3D-scan-and-reconstruction-realsense-d435i

3D scan and reconstruction system using Intel® RealSense™ Depth Camera D435i

This project works in the system:

Intel® RealSense™ Depth Camera D435i with its tripod
Open 3D 12.0
Python 3.8
Ubuntu 20.04 LTS
Processor: Intel® Core™ i5-9600K OC @5.1GHz
Motherboard: Asus ROG Strix Z390-E Gaming
RAM: Corsair Vengeance LPX 32GB DDR4 OC @4200MHz
GPU: NVIDIA GeForce RTX2060 Super, with CUDA 11.5
Storage: Samsung 850 Evo SATA III 1TB
Turn table (set the first loop of frame as the indicator of one rotation by changing the frame_loop_at)
Object above the turn table with the maximum recommended height of 25 cm

Procedure:

Place the camera 30 cm in front of the object.
Turn on the turn table with the object to be scanned on it.
Run 3D-Scan-Record-RGBD.py.
Input the distance (cm) from the object's centre to the camera.
Input the object's maximum radius (cm) from the centre.
Let the camera scan the object until the 360-degree rotation is achieved (hit spacebar to pause and ESC to exit). The program will exit automatically and save the capture in the dataset directory.
Run 3D-Construct-RGBD.py to begin the reconstruction system.
Program will show the results of the .ply file from the reconstruction system.

Note: if the object is quite complex in terms of the surface (rough surface with many indentations), the scanning process might need to be done twice, from a higher-level point-of-view and lower-level point-of-view to ensure all surfaces are already captured and not leaving holes in the reconstructed model.

Scan position example:

Depth scanning before (top) and after (bottom) optimization of the camera preset:

Result from default and custom/optimized camera preset and reconstruction configuration:

Time elapsed for each of the reconstruction pipelines and its RMSE (Root Mean Square Error):

aircraftyue/3D-scan-and-reconstruction-realsense-d435i

3D-scan-and-reconstruction-realsense-d435i