Uneven surface in the 3D model captured by the D405.

Camera Model

D405

LibRealSense SDK Version

2.58.4.0

Firmware Version

5.17.0.10

Operating System

Ubuntu 24.04

Kernel Version (Linux Only)

7.0.0-31-generic

MIPI Driver Version (NVIDIA Jetson Only)

No response

Platform

PC (AMD64/x86_64)

Connection Type

USB3

Language

Python

Application Segment

3D Scanning

Issue Description

The proposed system aims to convert the captured 3D data of an object into a mesh model representing its surface geometry. A Doosan A0509 robotic arm equipped with an additive manufacturing (AM) nozzle will then follow the generated mesh-based toolpath to deposit material accurately along the target surface.

Technical Problems

1. Insufficient geometric accuracy:The D405 does not provide sufficiently accurate object geometry when capturing the object within the 7–50 cm working range.
2. Unclear boundary detection,The object's boundary is not clearly defined in the captured 3D data, resulting in errors in the reconstructed geometry.
3. Surface irregularities: The captured surface contains significant noise and unevenness, producing a rough and inaccurate mesh that is unsuitable for generating a precise AM toolpath.
4. Accuracy requirement: The reconstructed geometry must achieve sub-micrometer-level accuracy . how can i solve those problem? please providing the proper process.

to meet the requirements of the proposed surface-following AM process..

Steps to Reproduce

step to produce mesh file(*ply).

1. captured 3d data from realsense-viewer
2. using cloudcompare for reducing noise

Expected Behavior

No response

Actual Behavior

No response

Relevant Log Output

Additional Context

No response