0
$\begingroup$

I am currently working on a computer vision project that involves analyzing video data of a person captured from a webcam. In this project, I need to compute the depth map or distance of a specific object (in my case, the nose of the person) from the camera. Unfortunately, I do not have access to any reference image with known depth or distance information for calibration purposes.

Considering the constraints of working with webcam-captured video data and the absence of reference images, are there any known or solved algorithms/methods that can help me achieve accurate depth estimation? I would greatly appreciate any insights, resources, or techniques that can be employed to estimate the depth or distance of the person from the webcam without the need for a reference image. Thank you for your assistance!

$\endgroup$

1 Answer 1

0
$\begingroup$

You can have a look at monocular depth estimation methods that only require a single image instead of a stereo pair. In particular, this approach can estimate depth without ground-truth depth maps making it unsupervised and easier to develop/train.

You can also start from a pretrained depth model, and then fine-tune on your webcam images.

$\endgroup$
1
  • $\begingroup$ What are some good pre-trained models for this purpose? Are there specific methods for capturing depth of "faces" in images? $\endgroup$
    – thedumbkid
    Commented Aug 2, 2023 at 6:31

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

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