Seeking Assistance with Eliminating False Positives in Motion Detection

General discussion about Blue Iris
User avatar
Sardon
Posts: 114
Joined: Sun Nov 19, 2023 10:50 am

Re: Seeking Assistance with Eliminating False Positives in Motion Detection

Post by Sardon »

Haha, that's highly unlikely!

By the way, I might need to deselect the option you mentioned earlier. When I stepped out to run errands, it didn't register my departure. Perhaps it was because I was locking up, which reduced the number of real-time images sent to the CP server and the fact that the option to begin analysis with motion leading image was still enabled . I'll disable it now and I'll check the .dat file take a look at what took place.

Also, I have another query. Is there any issue with sending a large number of real time images to CP, apart from the AI server processing and causing a CPU spike, once an alert is triggered it'll stop surely?

Right, kids are out and there are little humans being detected everywhere time to start increasing object size until it rules 'em out!
User avatar
TimG
Posts: 2098
Joined: Tue Jun 18, 2019 10:45 am
Location: Nottinghamshire, UK.

Re: Seeking Assistance with Eliminating False Positives in Motion Detection

Post by TimG »

Also, I have another query. Is there any issue with sending a large number of real time images to CP, apart from the AI server processing and causing a CPU spike, once an alert is triggered it'll stop surely?
No issue at all, you are just increasing cpu load etc. so it all depends if your pc can cope with it. You may have 10 "+real-time images", but it will stop on the first identification, and you will see an asterisk next to it in the DAT file. If it found something on the third image, there should only be approx 3 times/ images checked in the DAT file. You may however choose to force it to keep going until it reaches the end of your "+real-time images" count. That is done by adding something you won't see into "To cancel". Unicorn is a good one. Then it will keep going until it reaches the number of "+real-time images", and they will all be seen in the DAT file.

So if you choose 20 "+real-time images" and you have set your "analyze one each" to 500ms, then BI with "Unicorn" would send an image every 500ms, so taking 20 x 0.5 = 10 seconds worth of motion and images, and BI5 will decide which one has the best identification. The downside of that is that BI5 cannot create an Alert until 10+ seconds (in that case) has elapsed. I hope it is making sense :?

BI5 gives you options. For everything 8-)
Forum Moderator.
Problem ? Ask and we will try to assist, but please check the Help file.
User avatar
Sardon
Posts: 114
Joined: Sun Nov 19, 2023 10:50 am

Re: Seeking Assistance with Eliminating False Positives in Motion Detection

Post by Sardon »

Things appear to be progressing rather well. There hasn’t been much activity in terms of alerts, notifications, or sirens.

It’s become surprisingly quiet. I’m accustomed to receiving a constant stream of alerts throughout the day, so it’s quite a change. However, I believe I'm nearing the optimal configuration for my setup.

However, I wanted to discuss something that puzzled me yesterday. After stepping out briefly to dispose of rubbish, I returned to my front door and was perplexed as to why my CPAI failed to detect me. Since adjusting the settings, it has been quite reliable in detecting movements, not so much me leaving my home but anyone entering into the FOV heading toward my door.

I've attached a video link showing the .dat file and a playback using AI analysis:

Video Link: https://drive.google.com/file/d/101oNth ... sp=sharing

In the video, the .dat file shows no detections, yet the slow-motion playback reveals that it initially detected me with a confidence level of 72%, followed by 51%, and then in the 40% range.

I’ve marginally increased my AI confidence level to 55% from 50%. I’m uncertain why the system didn’t register the initial 72% detection as a valid trigger. Could this be related to disabling the 'being analysis with motion leading image' option?

At this point, I’m also considering setting the confidence threshold generally to 40% which I think will then capture any human-like figure, regardless of its nature. This might be the key to detecting everything from standard visitors to more unusual entities including aliens!

Interestingly, my alert frequency has reduced significantly, prompting me to routinely check if the system is still operational, which is quite funny.

I would appreciate your thoughts on the aforementioned missed trigger and my intentions to lower the confidence threshold. It seems that fine-tuning the settings to find the perfect balance is my next step.

But progress has been made for sure!

Kind regards,
SN
User avatar
TimG
Posts: 2098
Joined: Tue Jun 18, 2019 10:45 am
Location: Nottinghamshire, UK.

Re: Seeking Assistance with Eliminating False Positives in Motion Detection

Post by TimG »

My first observation is that triggering is very different for BI5 alone, or BI5 with CPAI:

BI5 alone - You spend all your time de-sensitising it to avoid nuisance triggers. Then it misses what you were looking for in the first place.
BI5 and CPAI - Turn the sensitivity up to 11 ! You no longer care about those nuisance triggers, as they will be checked by the AI and ignored if not what you are looking for.

This includes little gems like "Only trigger when object travels ?? pixels on sub stream". It made perfect sense for BI5 alone, but might cause you to miss something. If using CPAI turn it OFF. There are a lot of these little traps due to the sheer configurability of BI5.

