Center of object - not edge - triggers zone crossing

[CinciJeff]

I've been using object detection and zone crossing since these features were introduced and had always assumed that motion was triggered when any part of the detected object entered a zone. Even the documentation refers to it this way, using the terms "enter" and "travel between". So if the object is a box, I expected the detector to trigger if any corner or side of that box crossed between one zone and another.

After reviewing some recent video and not seeing the expected motion detection results I decided to set up an experiment. I set up an old camera in front of some poster board and drew some zones on it with marker. I then made actual zones in BI that matched the outlines. I then tied some thread to a black square of paper and recorded the square moving across the board between zones. This video allowed me to experiment with different settings and compare the results. From what I observed, it's not the edge, but the center of the box that must cross a zone in order for the detector to activate.

Is this common knowledge and I missed it somehow? I'm curious if anyone else had the same misconception I did. I can tell you that it significantly changes the way I will set up my motion zones.

Here are some screenshots of my test. I'd love to share video but I couldn't figure out how to export out of BI and still include the rectangles and motion. I suppose I'll need to use Snagit or some other video capture utility to do this.

Detection criteria is "A>B", MAKE time is 0.1s. Here is the box moving from Zone A (the entire board) into Zone B. Box is not orange.


And here it is after the center passes the edge of Zone B:


I plan to use this setup to test some other aspects of motion detection including Minimum Object Size and Minimum Contrast. It's a great way to experiment with these variables in a controlled environment.

 

[aristobrat]

From what I observed, it's not the edge, but the center of the box that must cross a zone in order for the detector to activate.

Is this common knowledge and I missed it somehow?

Have you tried re-playing the Alert videos with the option enabled that shows the motion detection in progress (just right-click on the video for this option)? I'd be curious to see how quickly BI is recognizing the movement as an object, and if the time that takes is allowing part of the object to enter the zone before it trips.

 

[CinciJeff]

Have you tried re-playing the Alert videos with the option enabled that shows the motion detection in progress (just right-click on the video for this option)? I'd be curious to see how quickly BI is recognizing the movement as an object, and if the time that takes is allowing part of the object to enter the zone before it trips.

Yes, that's how I was able to play with the settings and view the results. I slowly moved the square across the field and BI picked it up as an object (yellow square) shortly after it entered.
I was actually thinking the same thing as I was composing the post, so I went and changed the "MAKE" time from 0.6 to 0.1s. There was no difference in when it tripped.

 

[Edcfish]

This is purely speculation, but I would expect it has to do with the 'Object Moves (pixels): option under the Trigger -> Motion Sensor-Advanced -> Object Detection.
Interesting test. I look forward to more of your results.

 

[CinciJeff]

This is purely speculation, but I would expect it has to do with the 'Object Moves (pixels): option under the Trigger -> Motion Sensor-Advanced -> Object Detection.
Interesting test. I look forward to more of your results.

I actually turned that option off to isolate the zone crossing so it shouldn't be a factor.

I was able to capture a screen-grab video with Snagit, so I'll try to attach it. First time doing this so we'll see how it goes:
View attachment Motion detection test.mp4

 

[CinciJeff]

For the sake of completeness, here is the zone screen. I deleted Zone A for this screenshot since it obscured B, C and D:

 

[beepsilver]

Very interesting analysis! Which motion algorithm are you using? Edge Vector is supposed to distinguish between leading and trailing edges of motion. Perhaps the software identifies the leading/trailing edges, but still requires the object to cross center of the zone before activation.

I haven't had much success with Edge Vector when using multiple zones.

 

[IAmATeaf]

I also thought there had to be a gap between zones, so G for example would be the entire screen and any overlap between A and B there should be a gap?

Edited As I’ve been kindly informed that H is actually used for hotspots, thanks

 

[Il_Pres]

Hi would like to re-use this wonderful thread in order to understand if there's a way to consider the leading edge of object box to trigger zone crossing , rather than the center point... I'm trying to setup the best spot to trigger ALPR with a snapshot taken with the car near enough to the camera for the recognition to work solidly. Center point of the box varies a lot depending on the size of the moving vehicle... I would prefer a zone crossing trigger using the leading edge of the box. Any help please?

 

[wittaj]

Hi would like to re-use this wonderful thread in order to understand if there's a way to consider the leading edge of object box to trigger zone crossing , rather than the center point... I'm trying to setup the best spot to trigger ALPR with a snapshot taken with the car near enough to the camera for the recognition to work solidly. Center point of the box varies a lot depending on the size of the moving vehicle... I would prefer a zone crossing trigger using the leading edge of the box. Any help please?

A couple of different ways for you to accomplish this.

I found this to be the setup that worked best for me. The one zone approach at this location allowed me to catch the front plate of the cars going right to left and the back plate of the cars going left to right. Occasionally I would get a large truck or a very fast vehicle that was missed, but I also run this continuous motion in the event that happens or it missed the motion. And then at night I simply make the object size box a lot smaller like the size of the plate.



The other way is since this thread originated, BI now has DeepStack integrated with it. It has added the ability to only send when AI detects a vehicle. You can assign a leading edge trigger now as well with DeepStack, But some fields of view are still problematic and you could do as another member did here and create a custom model based on your field of view that will only take a snapshot when a plate is in view. Member said it took about an hour or so to train the custom model.

 

[Il_Pres]

Thank you! that is fantastic information to experiment on...

 

[bignose3]

HI,
I know this thread was started a long time ago but recently re-visited.
I had BI for years & always assumed it was leading edge but like iL-Pres I want it the ALPR so timing is essential

My problem is (& good for reading plates) is my camera is almost face on, so it seems to struggle to get the middle of the object as small in distance & big close up.

Early testing currently but struggling to get it right, especially as size difference with Cars & Van's

I use to have the APLR camera offset a little & worked well at getting the movement in the right place (I realize now more luck than judgement as center of car) but whilst very good on the LPR, not perfect, hence the re-positioning.

As with many, costly to send excessive APLR to which ever service so only want to send those most likely.
I also have cars in the driveway close that I am trying to not pickup, which should be possible as different movement pattern & distance but not the case at the moment.