 88 mapped to 5000)
you should be able to verify deepstack is running by opening the address:port in a web browser. I have also made the mistake of cut and pasting the string to start the docker container and not realising i started a scene version not the vision detection
 
chris.h said:
Hi @warbux I have the same setup with a single docker container for deepstack, the address I use is (deepstack running on port 88 mapped to 5000)
you should be able to verify deepstack is running by opening the address:port in a web browser. I have also made the mistake of cut and pasting the string to start the docker container and not realising i started a scene version not the vision detection
Click to expand...

Hi, thanks for the quick reply. I got it working. It was defaulting to the wrong port.

I wonder if these other AI servers are as good or better than deepstack?
 
blargrx said:
awesome, just updated. will start to play around with the new version.

Sorry for another dumb question, how do I find the updated version of the AI in the future. I'm still new to github, i clicked on releases from the download section of the readme but the most recent version was just 2.0
Click to expand...

I have attached the latest version of AITool Version 2.0.1016.7754 if you like to work at the bleeding edge ;) ( I have not personally installed it yet!)

Updated 28/3/2021
 

Attachments

Last edited:
Well, i did download the last version it gives me an error for connecting to blue iris:
I can connect locally normal, also local passcode is turned off...


192.168.1.8 is blue iris

remote registery service is enabled..

Give the current user (DESKTOP-S7AI4BS\Home Security System) access to '192.168.1.8': Attempted to perform an unauthorized operation. [UnauthorizedAccessException] (In BlueIris.RefreshBIInfoAsync:121 > BlueIris<>c__DisplayClass15_0<RefreshBIInfoAsync>b__0:121 > Win32.RegistryKey.OpenRemoteBaseKey(RegistryHive hKey, String machineName, RegistryView view) > Win32.RegistryKey.Win32ErrorStatic(Int32 errorCode, String str))
 
Last edited:
You must log in or register to reply here.
Top Bottom