In this project, the author used a surgery navigator probe developed by the group to sample a human face model, hence getting a dense unorganized point cloud. Then from these points, the 3D model of the surface could be reconstructed in the computer.

Notice here that the probe which could collect point coordinate on a surface was developed based on Computer Vision Algorithm. We are not using traditional laser sensors or handhold coordinate collector. Therefore to construct a surface based on datas collected by this machine is not that easy and the feasibility of this project direct reflect the preciseness of the probe and the effectiveness of the CV algorithm.

Also the preciseness and reliability of the navigatoris tested during the experiment. To realize the reconstruction, two existing algorithms, the Hoppe method and the Power Crust method is compared here. Finally this piece of work is related to the development of in-time surgery guide functionOf the navigator, so the coordinate matching work later is also considered and made a foundation.

Previous surface reconstruction methods have typically required additional knowledge, such as structure in the data, known surface genus, or orientation information. In contrast, the two methods outlined in this thesis requires only the 3D coordinates of the data points. From the data, the Hoppe method is able to automatically infer the topological type of the surface, its geometry, and the presence and location of features such as boundaries, creases, and corners. In the mean while, the power crust method does not depend in any way on the quality of the input point sample. Any input gives an output surface which is the 'watertight' boundary of a three-dimensional polyhedral solid. So the result of this project does not only suitable for human face model reconstruction but can also be put into the reconstruction of any other forms of object.

 

 

This is my undergraduate thesis work. In our lab, we have developed a Medical Navigator System based on binocular computer vision algorithm. The system are as follows:
Our medical navigator
See the left of the picture? This is the probe of the navigator system. There are 4 LED installed on the body of the probe. The binocular camera catches the infrared ray emitted by the LEDs, and the 2D position of every of these 4 light point are recorded by the computer and by means of Computer Vision Algorithm, the 3D coordinate of the pinpoint of the probe could then be calculated.

? Computer aided Surgery system. This is real hot topic right now. So I guess you could seek lots of more information on Google. I¡¯m just gonna stick to my work. Below are some types of medical navigator systems.

Medical Navigators

? Besides choosing the suitable algorithm and program, the hardest part of the project is to collect the data points, i.e. the sampling of 3D coordinates of the points on the object surface.

? I have to use the probe by hand and pin the point on the surface one by one. ?See the picture below: we devided the face into several different region so that we could focus on certain areas during certain period of sample session.
Human face

? Every sample session, I place the probe on one certain point and Dr Wang press to record the coordinate. We have used automatic means for data sampling such as counter, so that I could slash the probe on the surface while the computer automatically collect the data. However the plastic surface is very slippery and there are always curves, it¡¯s almost impossible to make a slash while not leaving the surface.

Sample by human hands
? Through hard work, we finally collected about 13500 points. Listed below:
point cloud
??I used C++ and VTK to develop the program to read in data and to use algorithms to reconstruct the 3D surface. The original 3D reconstruction algorithms in VTK is the Hoppe algorithm. However in our case it could only reconstruct the face kinda like a smashed tomato. -_-! Because points derived by using human hand and CV algorithms are bound to have errors and noises.
Error Points
After months of experiments, the final result was realized by using the adjusted Powercrust algorithm after doing the regional smooth. The results are very cool:
Result

 

 

 

 

SURFACE RECONSTRUCTION FROM UNORGANIZED
DENSE POINT CLOUDS
This is my undergraduate thesis.

Any Comments? Email me:)
3D Surface Bar
 

ACKNOWLEDGMENTS & Golden Memories

This project is developed together with Doctor Jianhua Wang. As the undergraduate thesis of my study, this project is challenging but rewarding. Alas, I could still remember the days when Wang and I have to take these thousands of points by hand. I put the probe on the surface and say: 'Ok' and he answered :'Ok' and click the mouse to let the camera catch the position of the probe. And now I even almost finished my graduate study.

 

Introduction

For more detail, send me an email to view the paper

( only Chinese version available)

My Work

Papers