工研院之三次元撷取与显示技术发展.doc

影像與識別 2000, Vol. 6, No.2 三次元擷取與顯示技術發展
工研院之三次元擷取與顯示技術發展
薛文珍　Wen-Jean Hsueh
工業技術研究院　光電工業研究所
http://www.oes.itri.org.tw/3D
ABSTRACT
This paper gives an overview of the research and development activities in 3D surface digitizing and autostereoscopic display conducted at the Industrial Technology Research Institute (ITRI) of Taiwan. As a major technology and consulting service provider of the area, for the past decade ITRI has developed 3D laser scanning digitizers ranging from low-cost portables, industrial CAD/CAM digitizing, to large human body scanner, together with 3D surface modeling software for total solution on reverse engineering to process large quantity of 3D data sets acquired from digitizers. On the other hand ITRI has for the last five years devoted to cultivate capabilities in the design of autostereoscopic display systems. Based on both hardware technologies in optics and electronics, and software technologies in 3D surface modeling and imaging, ITRI is now exploring innovative methodologies that provide higher performances, including hardware-based algorithms with advanced camera designs, optical tracking and motion capture for animation, and 3D display design concerning human factors. It is expected that both the needs for easy and fast high-quality 3D information and display in the near future will grow exponentially, at the same amazing rate as the internet and the human desire for realistic and natural images.
¨ INTRODUCTION
The Industrial Technology Research Institute (ITRI) of Taiwan has been devoting to the research and development of 3D surface digitizing and modeling for almost a decade, and in 3D autostereoscopic display for five years. As a major technology and consulting service provider of the area, ITRI has developed 3D laser scanning digitizers ranging from low-cost compacts, industrial CAD/CAM digitizing, to large human body scanner, with in-house 3D surface modeling software to provide total solution in reverse engineering. The core abilities to acquire 3D information from sophisticated-shaped objects and to process and manipulate its large number of 3D data have enabled applications in reverse engineering, body scanning, and 3D animation. With newly developed 3D display capabilities, ITRI wishes to integrate 3D input, output, and processing technologies to seek innovative applications in new domains.
¨ 3D GEOMETRIC MODELING
2.1 3D laser stripe digitizer
The concept of ITRI’s non-contact 3D digitizer is illustrated in Figure 1. A stripe of laser is projected onto the surface of an object, and the reflected light is captured by either of the CCD cameras to avoid occlusion. Based on optical triangulation and camera calibration parameters, the 3D coordinates of the illuminated data points on the object surface can be calculated. Integrating the digitizer head and a four-axis translation-rotation stage, one can fully digitize a 3D object automatically. The digitizer head has a depth of field of 170 mm, and the system scans at speeds up to 3,000 points per second and precision up to 50 mm. The digitization generates a large number of 3D coordinates from different viewing angles. A surface modeling process then takes place to make these unorganized 3D data useful. Thorough discussion of the design and analysis of the digitizer can be found in [3][7][8].
A low-cost compact version with only one CCD camera for PC-based applications was designed so that one CCD retains the information of two fields of view of the original two-CCD design to account for occlusion issues without compromising precision and accuracy [2]. We further design the scanner to equip with a color CCD camera for building a complete color 3D model.
2.2 3D surface modeling
After acquiring large number of 3D coordinates from many viewing angles, registration and merging are two fundamental steps to create a useful 3D surface model. A polygon-based method for describing a complete object is proposed [6]. Multiple range images are integrated into a single polygonal, usually triangular, mesh. Registration is to align the multiple range images into the same coordinate system. Merging then removes redundant data and stitches these images to a single mesh. Our program needs no additional information from the digitizer. Figure 2 shows the reverse engineering surface modeling result of an impeller and its rapid prototyping replica.
Surface modeling from 3D laser digitizer enables the capture of detailed description of the object surface, but the large number of 3D coordinates or polygonal information hinders efficient usage and effective applications of the surface models. We proposed a sequential decimation process to reduce the number of polygons in a triangular mesh after the registration and merging of multiple range images are performed [14]. An iterative vertex decimation method is used to remove vertices with minimum re-triangulation error. To reduce the distortion resulting from polygon reduction, vertices are characterized by local geometry and topology before the re-triangulation error is evaluated. This algorithm can be applied to not only triangular meshes generated from 3D digitizer, but also general volume meshes and terrain meshes. Figure 3 demonstrates results from our polygon reduction algorithm.
