Major motion sensitivity changes after upgrade to 1.27.

[arc]

First, note that my problem resulted after upgrading from 1.25 to 1.27, so I do not know whether this is related to the 1.27 release or the 1.26 realease.

I had all my cameras, zones and motion detection settings set up and working very reliably for over a year. After upgrading to 1.27, I initially got a weird combination of over-sensitivity and under-sensitivity. Major events were not being detected, but low-light very minor camera shimmer in the mornings would alarm like crazy (I do not know that the pixel shimmer was causing it, but it was the only thing I could think of for what woudl trigger the alarm since there were no motion events).

I wound up having to tweak a bunch of things and got something a bit better, but still have this weird result where a 15 second event is now lasting 4 minutes (no events occur over the last 225 seconds. I've checked all the settings on pre/post event image counts and such and nothing looks to be mis-set.

Has anyone else had similar issues or know of settings that might have changed or whose semantics might have changed that would cause previous ZM settings to be interpreted very differently in 1.27?

[TheKorn]

I ran into this issue when I upgraded from 1.25 to 1.26.5.

In your monitor definition, it's likely that the reference image blend percentages aren't set or are set to something goofy, since apparently those fields didn't exist in 1.25. I set mine to 6.25% for both reference & alarm, then everything behaved/detected as it did in 1.25. (I haven't even bothered to tweak them!)

[arc]

Good to hear I am not alone in having an issue. Those "Reference Image Blended" settings were high on the list of suspects, but the docs were not exactly clear on what impact these really had. I changed one of them, but not the "Alarm" one, so that might solve the issue.

Seems like the database migration scripts should be puting something more reasonable in the tables for new default values. I noticed other new settings that had the absence of a value also.

Thank you for confirming and pointing me in the right direction.

[mastertheknife]

There is a some difference in senstivitiy between ZM 1.25 and ZM 1.26/1.27 because the algorithms for motion detection were rewritten (optimized) in ZM 1.26. You can read more about it here: http://www.zoneminder.com/forums/viewto ... =9&t=17652

Another difference is most likely because the default blend percent changed and there is now a new setting for using different blend percent during alarm. To restore pre ZM <1.26.5 behaviour, set the two settings identical.

[TheKorn]

There is a some difference in senstivitiy between ZM 1.25 and ZM 1.26/1.27 because the algorithms for motion detection were rewritten (optimized) in ZM 1.26. You can read more about it here: http://www.zoneminder.com/forums/viewto ... =9&t=17652

Have to say (besides the obligatory fantastic work, of course!) that I didn't notice any change whatsoever between the motion detection routines between 1.25 and 1.26. I know that was more or less your intent in the re-write (re-write with close to identical output). They say the proof is in the pudding, and that's a heck of a lot of proof in my book!

BTW, ever thought about profiling zms/nph-zms? Never made any sense to me why playing back a stream takes twice as much CPU as capturing it!

[arc]

The "Alarm Reference Image Blend %ge" seemed to be the culprit of my problems, so thanks for the suggestion.

Related to the other comments, the other changes from 1.25 to 1.27 seem to be excellent improvements, outweighing the bit of a hassle I had to deal with.