National Rearch Council Canada Institute for Information T echnology Conil national职业性格
de recherches Canada Institut de technologie de l'information
3D Active Vision Systems for Industrial Design Applications *
Beraldin, J.-A., Gaiani, M.
January 2005
* published in SPIE: Electronic Imaging 2005. Videometrics IX. San Jo, California, USA. January 16-20, 2005. NRC 47405.
Copyright 2005 by
National Rearch Council of Canada
Permission is granted to quote short excerpts and to reproduce figures and tables from this report, provided that the source of such material is fully acknowledged.
Evaluating the performance of clo range 3D active vision systems for
industrial design applications
J-Angelo Beraldin a, Marco Gaiani b
a*Institute for Information Technology, National Rearch Council Canada, Ottawa, On, Canada
b**INDACO, Politecnico di Milano, Milano, Italy
ABSTRACT
In recent years, active three-dimensional (3D) active vision systems or range cameras for short have come out of rearch laboratories to find niche markets in application fields as diver as industrial design, automotive manufacturing, geomatics, space exploration and cultural heritage to name a few. Many publications address different issues link to 3D nsing and processing but currently the technologies po a number of challenges to many recent urs, i.e., "what are they, how good are they and how do they compare?". The need to understand, test and integrate tho range cameras with other technologies, e.g. photogrammetry, CAD, etc. is driven by the quest for optimal resolution, accuracy, speed and cost. Before investing, urs want to be certain that a given range c
amera satisfy their operational requirements. The understanding of the basic theory and best practices associated with tho cameras are in fact fundamental to fulfilling the requirements listed above in an optimal way. This paper address the evaluation of active 3D range cameras as part of a study to better understand and lect one or a number of them to fulfill the needs of industrial design applications. In particular, object material and surface features effect, calibration and performance evaluation are discusd. Results are given for six different range cameras for clo range applications.
Keywords: range cameras, accuracy, resolution, measurement uncertainty, calibration, industrial design, standards
1.INTRODUCTION
Active three-dimensional (3D) active vision systems or range cameras for short are concerned with extracting information from the geometry and the texture of the visible surfaces in a scene, processing the data and finally, communicating the results. With recent technological advances and improved demand for range cameras, the market place has been flooded with numerous 3D vision systems that address all sorts of needs [1]. In order to take full advantage of the 3D vision system
s, one must understand not only their advantages but also their limitations. This paper covers some important issues that must be addresd before embarking in a 3D project. In particular, resolution, uncertainty and accuracy of 3D information measurement in the context of clo-range active 3D systems are covered here. A number of examples illustrating the issues are shown. The goal is not to survey all commercial 3D vision systems or prent an exhaustive list of tests of the systems chon for this paper. Instead, some theory about 3D nsing is prented and is accompanied by lected results that should give the reader some pointers in order to become more critical when picking a 3D vision system. In particular, object material and surface features effects, calibration and evaluation of distance measuring devices, and performance evaluation of range camera systems are discusd. Six range cameras were tested, four are lar-bad and two fringe projection-bad systems. The tests were part of a study to better understand and lect range cameras to fulfill the needs of industrial design applications which are esntially aimed at re-designing a particular object or finding a mathematical model (NURBS) to a physical object [2]. A short review of active 3D methods is prented in ction 2. This is followed by a description of objects material and surface features effects on 3D measurements. Issues on calibration and evaluation of range cameras are discusd in ction 4. Section 5 is dedicated to the performance evaluation of six clo range cameras. Finally, in order to help the readers further their study of this topic; conclusion remarks are given and a number of references are listed.
* angelo.a
** marco.gaiani@polimi.it
2.ACTIVE 3-D SENSING
The desire to capture shape by optical means dates back to the beginning of photography [3]. It is only with the advent of compact and powerful computers that the process of capturing shape by optical means has regained substantial interest. For example, in the early 70's, Shirai [4] published a paper describing object recognition algorithms bad on range images generated by a triangulation-bad active range camera. Non-contact measurement techniques like tho bad on structured light (active 3D vision) and passive stereo are examples of fields that have benefited from technological advances of the last 35 years or so [5]. Active range cameras that u light waves for 3D measurements can be divided into class according to different characteristics. A number of taxonomies exist in the literature [5-6]. Here we summarize the main class and give the practical operating distance camera-to-object:
Triangulation: distance scanner-object about 0.1 cm to 500 cm
•Single spot (1D)
•Profile measurement (2D)
•Area measurement (3D really 2.5D: only surface measurements and not volume measurements) o Galvanometer-bad lar scanning
o Lar probe combined with translation-rotation motors, articulated arms and coordinate measuring machines (CMM), position trackers
o Multi-point and multi-line projection三生三世十里桃花插曲
o Fringe and coded patterns projection
o Moiré effect (shadow,…)
Time delay & light coherence义卖活动的意义
•Time of flight: 100 cm to veral km
o Single point and mirror-bad scanning
远宦帖•Puld lars
梦见月亮
•AM or FM modulation
o Area measurement using micro-channel plates or custom build silicon chips (puld or AM).
