Mobile 3D object measurement system based on active triangulation method.

Mutka, Alan ; Nizetic, Josip ; Curkovic, Milan 等

1. INTRODUCTION

3D shape reconstruction has been an important field of research for many years. There are a lot of reconstruction methods based on triangulation like basic stereo vision, stereo vision based on simple laser light source (Lubeley et al., 2005) or coded light (Wu et al., 2006). All these methods are divided in two basic groups--passive and active. Passive reconstruction method generates results from two or more images without influencing the scene with additional energy. Algorithms for passive reconstruction are very complicated and have problems with robustness. Active methods are using and additional energy source like a laser stripe or texture generated by video projector. Additional energy improves reconstruction robustness and provides much better results. Our method is based on active stereo vision using a video projector as additional energy source and Canon EOS professional digital camera. The Video projector generates red stripes which are moved over the measured object and images are taken by the camera.

The most difficult task in 3D shape reconstruction using projected stripes is stripe peak detection. In (Forest et al., 2004) laser stripe peak detection method using a FIR filter is presented. The peak of projected stripes must be detected in subpixel precision to obtain satisfactory results. This article describes a new method for peak detection based on kernel shapes. 3D shape reconstruction points calculated using triangulation equations (Labuz, 1988) contain a lot of falsely detected points, which must be recognized and removed using advanced filters like Beziere or LSP (Bourke, 2008). Surface mesh triangulation is generated using additional algorithms like Delaunay the triangulation method.

In section II. the system configuration for 3D shape reconstruction is introduced. Section III. describes kernel shapes for line detection and methods for locating peaks in subpixel precision. The filtering method for obtained 3D points is presented in section IV. CogniLine digitalization software, automatic calibration and results are presented in section V.

2. SYSTEM CONFIGURATION

System configuration for 3D shape reconstruction is presented in Fig.1. Camera and projector have aligned optical axes in order to simplify the geometry of getting 3D points. DistanceD, DistanceO, DistanceY and DistanceX are system parameters which are obtained by calibration. System calibration is extremely important to get correct results and new algorithms for additional tuning are developed to simplify and improve the calibration procedure (Iterative Closest Point Calibration--ICPC). The video projector generates vertical lines which are shot with digital camera and processed by line extraction image processing algorithms.

[FIGURE 1 OMITTED]

3. SUBPIXEL PEAK DETECTION

Detecting the projected stripe can be improved by using controlled light conditions or manipulating camera parameters that control light adjustment (Fig. 2.).

Red stripe detection is implemented in HSV color space. Experimental results show that the most accurate detection is obtained using just the V component of HSV space. Image is also converted to black and white using additional threshold to decrease computational time.

Fig. 3. shows kernel shapes that can be used for vertical line detection. Experimental results show that shape and orientation can have a significant effect on final detection results. The kernel most commonly used in image processing (Fig. 3e.) generates only 65% of maximum detected line points produced by our kernel (Fig. 3b.).

[FIGURE 2 OMITTED]

[FIGURE 3 OMITTED]

[FIGURE 4 OMITTED]

The vertical line must be detected in subpixel precision to obtain acceptable results. First, the optimal central pixel position is obtained using the weighted average algorithm (Fisher & Nadiu 1996). In the second step Blais and Rioux linear peak detector (Blais & Rioux 1986) is used since it provides the best results for subpixel position estimation (Fisher & Nadiu 1996). Due to its linear nature, the algorithm is fast and accurate what makes it suitable for multiple image processing.

4. 3D POINTS FILTERING

From detected 2D subpixel line points and known system parameters (Fig.1.), the triangulation method generates 3D points in space. Generated 3D line points usually diverge an ideal continuous line (Fig. 4a.). Applying algorithms based on Bezier curves (Bourke, 1996) and Least-square projection (LSP) methods, the obtained points are converted to a smooth line or surface (Fig. 4b.-4d.).

