For those sending the object in the notification URL to BI, they would probably still want the specific animal identified even if they chose the shortcut any mammal in the AOI.
Yet Another Free Extension for Blue Iris Adding AI Object Dectection/Reduction in False Alarms/Enhanced Notification of Activity/On Guard
- Thread starter Ken98045
- Start date
It is time for testing of a new Beta version. To get this test version go to: Ken98045/On-Guard. Download the "Code" zip file, and install it from src\installer.
There are quite few changes here some of them are visible, some not.
Here is a partial list of the changes:
1. If you have multiple AIs and one of them goes down, that AI will be taken out of rotation and the app will continue to function. However, one of the things I noticed when testing is that you must give an explicit IP address (192.168.1.2 for instance) for your local computer. Using "localhost" for the IP Address causes this to fail.
2. AIs requests are dispatched to the next AI available in the pool of free AIs. The previous version would go 1 - 5/repeat if you had multiple. This may be a slight performance gain.
3. Pictures are now displayed in date/time order. The previous version showed them in file name order. While it was often the case that this equaled date/time order, this wasn't always the case, particularly with respect to the test images.
4. There is now a display on the main screen of the last time it took the AI to process a picture. This was a request from someone here. It is shown the form of a color coded bar with the specific time overlaying that bar.
5. Several UI bugs were fixed. This includes the fact that the previous version did not update the search object list for an area if you updated an item. Also, things were fixed related to the MQTT settings.
6. Several other bugs were fixed.
7. Objects that are irrelevant to your areas are filtered out as the first step in object detection. For instance, if you never request "potted plants" as an object they will never show up in the UI. You probably won't notice this, but you should be aware of if. If any area in your camera references an object they will be displayed and outlined everywhere.
8. There was a problem where objects were excluded/included within an area if you have pictures at a different resolution from the area in which the defined. While this isn't often the case in "real life" was definitely a problem with the test images. Those images are all at a different resolutions.
9. The database installation included in setup is now SQL Server LocalDB 2016. There were some complaints that the previous 2019 version wasn't appropriate, and I think this is an easier version to install. So, running Setup you may well see the database install/modify itself.
I think that this version is quite stable, and does fix several bugs I noticed while testing. I would like a least a couple of you to test it before I make it a "release" version. The data from the 1.7.x versions is all compatible with this version (1.8.1).
There are quite few changes here some of them are visible, some not.
Here is a partial list of the changes:
1. If you have multiple AIs and one of them goes down, that AI will be taken out of rotation and the app will continue to function. However, one of the things I noticed when testing is that you must give an explicit IP address (192.168.1.2 for instance) for your local computer. Using "localhost" for the IP Address causes this to fail.
2. AIs requests are dispatched to the next AI available in the pool of free AIs. The previous version would go 1 - 5/repeat if you had multiple. This may be a slight performance gain.
3. Pictures are now displayed in date/time order. The previous version showed them in file name order. While it was often the case that this equaled date/time order, this wasn't always the case, particularly with respect to the test images.
4. There is now a display on the main screen of the last time it took the AI to process a picture. This was a request from someone here. It is shown the form of a color coded bar with the specific time overlaying that bar.
5. Several UI bugs were fixed. This includes the fact that the previous version did not update the search object list for an area if you updated an item. Also, things were fixed related to the MQTT settings.
6. Several other bugs were fixed.
7. Objects that are irrelevant to your areas are filtered out as the first step in object detection. For instance, if you never request "potted plants" as an object they will never show up in the UI. You probably won't notice this, but you should be aware of if. If any area in your camera references an object they will be displayed and outlined everywhere.
8. There was a problem where objects were excluded/included within an area if you have pictures at a different resolution from the area in which the defined. While this isn't often the case in "real life" was definitely a problem with the test images. Those images are all at a different resolutions.
9. The database installation included in setup is now SQL Server LocalDB 2016. There were some complaints that the previous 2019 version wasn't appropriate, and I think this is an easier version to install. So, running Setup you may well see the database install/modify itself.
