Why sometimes roll, pitch and yaw all report zero?

[Mona_J_1]

Hi,

I am not sure why sometimes all the information including pitch, roll, yaw report zero? Can you please assist me with this?

Here's an example photo:

Here's another example:

[samontab]

To actually get the 3D pose of your face, some features need to be detected, like eyes, mouth, etc. If some of those are not detected, then there isn't enough information to obtain the 3d pose even though the general face in 2d is detected. You can always use something like a kalman filter or a particle filter to continue tracking even when no detection is available.

[Mona_J_1]

I am not sure why the second image might have any problem, because the other images that are quite similar like that pose have the pose information. Also can you explain more about using Kalman filter for getting the pose information all the time without getting all zeros sometimes? Do you have a sample code to use with Intel RealSense 3D SDK? 

Thanks,

Mona Jalal.

samontab wrote:

To actually get the 3D pose of your face, some features need to be detected, like eyes, mouth, etc. If some of those are not detected, then there isn't enough information to obtain the 3d pose even though the general face in 2d is detected.
You can always use something like a kalman filter or a particle filter to continue tracking even when no detection is available.

[samontab]

Well, as I said before, it might be because some of the features, like eyes, are not being detected, even if the images look similar to you, they might be just different enough to not get a detection. If you don't have those features you can't really estimate the pose of the head, so you can't get the pose information of course.

Now, this is normal. In the real world you can't have a perfect signal. That's why there are some mathematical tools that you can use to filter these signals and get a continuous value even if there is noise or holes in the original data. Although I don't think this is integrated in the RealSense SDK. But all is not lost, you can use other libraries, like OpenCV or others, to implement this. The keywords for you to search are Kalman Filter,  Particle Filter/Sequential Monte Carlo method.