0
$\begingroup$

I have a dataset of images taken by Kinect (RGB and D)

How can I extract the following information only from the images:

  • Frame rate

  • Camera height

The dataset containes images of person (s) moving in the scene like here. So I do have the bounding box or/and the skeleton of the person.

How can I estimate them, is there any way to do it?

$\endgroup$
0
$\begingroup$

Height is easy. You know that the camera is at $(0,0,0)$. Just fit a plane to the ground pixels and measure the orthogonal distance from the origin to this plane.

Frame rate is more of an issue. It is correlated with the speed of the user moving the camera. I suppose an approach would be to actually take the Kinect camera, move it like a normal person does and record the frame rate. You could then compute the optical flow in your dataset and record the magnitudes of flow vectors. You could then train a classifier (e.g. random forest) to learn to regress the frame rate given the flow vectors. Assuming that the camera motion is very similar, you could then put your dataset into this classifier and regress the frame rates for the whole trajectory. The average over all sets of frames should give you, kind of, the frame rate.

$\endgroup$
  • $\begingroup$ But in this way it's going to give me the distance in pixel, right? I think I need more information? In those datasets they give the position of the person and sometimes the skeleton position.. that would help? For the frame rate seems to be a good method, I'll try it. $\endgroup$ – ROS_OPENCV May 25 '16 at 11:37
0
$\begingroup$

I finished developing a first version of estimating the height of the camera, and I thought to share it with you, maybe it will be usefull for you

https://github.com/JaouadROS/heightcameraestimation

$\endgroup$

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.