I think that this version is quite stable, and does fix several bugs I noticed while testing. I would like a least a couple of you to test it before I make it a "release" version. The data from the 1.7.x versions is all compatible with this version (1.8.1).
Vettester
Getting comfortable
- Feb 5, 2017
- 799
- 814
When I tried to run the setup file I got the following error.
I was able to update to 1.8.1 by running the installer.msi.
Last edited:
I have the new version running on a different PC. I split-up my cameras between the 2 versions of OnGuard.
Vettester
Getting comfortable
- Feb 5, 2017
- 799
- 814
Ken,
After installing 1.81. I'm seeing the follow error in my log files every time a trigger occurs;
Everything seems to be working, but the database downgrade appears to have failed. I tried to uninstall and reinstall and that didn't help.
EDIT: I resolved this by uninstalling 1.8.1 and then downgrading to an older version (1.6.1). This reinstalled SQL Server LocalDB 2019. I then uninstalled 1.6.1 and reinstalled 1.8.1 and it changed the database to SQL Server LocalDB 2016.
Last edited:
Three of my cameras can be triggered at the same time by the same event. The cameras send an image every 2 seconds. I have never been confident that I am using the correct trigger interval because with 3 cameras firing there may be little time between images. I have it set to 2 seconds. How is the trigger interval used?
I also had another Windows Server 2019 VM on my Hyper-V host so I decided to try it. Well it worked!!! So it turns out that I have some DLL issues on my main Windows Server host that I'll have to sort out. Thanks all.
I have more information about the problem I am seeing with 1.7.3.3 were objects are not being identified and no alerts are sent. Below is the portion of the log.
The top portion is from an image processed that did not find the person as an interesting object and for the second test I changed the name and ran it through again. There were 2 clips in a row were the person was not identified but in both cases there were identified on the second pass. The log is identical until it gets to "Adding interesting object before 2nd chance ignore: person". Is this were the code is getting hung up? Is this related to AOI where objects are ignored? Below are my AOI for this camera. The bush to the right and the dish at the bottom are AOI that I ignore objects. The red X is were the person was identified with an 81% confidence.
Do you see anything that needs modified in the AOI settings?
I had a question above about the snapshot interval. I changed it to 1 to see if that would make a difference. Below are those settings.
Once a camera stops logging objects as interesting it does not get corrected until an OnGuard restart or a AIO setting change.
I am running 1.8.1 for a single camera on a laptop over wireless so I don't have enough data to see if this problem exists in the beta.
The top portion is from an image processed that did not find the person as an interesting object and for the second test I changed the name and ran it through again. There were 2 clips in a row were the person was not identified but in both cases there were identified on the second pass. The log is identical until it gets to "Adding interesting object before 2nd chance ignore: person". Is this were the code is getting hung up? Is this related to AOI where objects are ignored? Below are my AOI for this camera. The bush to the right and the dish at the bottom are AOI that I ignore objects. The red X is were the person was identified with an 81% confidence.
Do you see anything that needs modified in the AOI settings?
I had a question above about the snapshot interval. I changed it to 1 to see if that would make a difference. Below are those settings.
Once a camera stops logging objects as interesting it does not get corrected until an OnGuard restart or a AIO setting change.
I am running 1.8.1 for a single camera on a laptop over wireless so I don't have enough data to see if this problem exists in the beta.
Ken, when viewing the image of the first log the bottom portion is not visible so you need to download it to see the object was identified as interesting and the MQTT message was sent.
DeepStack February 2021 Release - Fixes for Jetson, Windows Native and Docker Versions
Updated versions of DeepStack are available. I just installed the Docker CPU version. The processing time is up slightly but the accuracy has improved.
With the prior release from December I was only using 1 instance of DeepStack in Docker because the processing time of the other instances was almost double that of the first instance. With this release I am running 5 instances and the processing times are the same for all. With 5 instance the CPU usage of VMMEM increased from 3% to 8% and the RAM usage increased from 3.7GB to 4.3GB.
