Closed madansu closed 7 months ago
Please keep in mind that object masks filter out objects based on the bottom center pixel of the objects bounding box. Need to see an example screenshot with the mask editor
When grabbing screenshots, I noticed that some of the masks were out of whack. I rebult the object masks and will continue to monitor. I should have something to report back over the next day or so.
Here is an example of the issue i am seeing -
https://github.com/blakeblackshear/frigate/assets/31675110/27dbae43-bcbc-4289-9773-a9d13c54bbb3
The area where the car is detected, is covered by an object mask for cars, but the car is still detected there.
need to see screenshot of the object mask
Here is the object mask config for the camera -
objects:
filters:
car:
mask:
- 0,576,143,720,222,720,563,720,822,720,958,720,1072,384,974,315,899,262,797,232,682,180,514,173,184,216,82,250,0,388
- 111,137,147,222,0,262,0,155
- 552,95,612,92,726,100,759,108,812,115,823,131,699,141,549,148,431,158,314,147,312,105
- 842,134,1013,139,1033,245,744,188
Here are the screenshots for the same
right, like I said previously object masks block out objects based on the bottom center of the bounding box. Here we can see that the object mask you have for that car is not covering the bottom center of that cars bounding box hence that car is not being masked out
Thank you. I will try extending the bounding box lower onto the screen to see if that fixes the issue.
Here is the new bounding box for that car object -
Here is another example of this happening. For all the detections in this clip, the bottom center of the detected area is within the object mask zone defined.
https://github.com/blakeblackshear/frigate/assets/31675110/5605a756-4714-45d4-b680-9c6795faab6c
it would be best to see the debug live view or a snapshot with the bounding box.
I'm also wondering why you track cars in general when it seems the entire view is object masked. Required zones are the recommended approach if you only want to be notified when the car first enters the driveway for example
In the examples you are showing, we can't see what the label is for the event. Can you confirm that these are "car" events? If so, are you certain you restarted frigate after updating your object masks for car?
it would be best to see the debug live view or a snapshot with the bounding box.
I'm also wondering why you track cars in general when it seems the entire view is object masked. Required zones are the recommended approach if you only want to be notified when the car first enters the driveway for example
I am looking for cars in the street. My use case is to use HA to alert us when cars pass in the street during odd hours of the day (like between 11 pm and 6 am). But, I dont want it detecting cars on my driveway or neighbors driveways during that time, and hence the object masks for the car. I am not very interested in when cars enter my driveway, but more the above use case.
That is the object mask that is set in this case.
In the examples you are showing, we can't see what the label is for the event. Can you confirm that these are "car" events? If so, are you certain you restarted frigate after updating your object masks for car?
Yes, the object being detected is a car. After the update of the config file, the frigate instance was restarted.
Here is another example from earlier this evening
This issue has been automatically marked as stale because it has not had recent activity. It will be closed if no further activity occurs. Thank you for your contributions.
Describe the problem you are having
I have a camera pointing out to our street,and I have object masks setup to prevent detection of cars in certain areas (for example - cars in our driveway, or our neighbors driverways).
However, On any motion, Frigate detects cars in these specific areas.
This is happening on our garage camera.
Can you please help me look into this ?
Version
0.13 RC 1
Frigate config file
Relevant log output
Frigate stats
Operating system
Debian
Install method
Docker Compose
Coral version
USB
Any other information that may be helpful
Attached images of the object detection in the masked areas -