While holding down the control key double click on an alert. Alternatively you can drag and drop .dat files from your "new" directory in windows explorer onto the analysis pane (lower right).
Blue Iris Version 6.0.1.24 Now Supports Custom Models
- Thread starter MikeLud1
- Start date
I still think this is by design. I've got almost 350 "occupied" canceled alerts today and every one of them has something in them that AI picked up. However, none of them contain anything I have defined in the AI "To confirm" field.
I asked ChatGPT "What causes a canceled alert in blue iris to be occupied?"
Here's the response...
In Blue Iris, “Canceled (occupied)” on an alert basically means:
Here's the response...
In Blue Iris, “Canceled (occupied)” on an alert basically means:
Here are the most common reasons it happens, in plain terms
The main causes
1. Camera is already handling another alert
If motion triggers happen back-to-back (or overlap), Blue Iris may cancel the newer alert because:- The camera is still recording
- The previous alert hasn’t finished its post-trigger time
- The alert list/AI analysis queue is still busy
- High-motion scenes
- Short alert cooldowns
- AI + motion triggers combined
2. AI analysis (DeepStack / CodeProject.AI) is overloaded
If you’re using AI:- The AI server can only process so many images at once
- When it’s backed up, Blue Iris marks new alerts as occupied and cancels them
- CPU/GPU spikes
- Alerts firing but no confirmations
- Lots of “canceled (occupied)” during busy periods
3. Alert cooldown or “cancel duplicates” settings
Some alert settings intentionally suppress alerts:- “Cancel duplicates”
- Alert cooldown timers
- Break time between alerts
4. Continuous recording + alerts competing
If the camera is set to:- Continuous recording and
- Alerts with snapshots, AI, or push notifications
5. System performance limits
When the system is near its limits:- CPU maxed out
- Disk I/O bottleneck
- RAM pressure
How to reduce or eliminate it
Try a mix of these (you usually don’t need all):Alert tuning
- Increase minimum time between alerts
- Disable cancel duplicates only if you truly need every alert
- Increase post-trigger recording time so alerts don’t stack
Motion tuning
- Reduce false motion (trees, shadows, reflections)
- Use zones and minimum object size
- Lower frame rate for motion detection
AI optimization
- Limit how often AI runs (e.g., once every X seconds)
- Reduce snapshot resolution sent to AI
- Make sure CodeProject.AI / DeepStack isn’t CPU-starved
System tweaks
- Move DB to an SSD
- Lower camera FPS or resolution
- Enable hardware acceleration (Quick Sync / NVENC)
When it’s not a problem
If:- You’re still getting alerts for real events
- The canceled ones are duplicates or near-simultaneous motion
so maybe the "max connects" could come into play. On my setup it's not uncommon for 4 to 5 cams firing almost the same time when I car drives around house. I had mine set to 4, but it was at 32. I'll try bumping it up again and test tonight. All of my occupied alerts are cars, and I have cars in confirmed box. The saga continues 
Yes, it's quite the rabbit hole. I've spent the last 30 minutes responding to ChatGPT questions.
The conclusion so far is:
If you want this almost gone
Your best options (ranked):- Tight motion zones + 1.0s object duration
- Alert break time ≥15s
- Day/Night profiles
- P-core affinity
- GPU-based AI (NVIDIA or CodeProject AI)
Well, some good news to report this morning. After changing my "max connects" back to 32 I had ZERO occupied alerts last night. At first, I thought I must not have had many cars go by, when I actually had more than usual. All 5 cams that I tested and have .dat analysis enabled did not have any. I will keep an eye on it and hopefully this was MY issue anyway. BTW I was running 4 max connects before. Running 6.0.2.1
I changed mine from 4 back to 32 and I'm still seeing them. I think the difference between yours and mine is that you have an external GPU and I don't.
Hmmm, I changed my demo machine back to 32 this morning and it uses the UHD onboard 770. So far I have over 175 alerts and NO occupied.
Running 6.0.2.1 and using yolov8s 13 cams running AI.
I run this on about all of them.
Also, this machine is all motion only, no continuous and it really don't care if I leave the demo running as this is my main pc.
Last edited:
So I've played around with this and with the max connects set to 32 on my i7-14700 with 32 gigs of memory the total RAM usage jumped up to the mid 90% range and I was still getting occupied canceled alerts. I lowered the max connects to 12 and that seems to have done the trick. Total RAM usage is back in the 50% range and I haven't seen an occupied canceled alert in over an hour.
Here is a good one, just had one on both systems and almost same time. Both should have been confirmed easily as a person. Also, there wasn't much other AI activity going on then as well.
Now I am thinking down other ideas...camera? network? LOL Keep in mind, the AI did confirm me out there several times in this period as I was taking 3 dogs out through out a 10 min. span.
Now I am thinking down other ideas...camera? network? LOL Keep in mind, the AI did confirm me out there several times in this period as I was taking 3 dogs out through out a 10 min. span.
Last edited:
Following from the sidelines with great interest. On my system (i5-8500, 16GB RAM, integrated UHD graphics 630 only), Bi definitely struggles more with more than 4 max connects in the general AI configuration (I find that 2-4 seems to be about optimal, with no significant difference between the two settings). When I try to run 8 or 12 max connects, the CPU usage spikes to 100% for long periods of time on every object detection (even though i have it set to use the built-in GPU for AI), and I will see even more falsely "occupied"/cancelled alerts (where it clearly finds an object that it properly classifies as an alert type object, within the desired zones, yet BI shows the timer symbol for a static object). Clearly, having a modern, capable Nvidia card is of significant benefit here, but it's disappointing, as I never had this much trouble running CPAI with my current system, and it's hard to justify the Nvidia card given the current runaway pricing for those and additional power consumption. Really hoping that Ken is hearing from enough people complaining about this issue, so he can work on this rather than the somewhat "fringe" features he has been adding recently.
I agree completely. My main BI system is a I7-6700k 16MB and I used to run CPAI on it ok but was limited to about 5-6 cams. That was using CPU, the onboard GPU would just bog down. With what prices are of video cards, I would spend that money to invest into a 12th gen I7 or newer with 32MB ram. I have proven that type of system will run the built in AI of BI. Ken even states in the help, some may benefit running CPAI over the built in. Would I go out and upgrade to a newer system? Only if I had too. It would make it W11 capable (without hacks) and the 12th gen I7-12000 65w is good on power usage. Keep that in mind as there are several I7 or I9 variations out there.
Changing the max connects didn't help much. After letting this run most of the day I still have a significant number of occupied canceled alerts.
I almost wonder if this was happening in ver 5 and I didn't notice it.
I get this no matter what settings or AI provider I use.
???
I get this no matter what settings or AI provider I use.
???
Well, that got me thinking. I went to a couple of my cams that did get the occupied alerts and switched those 2 cams to bi-combinedv5 model and no primary. So far I have no occupied running this:
Will let it run that way tonight and see what happens.
Keep in mind that there seems to be significantly less of these occurring during the dark hours.
Just to update everyone on the occupied alert issue. I have found that by going back to the bi-combinedv5 model and not using the primary model has almost eliminated all of my occupied alerts. The one garage cam that I forgot to switch models did have 4 or 5 and I think its due to having so many objects in there for the AI to detect. I have now switch it as well and will keep testing.
6.0.2.1
My setting for this test:
6.0.2.1
My setting for this test:
Last edited:
I'm testing a different solution with yolov10s that looks promising. So far it's been running for 16 hours with 0 occupied canceled alerts. I'm going to let it run at least 24 hrs before I post anything.