Subscriber Discussion

How Does Motion Detection Work?

JH
Jay Hobdy
Oct 05, 2017
IPVMU Certified

Here is the scenario: license plate reader camera mounted on a thin pole that seems to move with the wind. I think the pole is actually rigid conduit. Not our doing. (we tend to use something a tad stronger than conduit for a pole...) The camera is connected to Milestone license plate recognition software. When the camera moves, the software tries to read the ground, and we get hundreds of false reads.

 

I assume motion is based on pixels changing. Since the entire camera is moving, every pixel is changing. My first thought was to adjust sensitivity but after thinking about it, I wonder if sensitivity adjustments would do any good?

 

Is it based on how many pixels change, or how much they change? 

Avatar
Ethan Ace
Oct 05, 2017

Is it based on how many pixels change, or how much they change?

Yes.

It's both. Off the top of my head, sensitivity refers to how much a pixel has to change in order for it to be considered "changed." Threshold is how many pixels must change in order to trigger.

I would think sensitivity would make a difference here. You should be able to immediately tell the difference, as lowering sensitivity should show fewer pixels changing in the preview when the pole is wobbling (highlighted in green on the image).

U
Undisclosed #2
Oct 05, 2017
IPVMU Certified

It's both. Off the top of my head, sensitivity refers to how much a pixel has to change in order for it to be considered "changed." Threshold is how many pixels must change in order to trigger.

IMHO, there is a temporal aspect to triggering as well, i.e. how quickly does the change occur.

If there wasn't then one would expect many false alarms from dawn to dusk, aside from any due to the day/night switchover.

U
Undisclosed #2
Oct 05, 2017
IPVMU Certified

Eh, maybe you're right, and there is no time aspect.  

Avatar
Josh Hendricks
Oct 06, 2017
Milestone Systems

This is a good explanation for how the software-based motion detection works. In addition, that checkbox for manual sensitivity when unchecked means we are doing some math in the background and trying to filter out noise. This should make for more consistent motion detection sensitivity between day and night, but I doubt it would do much for when the camera itself is waving in the wind.

One thing that doesn't make sense to me though is why motion detection should have anything to do with your license plate recognition. There is no connection between motion detection and when we start attempting to read/detect plates. If the camera moves, it may be that it results in some artifacts which the LPR engine attempts to read, but if it's trying to read the pavement I would expect to see very low confidence reads which should be ignored if your minimum confidence level is high enough. Typically I see the minimum confidence level set to around 80-85% which means the low confidence readings below that would be disregarded and you should not see them at all.

U
Undisclosed #1
Oct 05, 2017

Generally yes, the simplest way to detect motion is simply to check how many pixels in two consecutive images have changed. With sensitivity and other adjustments - depending on the algorithms - you can define what you consider motion, ie. perhaps how big of a "blob" is considered motion and how sensitive it is to lighting changes etc. Moving the entire camera is surely quite disruptive for simple detection algorithms.

New discussion

Ask questions and get answers to your physical security questions from IPVM team members and fellow subscribers.

Newest discussions