Skip to content

Reconstruction of office from a set of 3D point clouds

Notifications You must be signed in to change notification settings

weitat95/room3DReconstruction

Repository files navigation

Reconstruction of office from a set of 3D point clouds

Goal:

Reconstruct an office room from a set of 3D point clouds acquired from an Intel RealSense depth sensor that was moved to scan different views

Data:

The data comes from a set of 3D point clouds acquired by an Intel RealSense depth sensor. It consists of N=40 frames of 3D point cloud data, where each frame is from a different viewpoint. The data is stored in a cell structure with N cells, each with 2 entries: X.Color which is the RGB value of each point and X.Location, which is the XYZ location of each point (and X is the cell Index. The original RGB or XYZ images are 480 rows by 640 columns by 3 channels but are encoded as a list of points(480*640,3).


Image from the left: The image at frame 12, point clouds extracted.


Matching of SIFT descriptors with matched points in green and unmatched points in red.


A complete point cloud model after merging 40 frames.

Left, front and right wall extraction from the room.

Releases

No releases published

Packages

No packages published

Languages