•Interferometric and Holographic: wide distance range
Surprisingly, only a few optical methods exist to measurement surface information (3D images) in the range from 0.1 cm to about 500 cm. But as surveyed in reference [1], a large number of systems are offered on the market especially tho bad on optical triangulation. System level differences and performance in terms of spatial resolution, overall accuracy and speed distinguish tho commercial offerings between them. We first look at some issues related to the impact of the interaction light-matter on the quality of the 3D measurements.
3.OBJECTS MATERIAL AND SURFACE FEATURES
It is said that with structured light (active) approaches, minimal operator assistance is required to generate the 3D coordinates, and that the 3D information becomes relatively innsitive to background illumination and surface texture. The first comment is indeed true if you compare to methods bad on contact probes and rulers. But one must be aware that not all the 3D information i
s reliable [7-10, 15]. The latter comment about surface texture is somewhat true as long as focusing and image processing techniques are ud wily. Furthermore, one should remember that the underlying hypothesis of active optical geometric measurements is that the imaged surface is opaque and diffuly reflecting. Hence, not all materials can be measured accurately like vapor-blasted aluminum. Problems ari when trying to measure glass, plastics, machined metals, or marble. As reported by Godin [11], "marble departs from this hypothesis, and exhibits two important optical properties in this context: translucency, and non-homogeneity at the scale of the measurement process. This structure generates two key effects on the geometric measurement: a bias in the distance measurement, as well as in increa in noi level, when compared to a reference opaque surface". This paper shows that with marble, the noi or range uncertainty varies with the spot size and a systematic bias appears in the surface location measurement.
Similar considerations can be drawn when measuring an object made of a plastic material. Figure 1a) shows the level of noi (uncertainty) when measuring a plastic soap bottle. Artificial shading of the 3D data is ud to enhance the surface details. Figure 1c) shows an error histogram resulting from the alignment of two range images. The bottle was then painted with a primer and the measurement repeated (Figure 1b). The error histogram resulting from the alignment of the two new
麻将咋玩range images is shown in Figure 1d). The histograms contours remain bell-shaped but the standard deviation drops from 147 μm to 54 μm. The alignment was performed with Innovmetric PolyWorks Modeler TM software.
a) b)
英特尔至强处理器
c) d)
Figure 1. Measurement of a soap bottle using a 3D lar scanner with and without primer paint, a) alignment of 2 views (raw plastic), b) alignment of 2 views (primer applied to bottle), c) histogram showing the error between the images (ca a) after alignment (1 sigma = 147 μm) and d) histogram showing the error between the images (ca b) (1 sigma = 54 μm). Additionally, the type of feature being measured is an important factor affecting the accuracy of a machine vision system. The accuracy of active 3D cameras drops when measurements are performed on objects with sharp discontinuities such as edges, holes, and targets. This means that systems bad on only range will not provide sufficient data for the applications. The integration of range and intensity data to improve vision-bad three-dimensional measurement has proven to work well [8]. Therefore, lecti
ng a vision system for a particular application must take into account the ability of the system to measure the features of interest with the required accuracy. In a large number of applications (like industrial design), different types of features are required to fully reprent an object. In the processing steps, an object is reprented by geometric entities: vertices (points), boundaries (edges), and surfaces. In addition, topological parameters, or the relationships between the entities, are also part of the object reprentation. In some objects, such as polyhedron types and simple sheet metals, vertices and edges may be sufficient. However, many other manufactured objects will also require curved and free form surfaces to be measured. The capabilities of a vision system to extract and to measure accurately the different types of primitives vary from one technology to another. In addition, many applications (like found in cultural heritage) do not allow any alterations to the object to suite the vision system, e.g., by placing markings or changing the reflectivity of the surface. Finally, one should also quantify the systematic errors and uncertainties introduced by the 3D modeling or inspection software.
