Hey Team,

I'm new here and currently working with a dual Kinect v2 setup. I've been getting great results so far and really enjoying the process.

I had a quick question regarding AI joint calibration:
Is there an ideal method or recommended approach when recording the calibration video? At the moment, I’ve just been walking around and pulling various poses. I'm wondering if there’s a specific set of movements or workflow that would help achieve optimal results with the AI joint detection?


Many thanks,
Lewis

Hello,

Below you can find core recommendations about AI-based calibration:

1. Calibration won't work if person stays at the same place all the time. Thus person should move around the whole capture area.
2. Person should be fully visible in all cameras.
3. It is recommended to put hands up from time to time - to cover all capture volume with joint positions. Hands up should not be cut out of cameras' view, so please hold hands up only near the center of capture area.
4. There should be no other people in the background. As AI tracking often selects "wrong" person - it may prefer person in the background to person close to camera. This adds noise to input data.
5. Color difference between person and background helps tracking a lot. So it is preferable to use clothing colors that differ from background.

I must say I used that too it really awesome blowed my mind. I noticed I hate light calibration I always make mistakes because room not so good for it reflection take by cams of light make calibration gg and give me awful result but this method I using on my new up coming come sing with me at dj table make my mouth open :O
Its super easy its super fast its super effective corrective get all cam position awesome correcly even this room. only got OP score good all perfect nailed it.