Since the standard Bezier method approximates every point with other points from the point cloud, this can in some cases lead to unwanted consequences. One of them is great computational complexity when a large point cloud is used. To preserve the authenticity of the surface, every point is approximated using points from given radius R. Complexity difference between the standard Bezier method and Bezier limited with radius R is presented in equation (1).

[THETA]]new(n) [congruent to] Fehler! Textmarke nicht definiert.Fehler! Textmarke nicht definiert.Fehler! Textmarke nicht

definiert. [(R/n).sup.d] x [THETA]]std(n), d-dimension [member of] {1,2} (1)

Fehler! Textmarke nicht definiert.Fehler! Textmarke nicht definiert.Fehler! Textmarke nicht definiert.[THETA]std(n)--number of FLOPS in standard Bezier

[THETA]new(n)--number of FLOPS in Bezier with radius R

5. COGNILINE

CogniLine is a software solution intended for 3D surface reconstruction, especially body parts digitalization. The great advantage of this system over other solutions on the market is the accessibility, mobility, quality and robustness. Surface reconstruction can be obtained using any digital camera and video projector. A small deviation in projector angle, or a drift from the ideal system geometry, significantly affected the results. Therefore system configuration and calibration is simplified and results are improved using special algorithms for parameter tuning.

[FIGURE 5 OMITTED]

The ICPC method is based on matching two clouds of points using Iterative Closest Point(ICP) algorithm. ICPC matches a predefined virtual 3D point model with the real 3D point model reconstructed from measurements and the system is automatically calibrated to obtain a perfect match.

For the purposes of displaying the obtained results, it is necessary to convert 3D points into mesh. VTK (Visualization Toolkit) (Schroeder et al., 2007) is used to triangulate and filter points using special correction filters to get satisfying visualization results. Final 3D face reconstruction result is presented in Fig. 5.

6. CONCLUSION

This paper describes a simple and robust 3D object measurement system based on active triangulation method. By using off-the-shelf video projector, professional digital camera and advanced CogniLine software, satisfying results for practical applications are obtained. Information from the real world must be collected accurately to get as much information as possible, and then additional processing is carried out to get desired result. CogniLine is tested on face reconstruction problem but it can be used for any medical and other practical application where a simple and robust method for 3D reconstruction is needed.

7. REFERENCES

Blais, F & Rioux, M. (1986). Real-Time Numerical Peak Detector, Signal Processing 11, 145-155, 1986.

Fisher, R.B.; Nadiu, D.K. (1996). A Comparison of Algorithms for Subpixel Peak Detection, Dept.of Artificial Intelligence, University of Edinbourgh, UK, http://citeseer.ist.psu.edu/782699.html

Forest, Josep; Salvi, Joaquim; Cabruja, Enric & Pous, Carles (2004). Laser stripe peak detector for 3D scanners. A FIR filter approach, Proccedings of ICPR'04, Cambridge, UK.

Labuz, J. Triangulation of surface points with cameras and projectors, Dept. of Electr. & Comput. End., South Carolina Univ., Columbia, SC, The proceedings of the 20. Southeastern Symposium on System Theory, 1988.

Lubeley, D.; Jostshulte, K.; Kays, R.; Biskup, K.; & Clasbrummel, B. (2005). Laser pattern based 3D object measurement system for medical applications, Proccedings of EMBEC'05, Prague.

Paul Bourke: Bezier curves, 1996. Available from:http://astronomy.swin.edu.au/~pbourke/curves/bezier Accessed: 2008-07-24

Schroeder, William; Martin, Ken; Lorensen, Bill (2007) The Visualization Toolkit An Object-Oriented Approach to 3D Graphics, 4th Edition, Kitware, Inc. , ISBN 1-930934-19-X

Wu, H.B.; Chen, Y.; Guan, C.R. & Yu, X.Y. (2006). Measurement Technology by Structured Light Using Stripe-Edge-Based Gray Code, Journal of Physics: Conference Series 48 (2006) 537-541