Smart Motion Detection generating a lot of false positives
- Thread starter tuvow
- Start date
CCTVCam
Known around here
- Sep 25, 2017
- 2,834
- 3,712
If that's true then person detection on the camera is fooled by a flapping piece of black fabric which is pretty gawdawful.
Not really. Without it, you wouldn't get images of people like this (not the best camera - this is from a news report in The Daily Mail Newspaper so all credits to them for photo - Article here: Gun-toting robber pictured working himself into a frenzy :
The whole point of running AI on top of the cameras software is to provide that additional more intelligent filter that can compare an image with millions of photos from different angles of every known object to help remove the false positives. If Dahua were to narrow the incamera software triggers down too much then criminals dressed in balck etc wouldn't trigger alerts and people would be moaning the motion detection was missing triggers. It's always a balance and no doubt camera software tries to include everything that could be a real trigger, whereas the purpose of additional AI, is to cut out the false triggers by more accurately identifying what probably isn't.
mephisto_uk
Getting the hang of it
One thing I'm realising is that the detection is not good for triggering recordings. I've been testing for a few days to trigger when motion is detected and it misses a lot of post/during movement when someone parks a car for a few seconds and then step out after 10-15 seconds for example. It misses a ton of footage, so I guess constant recording with motion alerts may be the best way for me. I was hoping to save on storage by not having as much footage written to disk, but I guess there is no good balance between detecting, recording footage and disk space used.
Not many here will advocate for recording on motion only. Way too many variables.
The balance is to use substreams and record 24/7 substream and then mainstream on triggers.
Or set up a lengthy record time after motion if doing motion only, but then again you could miss something.
The balance is to use substreams and record 24/7 substream and then mainstream on triggers.
Or set up a lengthy record time after motion if doing motion only, but then again you could miss something.
looney2ns
IPCT Contributor
With proper tuning of cameras, IVS rules, and or BI motion detection, it can be pretty darn accurate. But it takes some time to dial everything in properly.
I have two cameras in my back yard that trigger on BI motion only, not recording 24/7. And after proper tuning of the BI motion, it never misses a capture.
But you have to spend some time to get it dialed in.
I use substreams for "thumbnail" views and motion detection for my cameras without IVS to decrease CPU load. I see a lot of people also use substreams for 24/7 recordings to make sure nothing gets missed but is there really a purpose for that with that low resolution? (I have set up "D1"quality for my substreams) So hypotetically, if your IVS/ motion detection/deepstack/whatever misses a trigger with a face ID of a "perp", then your 24/7 low res recordings won't help you out? Or am I missing the point here
Something is better than nothing, but if you run a higher bitrate on the D1 it can look pretty good without a lot of CPU load and is better than nothing...
Now this applies to plates but the concept is the same
ipcamtalk.com
Now this applies to plates but the concept is the same
How much resolution is really needed for LPR
We have folks come here all the time thinking that more MP is better, whether it be for general purposes or for LPR. Those of us that have been around long enough know that sensor size is more important than MP. Those that have been here awhile know that I share a representative sample of...
looney2ns
IPCT Contributor
Thanks @looney2ns , I actually used that guide when I set up substreams and D2D.
I am still learning all the cool "new" features of BI5. I have actually been using BI since 2014 but that has been BI3 (I configured it back then and have basically not played with it, it has just worked.)
Now I set up a new 8th gen i5 with 16gb ram and SSD hd, using substreams and direct2disk, along with Intel hardware acceleration and have set up a couple of new Andy cams with IVS and Onvif triggers.
It's amazing how efficient all these "tweaks" are. When running BI as a windows service and only accessing via ui3, my CPU load is rarely above 10-12% . Just realized this may be slightly OT
I am still learning all the cool "new" features of BI5. I have actually been using BI since 2014 but that has been BI3 (I configured it back then and have basically not played with it, it has just worked.)
Now I set up a new 8th gen i5 with 16gb ram and SSD hd, using substreams and direct2disk, along with Intel hardware acceleration and have set up a couple of new Andy cams with IVS and Onvif triggers.
It's amazing how efficient all these "tweaks" are. When running BI as a windows service and only accessing via ui3, my CPU load is rarely above 10-12% . Just realized this may be slightly OT
looney2ns
IPCT Contributor
mephisto_uk
Getting the hang of it
Did you have IVS for those cameras before using BI for motion detection? How is the additional CPU load looking like?
I'm thinking about for the time being just getting a couple of very large HDDs and leaving 24/7 until I learn more, Maybe that is cheaper than getting hardware to cope with additional load of software detection?
looney2ns
IPCT Contributor
The only reason I'm using BI motion detection on those two cameras is because those cameras do not have AI IVS.