One other common issue in 3D digitizing is finding the next best view to efficiently digitize the object by calculating the next best position or a working path of the sensor, which has been well addressed by peers. In developing our low-cost scanning system that has significant less degrees of freedom in movement, however, we tackle a similarly important issue of deciding the best positions of the object for efficient scanning in building an effective 3D surface model. A low-occlusion approach is proposed to find the best viewing position for scanning by considering the position of the object instead of the sensor [1]. The efficiency improves significantly by combining carefully planned working path of the sensor and optimum positions of the object.
2.3 Applications
An important and original application of the 3D digitizers is in reverse engineering as already illustrated in Figure 2. We have consulted a wide range of traditional to hi-tech industries that adopts 3D digitizing as an indispensable tool for computer-aided design, manufacturing, and inspection, including applications in parts design, industrial design, tooling, sculpture, ergonomics, textile, and foot wear.
A good example is in the measurement and inspection of tires [10]. A 3D digitizing sensor is installed on a 4-axis mechanism to scan the whole tire surface in radial and circumferential directions. The laser stripe generates a section curve in the radial direction. The geometry of the tire can thus be evaluated, including width, diameter, circumference, arc value, and roundness. The dimensional error is always less than 0.15mm. Tread wear can also be quantified by comparing the digitized surface data to its original data, and the resolution is up to 0.1mm.
It is only natural that 3D scanning find applications in human-related measurements because of the abundant information 3D digitizers are able to provide dealing with biological and highly diversified subjects compared to the limited data from using conventional meters. As shown in Figure 4(a), our human body scanner consists of six 3D digitizers and three vertical translation axes. The range of measurement is 1900mm in height, 900mm in width, and 500mm in depth. It takes approximately 8 seconds to capture whole-body data of a 180-cm tall subject with a per-4 mm profiling, a horizontal resolution less than 2 mm, and a range resolution less than 1 mm. The body scanner satisfies the requirements for speed, safety, comfort, and efficiency, and has been used in consumer-product design and health care applications. Figure 4(b) illustrates the rendered results of scanning of a human subject [15].
To obtain realistic visual effects, more and more motion picture and animation productions adopt reverse engineering technologies. 3D digitizers and motion capture systems are two important tools in this domain. Manipulating large quantity of scanned 3D data, however, is highly inefficient and difficult for animators when using 3D digitizer to build models. In the 3D animation process, two issues are of major concerns: fast surface reconstruction and easy surface structure manipulation. Fast surface reconstruction saves time and thus cost. Easy surface structure manipulation asks for continuity maintenance and flexible orientation of coordinate system for merged surface reconstruction. We proposed a fast surface reconstruction pipeline of the 3D digitizer and used a motion capture system to integrate surface modeling technologies [9]. Figure 5 shows the results after clustering, surface reconstruction, and motion capture integrated animation rendering.
¨ AUTOSTEREOSCOPIC DISPLAY
Having experimented on several different types of autostereoscopic displays including projection 3D display [13] and double-panel 3D display, we now focus on the design of a system based on micro-retardation array to enable real-time motion parallax with infinite viewing zones limited only by the resolution of the tracking device, as shown in Figure 6. The objective is for the system to accommodate multiple viewers with independent tracking, maintain low degradation in resolution, be compatible with widely-used field-sequential stereo format, and be easy to manufacture in quantity [12].
3.1 Micro-retardation array
Because micro-polarization array requires further alignment and causes brightness degradation, we adopt micro-retardation array as the key component of the system. An LCD panel with the micro-retardation array can display stereoscopic images viewed with or without special glasses by encoding left and right images with alternate horizontal stripes of different polarization states. The micro-retardation array is a polymeric film of alternate half-wave and zero retardation stripes. The width of each stripe is of the order of hundreds of microns, as shown in Figure 7. To fabricate the micro-retardation array, we first stretch a polymeric material to get a half-retardation film, and then treat the film stripe-wise to erase the retardation of the treated area. Micro-retardation arrays are commercially manufactured using chemical processes. We instead fabricate micro-retardation arrays using CO2 laser heating process on account of its low-cost, easy to handle, and environmental-friendly advantages.
