Nowadays, depth cameras (e.g. Kinect), are capable to map the entire room. It is doing so, by generating point clouds (x,y,z) of the object, in front of them. But it still maps only one angle of the object in front. In order to receive 3D mapping of the object, it is necessary to picture from every angle. After all the data is received, all the angles should be transferred to one coordinate system – this process is called registration
Nowadays, depth cameras (e.g. Kinect), are capable to map the entire room. It is doing so, by generating point clouds (x,y,z) of the object, in front of them. But it still maps only one angle of the object in front. In order to receive 3D mapping of the object, it is necessary to picture from every angle. After all the data is received, all the angles should be transferred to one coordinate system – this process is called registration.
Our project answers the problem of achieving high quality registration. Our main goal is reaching a perfect match, meaning our reconstructed mapped object is identical to the original object. In the attempt to reach a perfect match, we overcame difficulties such as: unrelated depth camera error, limitation of the current registration algorithms, writing algorithm that support receiving and analyzing data from 4 cameras in real time.
During our project, we achieved high level matching of fixed objects. We built an algorithm that supports data processing in real time. And also, we designed a unique method to conduct registration. As part of our project, we used the Kinect camera, and doing so, we had to create a connection between our workspace and camera API. We created a Mex file that connects the MATLAB with the Kinect.
The System
Flow Chart
Calibration Process
The object , the calibration is done with , is a board
Calibration Results
ICP Results
Final Results
