1.25 => 1.26.5: incredible false positives for motion- HELP!

Forum for questions and support relating to the 1.26.x releases only.
Locked
TheKorn
Posts: 142
Joined: Fri Aug 02, 2013 12:39 pm

1.25 => 1.26.5: incredible false positives for motion- HELP!

Post by TheKorn »

So I recently upgraded zoneminder from 1.25 to 1.26.5. Was running great under 1.25, had things dialed in to the point where I'd get a false positive maybe once a month.

Under 1.26.5 though, all hell has broken loose. With the same zone settings (not to mention cameras... :D ) as I was using under 1.25, I'm getting false positives all the time! And for the life of me I can't figure out what's causing it!

In an effort to see like a computer, I took the previous frame and subtracted the frame that zoneminder alarmed on. (photoshop, two layers, difference blending.) Using that, anything that's the same is black, anything that's different is white. The analysis confirms what my eyes are seeing, that there's almost no difference between the non-alarmed and the alarmed images! Worse, what trivial differences there are between the two images do not in any way correspond with the area that zoneminder highlighted in its analyse image!

What's going on!?? Is there some snafu with upgrading from 1.25 to 1.26.5 that completely breaks motion detection?

They say a picture is worth a thousand words, so here are three...
Attachments
Frame 30 (last frame before alarm)
Frame 30 (last frame before alarm)
030-capture.jpg (29.21 KiB) Viewed 2067 times
Frame 31 (first frame of alarm)
Frame 31 (first frame of alarm)
031-capture.jpg (28.98 KiB) Viewed 2067 times
Frame 30 subtracted from frame 31.  (black = no difference, white = extreme difference)
Frame 30 subtracted from frame 31. (black = no difference, white = extreme difference)
031-minus-030.jpg (59.55 KiB) Viewed 2070 times
Last edited by TheKorn on Wed Jan 15, 2014 6:49 pm, edited 1 time in total.
TheKorn
Posts: 142
Joined: Fri Aug 02, 2013 12:39 pm

Re: 1.25 => 1.26.5: incredible false positives for motion- H

Post by TheKorn »

three more, for the complete picture. Note that the blobs that zoneminder drew correspond to almost nothing in the changes picture!

Any ideas of things to check? It seems as though motion detection has gone crazy haywire here. But since we're on 1.26.5 and it hasn't come up before, seems as though this is a migration issue rather than a core "motion detection is broken" kind of issue.

Thanks in advance!
Attachments
ZM's analysis image of frame 31, showing the blob that triggered the alarm.
ZM's analysis image of frame 31, showing the blob that triggered the alarm.
031-analyse.jpg (52.36 KiB) Viewed 2067 times
Statistics generated by ZM
Statistics generated by ZM
29912_stats.png (161.26 KiB) Viewed 2067 times
the zone definition
the zone definition
zone_definition.jpg (145.53 KiB) Viewed 2067 times
bb99
Posts: 943
Joined: Wed Apr 02, 2008 12:04 am

Re: 1.25 => 1.26.5: incredible false positives for motion- H

Post by bb99 »

Try changing the value for: "monitor::buffer::Alarm Frame Count" to a value of 4. I've never had to go that high (usually 2 or 3 stops this) but your example shows 3 consecutive false alarms.
TheKorn
Posts: 142
Joined: Fri Aug 02, 2013 12:39 pm

Re: 1.25 => 1.26.5: incredible false positives for motion- H

Post by TheKorn »

bb99 wrote:Try changing the value for: "monitor::buffer::Alarm Frame Count" to a value of 4. I've never had to go that high (usually 2 or 3 stops this) but your example shows 3 consecutive false alarms.
Root cause turned out to be the upgrade putting in no value at all for "alarm reference image blend %-age" nor for "reference image blend %age" for each monitor. Once I put in values for that for all monitors, back to well defined (for my intents, identical to v1.25) behavior.

Have to wonder out loud why these have no default values, since they aren't maintained from an existing zoneminder install? (And why does the color space default to 8 bits instead of 24, which would be the more common default?)

Small bumps in the road in the grand scheme of things, but seems like all three of these could be better handled on a database conversion. Or at least documented. (To be fair, the color space thing was documented in the forum. The other one I found randomly surfing threads in the forum that had nothing to do with this!)
Locked