1
$\begingroup$

You may think this is a duplicate, but my situation is different than previously asked questions. The only information I have is the width and height of the bounding boxes of detected people. The dataset I'm working on has images captured in different environments (street, garden, mall, ...). In other words, there is no fixed object in all images I can use as scale. The angle at which each image is captured varies drastically from almost parallel to the ground to prependicular to the ground. This means that bounding boxes don't always mimic the real size of a person. My objective is to measure the real distance between people. Any hints on how I can do this ? I am starting to think it is not feasible

Thank you!

$\endgroup$
3
  • $\begingroup$ Please, give us some image examples to think about. What do you know about the camera? Could you estimate based on optics? $\endgroup$ – P. C. Mar 9 at 12:40
  • $\begingroup$ I have no information about the camera. No focal length, no angle, nothing. The dataset is VisDrone2019. If google it you'll get sample images. The only information I have are the bounding boxes provided by the dataset. $\endgroup$ – Enzo1912 Mar 9 at 13:57
  • $\begingroup$ It's not possible, to get accurate measures with drone images you will need to estimate the parameters of the camera and distance between camera and the object. Without a reference, this won't be possible.] $\endgroup$ – Pedro Henrique Monforte Mar 10 at 20:42
0
$\begingroup$

Distance in an image is based on pixels. You will need to convert this distance to the actual real world metric like cm or inch. To make this happen, you will also need a reference object in your image whose real world dimensions are already known to you. Or you will need the distance between the object and camera. If you manage to get this distance using your camera lens, then you won't need any reference object. And then, it is very complex to get distances between objects placed in 3D space.

This is a great article that explains the aforementioned approaches in 2D space. It might be a good start for you.

$\endgroup$
1
  • $\begingroup$ Thank you for the response. I have read this the article before. My issue is that I have no information as far as the real world dimension is concerned. I only have bounding boxes of people whose sizes vary especially when having children and people crouching/sitting. $\endgroup$ – Enzo1912 Mar 9 at 14:27

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.