D405 accuracy with different skin colors
Hi, I would like to use a depth camera to track a face (3D) within a distance of approx.10 cm, no matter what skin color and preferably with different light conditions. In initial test runs using the D405 with light and dark skin tones under good light conditions, i get deviations in accuracy.
Is there a software, hardware or config solution here to increase the accuracy of the 3D dataset?
Hi Robert Hennel As the D405 model is not equipped with a projector component to cast a semi-random infrared dot pattern onto smooth surfaces to add texture to them to aid depth analysis, the face analysis may benefit from adding an external visible pattern projector to the scene. It has to be a visible pattern because the IR-Cut filter on the D405's sensors exclude infrared frequencies from the camera's perception.
Inte's white-paper guide about use of projectors with 400 Series cameras provides information about visible pattern projectors at the link below.
Hi MartyG, thank you for your answer.
Are there already implemented projects that use an external visible pattern projector and have gained experience? It would also be interesting to know if intel is working on a camera that has similar specifications to the D405 model (size, ideal range, accuracy and FOV) and has an emitter built in?
The model of visible projector illustrated in the projector paper appears to be an AAXA M6 mini-projector.
However, the paper states that a standard digital video projector can be used. The most important factor is that a suitable texture is chosen to cast from the projector.
My research did not locate an example of a visible pattern projector being used with a RealSense camera to add a non-infrared texture to a surface.
In regard to the experiences of other RealSense users with low-detailed objects used with D405, the links below may be helpful references.
I cannot comment on possible future plans for the RealSense camera product range, unfortunately.
Please sign in to leave a comment.