It's all not accurate enough. Depth sensing needs sub-millimeter accuracy in order to be really useful. Just observe how most companies show false-color depth maps which are useless to really understand their quality. A simple Phon-shaded mesh will very clearly show the shortcomings...
Indeed you are right. Could you point us to a company that shows clear and useful depth results when talking about their 3D solution? To my knowledge, there is no clear figure of merit defined for 3D performance - therefore every company gives different arguments
I don't have a link right now, but simply take your depth map, triangulate it and render it using some simple Phong shading - that certainly is not quantitative but gives a very good impression to the human eye on what to expect from the depth data, as it will clearly show the surface curvature and with that the amount of detail and noise in the data. Sure, you can also game this approach by choosing some very non-standard lighting, special scenes/subjects or some very low Phong coefficient. But to me it seems that companies do not even try to convey a clear result. Rainbow-color-coded depth maps are possibly the worst way to show the depth quality.
Interesting that they don't see to offer any TOF camera anymore?
ReplyDeleteIt's all not accurate enough. Depth sensing needs sub-millimeter accuracy in order to be really useful. Just observe how most companies show false-color depth maps which are useless to really understand their quality. A simple Phon-shaded mesh will very clearly show the shortcomings...
ReplyDeleteIndeed you are right.
DeleteCould you point us to a company that shows clear and useful depth results when talking about their 3D solution?
To my knowledge, there is no clear figure of merit defined for 3D performance - therefore every company gives different arguments
I don't have a link right now, but simply take your depth map, triangulate it and render it using some simple Phong shading - that certainly is not quantitative but gives a very good impression to the human eye on what to expect from the depth data, as it will clearly show the surface curvature and with that the amount of detail and noise in the data. Sure, you can also game this approach by choosing some very non-standard lighting, special scenes/subjects or some very low Phong coefficient. But to me it seems that companies do not even try to convey a clear result. Rainbow-color-coded depth maps are possibly the worst way to show the depth quality.
Delete