We demonstrated a green fabrication process for micro-retardation arrays with high contrast ratio and well-defined stripe boundaries. It is shown that by accurately controlling the power and spot-size of a CO2 laser, the retardation property of a polymeric film, such as PC and ARTON, can be tailored within a localized area without altering the retardation of the untreated area. By erasing the birefringence effect of polymeric films, the laser heating process can achieve fairly good results. The residual retardation of PC-based films can be as low as 6 nm. There are, however, two issues to be noted considering fabrication in quantity. Firstly, the polymeric film material should have low glass-transition temperature and low wavelength dispersion of birefringence. Secondly, fabrication throughput may be raised by, for example, scanning with multiple laser beams.
3.2 System performance analyses
The major factor that induces crosstalk (and thus reduces the contrast ratio) between viewing zones in the system is scattering of the Fresnel lens. The smaller the pitch of the Fresnel lens, the more serious the crosstalk. It can be decomposed into two parts. One is the scattering of the projected light into multiple viewing zones, and the other part is caused by the nature of de-polarization of the scattered polarized-light. These are current issues to work on. In addition, the contrast ratio of the system is very sensitive to the vertical position of the eyes. Therefore, the vertical dimension of the light source may have to be small, and a one-dimensional diffuser may be required to expand the viewing zone vertically.
¨ OUTLOOK
It is expected that the need for easy and fast high-quality 3D information in the near future will grow exponentially, at the same amazing rate as the internet and the human desire for realistic and natural images. 3D imaging can be achieved from many sources, including active scanning for model-based 3D and passive capture for vision-based 3D. Either one has its advantages, and integration at some level is expected to be a certain route towards advancement of the technology. The requirement of performances will certainly go in the direction of real-time, dynamic, high-resolution, and high-accuracy 3D imaging.
Based on both hardware and software technologies in scanning, merging, registration, surface fitting, compression, and display, ITRI is now exploring innovative methodologies that provide higher performances. A project involving hardware-based speckle image correlation algorithms with advanced camera design [4][11] wishes to provide high-speed 3D imaging capabilities. Effort in animation application will continue with integration of capabilities in high-speed tracking for motion capture. Acquiring 3D information from vision-based cameras will also be pursued. Topics in creating natural 3D color will be explored. Integration among 3D input, processing, display, and human interface technologies to spawn more innovative ideas in human-centered technologies is underway.
¨ ACKNOWLEDGEMENTS
The collected effort was made possible by continuous funding and support from the Ministry of Economic Affairs of Taiwan, ROC for the past decade and continuing. The author would like to acknowledge contributions from all members of the Optical Inspection Department and the Stereo Lab at the Opto-Electronics & Systems Labs of ITRI, with special thanks to Mr. Hsien-Chang Lin, Mr. Kuen Lee, Mr. Chia-Chen Chen, Mr.Chao-Hsu Tsai, and Dr. Bor-Tow Chen for generous help to this article.
¨ REFERENCES
[1] Chen, B.-T., W.-S. Lou, C.-C. Chen, and H.-C. Lin, “A 3D scanning system based on low-occlusion approach,” 2nd 3DIM Conf. 3-D Digital Imaging and Modeling, 506-515, Ottawa, Canada, 1999.
[2] Chen, B.-T., W.-S. Lou, C.-C. Chen, and H.-C. Lin, “Low-cost 3D range finder system,” SPIE Proc. Input/Output and Imaging Technologies, 3422:99-107, Taipei, Taiwan, 1998.
[3] Chen, B.-T., W.-S. Lou, C.-C. Chen, and H.-C. Lin, “3D digitizer: method and analysis,” 10th IPPR Conf. Computer Vision, Graphics and Image Processing, 406-412, Taichung, Taiwan, 1997.
[4] Hsueh, W.-J. and D. P. Hart, “Real-time 3D topography by speckle image correlation,” SPIE Proc. Input/Output and Imaging Technologies, 3422:108-112, Taipei, Taiwan, 1998.
[5] Hsueh, W.-J.