Release 0.10.0

[blakeblackshear]

Placeholder for 0.10.0 release

Some potential ideas:

• In progress events
• Avoid copying cached segments unless events in progress
• Stationary object detection
• Motion data for recording segments

[netlify]

✔️ Deploy Preview for frigate-docs ready!

🔨 Explore the source changes: e5714f5

🔍 Inspect the deploy log: https://app.netlify.com/sites/frigate-docs/deploys/620931a54d0f24000728333f

😎 Browse the preview: https://deploy-preview-1927--frigate-docs.netlify.app

[Raphael909]

Thank you and congratulations on 0.9.1 release.
I was wondering what are your thoughts on this here #174.

[blakeblackshear]

I was wondering what are your thoughts on this here #174.

It's from quite a long time ago. Still want to improve tracking, but this may not be how I end up doing it.

[eatoff8]

Thanks for all your hard work Blake, intereseted to see how the stationary tracking will work. I currently use DOODS for scanning a room to make sure its empty before turning off the lights. Hopefully stationary tracking will allow me to replace that.

[shubness]

+10000 Stationary object detection

[tuxut]

Hi,
I have experimented with stationary object detection on the basis of the MQTT event messages send by frigate. I want to drive a tasmota based relais to illuminate my driveway and often the light is switched on by the redetection of my both cars parking in front of the camera.
To test the algorithms I build a pyqt and opencv based live view (to view the RTMP-Stream) where I overlay the detected objects to the "real" live video of 30fps and add the identified stationary objects with their score by using the alpha channel (better rated objects are brighter). If I let run the test view for a longer time the both cars are get scored up to a value that let them clearly identified as stationary. This development is in a early stage if there is interest, I can share what I have done so far. The algorithm is simple: any object is compared to what have been seen before and if the distance of the center is within a range and the area is below 10% in difference the score for the prior detected objects is increased by 10 all other (not detected objects) are decreased by one. The area size and center position is also updated with the newly get event by averaging. If two objects are moved into another after time they are merged to one object and the score of the two is added for the object that will survive. I will test if it works well to merge the two objects by their score. Up to now they are merged by 10% of the lower rated object and 90% of the higher rated one. What do not work well: in a zone where often objects pass by an object "zone" is identified that has a too high score to be ignored due to the fact that motion and detection here is to frequent. In my case I can define a "zone" that tells my algorithm where not to score (or reduce the given score) object. Or to define where stationary objects are more likely to get a better separation. After all I have to say that frigate is a real enrichment and I have fun with it. Would be nice if I can contribute with this. But up to now I am not sure how to set up a proper development system. Would be nice if there will be better documentation about this. To use the docker image "as a user" is a piece of cake even if I had not much experience with docker (as a developer).

[blakeblackshear]

There is a contributing guide in the docs. It sounds like you are describing an alternative tracking system. Stationary object tracking should be relatively straightforward to incorporate into frigate again.

[maxi1134]

How to use this exactly? I don't seem to find any documentation!

[NickM-27]

How to use this exactly? I don't seem to find any documentation!

It's not clear what you're referring to (frigate in general?) but the docs are https://docs.frigate.video

[maxi1134]

Oh sorry! I thought I was sure I was on a thread for the RTSP restream function that was planned for 10.0