CodeProject AI and final Alert image

[cjowers]

Hello,

Is there a way to make AI scan only the Alert image for objects (checking the whole image) without creating an additional motion zone?


I'm new to codeproject AI and struggling with the below example:

I have an LPR scenario where I need my motion zone trigger on the right side of the screen (call it zone 'RIGHT', where the front bumper of a car as it is driving away), but I actually want AI to check the left side of the screen, call it zone 'LEFT' (for the rear license plate).

so I achieved this by having an extra zone 'ALL' (full screen), which is used by the AI settings. it now finds the objects as desired, but it seems to be messing up my final alert image...

Since the extra zone 'ALL' is still a motion trigger, and AI uses motion as a trigger (rather than my alert image), the AI appears to be changing my final Alert image from my original alert image (which is tightly controlled by motion zone 'RIGHT') into a different AI-processed image (which could be anywhere prior, any object found after zone 'ALL' is triggered).

the sequence goes:
-some motion is triggered by zone 'ALL'
-AI start processing images
-Alert image triggered by zone 'RIGHT'
-Alert image (hi-res) shows up in sidebar (exactly as desired, where leading edge of motion is in zone 'RIGHT', license plate is in left side of image)
-AI confirms an object in some of the image sequence
-AI changes the Alert image in the sidebar to be an earlier image in the motion sequence where it first found an object.

thanks

 

[CrazyAsYou]

Can you attach a screen/camera view or two, maybe with a line or two where you want the left/right divide?

 

[cjowers]

It seems that it was the circled setting that allowed AI to 'pick' the best image of the set for my final alert. When it is set to '0', there is nothing to replace, so that is probably what I want. (although it would be nice to scan multiple images for 'just' detection accuracy, not for changing the alert image)

And if I de-select all the zones under AI, it seems to scan the entire image (still need to verify after a specific event occurs, but so far so good!).

I'm still fighting with common LPR challenges (tree shadows on a windy day, extreme lighting changes everytime a cloud passes, parked cars changing the path of travel, headlight reflections changing trigger timing, etc) but overall its pretty good.