Best camera for scanning faces (ear to ear)
Hello,
This is a follow-up of my question in :
https://forums.intel.com/s/feed/0D50P0000490VBoSAM
I am working with the high accuracy preset with the following parameter values :
Disparity Shift : 30
Laser Power 250mW
Depth Unit Scale: 0.0001
I am not sure whether this is enough to reach accuracy requirements of upto 1mm. Is there a preset existing for face scan from ear to ear ?
@MartyG I have been through the pdf you suggested, but don't get to our accuracy requirements. Thanks also for pointing out CloudCompare we are using this for comparison.
Ideally we would like to find deviation of points in a volume grid for each camera. Does Intel have any white paper for this?
Thanks
-
The latest new RealSense L515 camera, which is based on lidar sensing instead of the Stereo sensing of the 400 Series cameras, can capture 23 million accurate depth points per second of data. A company called LUMOplay who assessed a pre-release unit of the L515 recently gave its image quality an excellent review.
https://github.com/IntelRealSense/librealsense/issues/6214
Pre-orders for the L515 placed now will not ship until after July 26 2020, though it is designed so that you can prototype now with the existing 400 Series cameras and then seamlessly drop an L515 into a project to replace it. The RealSense SDK supports 99% of its functions for the L515 at its launch.
In regard to improving face scanning results right now with a 400 Series camera, Dorodnic the RealSense SDK Manager suggests in the link below to use post-processing in combination with a short-range preset to enhance the face image.
https://github.com/IntelRealSense/librealsense/issues/1021#issuecomment-358223376
A link to a short-range preset file can be found by visiting the link below and scrolling down to a page section headed 'Additional Presets not in the Viewer':
https://github.com/IntelRealSense/librealsense/wiki/D400-Series-Visual-Presets#preset-table
Intel has a RealSense white paper on depth post-processing.
https://dev.intelrealsense.com/docs/depth-post-processing
A RealSense community member offers their own advice in this discussion for parameters to adjust that helped them to achieve a better short-range image:
https://github.com/IntelRealSense/librealsense/issues/1021#issuecomment-365656575
Intel's RealSense white paper on improving Subpixel Linearity may also be of interest to you.
-
@MartyG : Thanks a lot for the detailed answer. The Short-Range-Preset.json combined with the subpixel linearity improved the results dramatically. I will also read the depth-post-processing bit carefully, we are already using this but perhaps the parameter values suggested could improve the results quite a bit.
I would keep you posted once we have incorporated these changes into our software about the accuracy levels we could reach. Thanks once again.
-
Another question, as I understood from the forums is that these presets are computed using machine learning algorithms. Since we use the device for medical application, would it be possible to know how this is done and whether any guarantees on results obtained using the Short-Range-Preset.json preset can be given.
Thanks.
-
The Advanced Mode settings use machine learning algorithms to govern how they interact with each other. The Visual Presets themselves are not algorithm generated. They are just a text-file list of settings that are applied to the Advanced Mode functions when the file is loaded, and then the Advanced Mode functions should adjust according to their machine learning routines.
The workings of camera algorithms are typically not made public by Intel because they are Intel's intellectual property. So this makes it difficult to provide precise guidance about how they interact. Intel have not documented a lot of the Advanced Mode functions simply because the complex interactions between its settings are an automated process. RealSense users are free to do trial and error experimentation to work out how the change to one setting affects other settings and the overall image that is output though.
The official Advanced Mode documentation says: "Advanced mode hardware APIs are designed to be safe, in the sense that the user can't brick the device. You always have the option to exit advanced mode, and fall back to default mode of operation. However, while tinkering with the advanced controls, depth quality and stream frame-rate are not guaranteed. You are making changes on your own risk".
https://github.com/IntelRealSense/librealsense/blob/master/doc/rs400/rs400_advanced_mode.md
Please sign in to leave a comment.
Comments
5 comments