Improving Mesh Quality for Scanning with Intel RealSense D415 Cameras

RecFusion
1

Hello,

I am currently working on a challenging project that involves scanning premature babies using four Intel RealSense D415 cameras. The objective is to generate a 3D mesh representation of the babies to measure parameters such as length, skull circumference, and volume. I have successfully implemented the capture process, but I’m facing difficulties in improving the quality of the generated mesh.

To provide a visual reference, here’s an image of the mesh without any post-processing or filters applied:

image
image1920×1182 82.4 KB

While I have attempted to apply various filters to enhance the mesh, I’ve noticed that these filters tend to reduce the quality and fine details of the mesh, which is crucial for accurate measurements.

I’m seeking advice and recommendations on how to improve the quality of the 3D mesh generated by the RealSense D415 cameras. Specifically, I’d like to preserve as much detail as possible while still achieving a high-quality mesh for accurate measurements.

Any insights, suggestions, or best practices for post-processing the mesh in my Qt QML GUI would be greatly appreciated. Thank you for your assistance.

2

One thing you can try is to configure the Intel RealSense sensors to only provide confident depth values to reduce noise. There’s a high quality preset you can use for this and also some of the other parameters such as the laser strength can provide an improvement. The effect of the different settings can be easily assessed in the RealSense Viewer provided with the RealSense SDK. Also, try to move the sensors as close to the object as possible.
On the acquired point clouds you can add additional filtering such as outlier removal, removing measurements on the contour of objects and perhaps bilateral filtering for smoothing.

How are you using the RecFusion SDK for the stitching? The use case of multiple static cameras and static scenes is not the primary application of the SDK and due to the limited amount of data the results are more noisy than typical scans with a moving sensor.

We do have other SDKs and solutions which are more adapted to your use case. If you want to have a discussion on this, please send me a direct message.

1 Like