Now I will go back and read your post more thoroughly :idea:
Forum Moderator.
Problem ? Ask and we will try to assist, but please check the Help file.
User avatar
TimG
Posts: 2098
Joined: Tue Jun 18, 2019 10:45 am
Location: Nottinghamshire, UK.

Re: Seeking Assistance with Eliminating False Positives in Motion Detection

Post by TimG »

Note that the video check isn't the same as the DAT check. The video is checking every frame, but the DAT file shows the series of individual discrete fixed images sent at the fixed "analyze one each" time interval that were sent to CPAI. I can see that your DAT file had the ID (marked with asterisk) at the end which is odd. If it was at the beginning, it would stop as soon as it found something. You could then force it to carry on by adding unicorn to cancel.

If you look at my DAT file, you can see:

Screenshot 2024-01-27 115709.png
Screenshot 2024-01-27 115709.png (108.86 KiB) Viewed 889 times
1. I have "boat" in Cancel to keep it checking. Unicorn will do fine.
2. My GTX1650 in conjunction with an Intel i7-8700k sorts CPAI identification out in 30+ms.
3. I send an image "analyze one each" every 250ms (The difference between every time point in the DAT file).
4. You can't see that there are 20 "+real-time images", but detection time is 20 x 0.25 = 5 seconds.
5. Break time is set to 5 seconds.
Forum Moderator.
Problem ? Ask and we will try to assist, but please check the Help file.
User avatar
Sardon
Posts: 114
Joined: Sun Nov 19, 2023 10:50 am

Re: Seeking Assistance with Eliminating False Positives in Motion Detection

Post by Sardon »

TimG wrote: Sat Jan 27, 2024 11:27 am BI5 and CPAI - Turn the sensitivity up to 11 ! You no longer care about those nuisance triggers...
What am I supposed to be adjusting to 11 ?
User avatar
TimG
Posts: 2098
Joined: Tue Jun 18, 2019 10:45 am
Location: Nottinghamshire, UK.

Re: Seeking Assistance with Eliminating False Positives in Motion Detection

Post by TimG »

Oh dear,

My comedy movie reference failed: Spinal Tap.
Forum Moderator.
Problem ? Ask and we will try to assist, but please check the Help file.
User avatar
Sardon
Posts: 114
Joined: Sun Nov 19, 2023 10:50 am

Re: Seeking Assistance with Eliminating False Positives in Motion Detection

Post by Sardon »

How embarrassing for me; I've never come across that term before.

To quickly recap: The BI is set to maximum sensitivity. The only setting I can adjust is contrast, as I've had to tweak the minimum object size. I'll enter 'unicorn' in the cancel field so it analyses all images, and I'll keep an eye out for a GPU. I might soon have a spare 3080, which I could allocate to the BI machine if I upgrade to a 4080 super. It might be a bit excessive - it idles around 111 watts, but boasts an impressive CUDA core count of just over 10,000.

For the time being, I have no alternative but to rely on the CPU, with response times ranging from 120ms to a maximum of 170ms. The performance you're achieving with that GTX 1650 is indeed very impressive.
User avatar
Sardon
Posts: 114
Joined: Sun Nov 19, 2023 10:50 am

Re: Seeking Assistance with Eliminating False Positives in Motion Detection

Post by Sardon »

I believe the missing piece of the puzzle in this situation is a GPU. Based on my online research, all tests and reviews seem to solely refer to the GTX 1650.

I've not found anything else that suggests the use of a more capable GPU.

As Tim suggested, the 1650 appears to be sufficient for running CPAI which clearly he has demonstrated according to his .dat file reduces the latency significantly.

Using my 3080 would be overkill for this purpose.

However, it's crucial that I do reduce the latency; after inputting "unicorn" in the cancel field and reviewing the .dat file, it's evident that the system is now comprehensively scanning the total amount of images I specify in the real-time field, but this results in considerable delays in sending alerts to my phone.

A GPU capable of delivering low latency is essential here. My best option seems to be acquiring a GTX 1650 for this purpose and not using my 3080.

I was trying to find benchmarks on cards such as my 3080 that have been used for this purpose and I've found nothing.

eBay it is....
User avatar
Sardon
Posts: 114
Joined: Sun Nov 19, 2023 10:50 am

Re: Seeking Assistance with Eliminating False Positives in Motion Detection

Post by Sardon »

Quick question I forgot to mention earlier.

To activate alerts, whether push notifications or sound alerts, where is the most suitable location to define your preferences, as I've just found you can set this in two different locations.

At present, I have specified push and sound alerts in the Alerts tab, under 'Actions' on Alert.

However, I have recently seen the option under the Trigger tab, then 'When Triggered', followed by 'Immediate Actions'.

Just wondering if this makes any difference in terms of processing speed or doesn't it matter ?
Post Reply