Updated versions of DeepStack are available. I just installed the Docker CPU version. The processing time is up slightly but the accuracy has improved.
With the prior release from December I was only using 1 instance of DeepStack in Docker because the processing time of the other instances was almost double that of the first instance. With this release I am running 5 instances and the processing times are the same for all. With 5 instance the CPU usage of VMMEM increased from 3% to 8% and the RAM usage increased from 3.7GB to 4.3GB.
Last edited:
Anyone have good settings on the Blue Iris side for this type of situation? My front door camera faces directly down the sidewalk leading up to my house. I easily get snapshots taken of people walking back and forth, left to right or right to left in front of my house, but when someone walks straight down the sidewalk to my house, or straight away, blue iris doesn't seem to detect this as motion. I've already made the settings almost as sensitive as possible, but these two cases don't seem to trip the trigger. So anyone approaching or leaving doesn't even have a chance to get analyzed by deepstack since no screenshots are taken
If your object size and contrast are low check the make time and object must travel X pixels. Try lowering 1 or both of those. Do you have more than 1 zone setup? If you do check if you use zone crossing. I know some of the tutorials say not to use zones but I don't want the camera to trigger from motion in some areas so I adjust the mask for the zone to ignore those areas.
left <> right motion is much easier to detect. Straight incoming/outgoing is more challenging. Especially at night.
Does your camera have the ability to utilize IVS tripwires/intrusions instead of motion detection?
Installed v1.8.2 tonight. Actually found a reason to start using MQTT 
Works fantastic. No errors after a couple walk tests. I did notice either OnGuard or Deepstack have a better time find something of interest when in B&W mode instead of color. Will mention if something bad pops up.
Works fantastic. No errors after a couple walk tests. I did notice either OnGuard or Deepstack have a better time find something of interest when in B&W mode instead of color. Will mention if something bad pops up.Sorry, I've been offline for a few days. I did want to let everyone using 1.8.x that I have posted 1.8.2.2. This is not a release yet (probably late today).
This change increases the buffer space (memory allocated) for the Windows functionality that monitors file directory changes (new pictures). I noticed that occasionally that functionality would completely stop working (no new pictures recognized) after either a lot pictures are processed or there is a lot at once. I can't duplicate the exact circumstances since it is semi-random. In any case, I increased this buffer area from the default of 8K to 2MB. I am hopeful that that will solve this problem, but it may be a few days before I can be completely (well, relatively) sure that this solves the problem.
I will try to get back to people on previous messages later today or tomorrow. This assumes that we won't lose internet service since we are supposed to get heavy snow accompanied by high wind. It hasn't started yet, but sometimes is starts quickly.
This change increases the buffer space (memory allocated) for the Windows functionality that monitors file directory changes (new pictures). I noticed that occasionally that functionality would completely stop working (no new pictures recognized) after either a lot pictures are processed or there is a lot at once. I can't duplicate the exact circumstances since it is semi-random. In any case, I increased this buffer area from the default of 8K to 2MB. I am hopeful that that will solve this problem, but it may be a few days before I can be completely (well, relatively) sure that this solves the problem.
I will try to get back to people on previous messages later today or tomorrow. This assumes that we won't lose internet service since we are supposed to get heavy snow accompanied by high wind. It hasn't started yet, but sometimes is starts quickly.
Sorry I'm late getting back to you. I did try the latest (1.8.2.2-- which is not yet a release) appears to work with both. Right now I'm trying to catch up with messages and get 1.8.2.2 released plus working on 2.0.
Yes. I looked at it briefly and it doesn't seem too bad, but you are the first to ask so it isn't a super high priority. Right now I have some higher priority items to work on. Right now that means making a release of 1.8.2.2, and making sure it is stable.any chance for telegram integration ?
The next release (2.0.0) is also on track. In theory it should run on Linux and Android once I get the settings information saved in an xml document rather than the Windows registry.