4.CALIBRATION AND EVALUATION OF DISTANCE MEASURING DEVICES
A measurement result has only a meaning if its uncertainty is known. Here we give a famous quotation taken from Lord Kelvin: "When you can measure what you are speaking about, and expres
s it in numbers, you know something about it; but when you cannot measure it, when you cannot express it in numbers, your knowledge of it is of a meagre and unsatisfactory kind: it may be the beginning of knowledge, but you have scarcely, in your thoughts, advanced it to the stage of science. - Sir William Thompson, Lord Kelvin (1824-1907)". It should summarize the importance of knowing how a 3D system measures physical quantities.
The statement of uncertainty is usually bad on comparisons with standards traceable to the national units (SI units) as requested by ISO 9000-9004 [10]. For example, manufacturers of theodolites and CMM manufacturers u specific standards to asss their measuring instruments. A guideline called VDI/VDE 2634 is being prepared in Germany for particular optical 3D vision systems. It contains acceptance testing and monitoring procedures uful for practical purpos for evaluating the accuracy of optical 3D measuring systems bad on area scanning. The guideline applies to optical 3D measuring systems bad on area scanning, which works according to the principle of triangulation, e.g. fringe projection (e ction 2), moiré techniques and photogrammetric/scanning systems bad on area scanning [12]. Though no internationally recognized standard or certification method exists to evaluate the accuracy, the resolution, the repeatability or the measurement uncertainty of range cameras, for the time being, the ur has to d
evi techniques to ensure a confidence level on what is being measured. Calibration and evaluation methods along with definitions of terms are fundamental in a standard. The calibration of any range camera is concerned with the extraction of the internal parameters of the range camera when a mathematical model exists (focal length, lens distortions, scanning parameters, etc.) or with a mapping of the distortions in a look-up table when some elements cannot be modeled very well. After this pha, the range measurements are available in a rectangular co-ordinate system. Following the calibration process, the accuracy, repeatability, resolution and measurement uncertainties of a range camera can be determined. This is the evaluation stage. For instance, definitions for the terms can be found in the VIM standard for metrology [13]. Usually, accuracy is measured with different positions/orientations of a calibration object (avoid using the same control points). One must also make sure that it was built with a piece of equipment that is traceable to a known standard.
A number of methods are available for the laboratory. Some evaluation methods can be quite accurate but cumbersome to u on a remote site. In practice, an object that is distinct from the calibration equipment and for which the accuracy is ten times better than that of the range camera will be employed in such an evaluation. Unfortunately, it is not easy to bring to a site an evaluation o
bject, especially if it has to be accurate. For instance, a ball bar made of a stable material with a low thermal dilatation coefficient, e.g., Invar (linear coefficient of 2 ppm/o C) can be ud as a compact traveling standard. Figure 2 shows a photograph of a ball bar made of a very stable material. This type of object can be ud to verify the accuracy of a range camera.
Figure 2. Certified Bar Ball with a nominal length of 200 mm. Effective distance uncertainty: 0.43 μm.
a)
b) Figure 3 Calibration and evaluation: a) laboratory at NRC-IIT for 3D vision systems, b) test objects ud at NRC-IIT, Coordinate Measuring Machine ud to verify our 3D objects at NRC-INMS laboratory.
A laboratory at the Institute for Information Technology (IIT) of the NRC has been dedicated to calibration and evaluation of machine vision nsors and systems (e Figure 3a). The objectives are 1) to perform preci calibration of various types of nsors and systems, 2) to monitor nsor stability over time and under variations in environmental conditions such as temperature and ambient light, 3) to evaluate system geometric measurement accuracy on a wide range of specially designed standard objects and high-precision positioning devices, and, 4) to validate computer vision algorithms, such as target and edge measurement, multi-view registration, model-bad recognition, and nsor fusion. Objects with known features and surfaces were manufactured from stable materials, with tight tolerances and measured with a CMM calibrated to 1 μm in NRC's standards laboratory (INMS). Some of the objects are ud for calibration and evaluation of vision systems (e Figure 3b).
宁海温泉
a) b)
c) Figure 4. Test objects and processing: a) pyramid with steps of different heights, b) flat target plate compod of accurately positioned high contrast targets, c) 3D coordinate target extraction after processing of intensity image generated by lar scanner.
