Reference Image Blend Percentage aka Vapor Trails
From ZoneMinder Wiki
There is one important adjustment that is available for Fine Tuning the Zone sensitivity and selectivity, and it will not be found in the Config pages for a zone.
In the Console screen, click on the ID number for any camera. This is the far left column.
Select the Misc Tab, and one of the options will be: "Reference Image Blend %ge". This parameter determines how much of any previous alarm pixels will be averaged into your current reference image. If you have Outlining enabled, and you are seeing the outlines of your alarmed areas or blobs in your images, then you have probably seen the vapor trails, or ghosts that follow any moving object across the screen. This ghosting can significantly increase the area of the alarm pixels, or blobs when the object moves quickly across the camera, as an automobile would.
As an object moves, it will change some pixels on the leading edge, and on the trailing edge, some pixels are changing back to the background, and these get interpreted as changed pixels as well. So there will always be a "ghost" that follows a moving image. The Ref Image Blend parameter exaggerates this effect, and can make your alarmed areas and/or blobs uncontrollably large.
One downside is that this parameter is set on a per camera basis, and not on individual zones.
The official docs say:
Reference Image Blend %ge – Each analysed image in ZoneMinder is a composite of previous images and is formed by applying the current image as a certain percentage of the previous reference image.
Thus, if we entered the value of 10 here, each image’s part in the reference image will diminish by a factor of 0.9 each time round.
So a typical reference image will be 10% the previous image, 9% the one before that and then 8.1%, 7.2%, 6.5% and so on of the rest of the way.
An image will effectively vanish around 25 images later than when it was added.
This blend value is what is specified here and if higher will make slower progressing events less detectable as the reference image would change more quickly. Similarly events will be deemed to be over much sooner as the reference image adapts to the new images more quickly.
In signal processing terms the higher this value the steeper the event attack and decay of the signal. It depends on your particular requirements what the appropriate value would be for you but start with 10 here and adjust it (usually down) later if necessary.
However, I have found that values greater than 10 work quite well. Indeed, I have used a value of 50, which shouldn't work, but is giving me acceptable results.