Stereoscopic Scanner in Quality Control

A very simple method of making stereoscopic image pairs from small objects by means of flatbed scanners has been recently developed. Due to the non-parallel optics of modern flatbed scanners using CCD-elements, objects can be scanned with slightly different viewing angles by simply putting them on different lateral positions on the glass surface. As a consequence of the high resolution and the good optical properties of modern flatbed scanners, this method is particularly well suited for applications in quality control. After a brief description of the principle, a new low cost stereo / 3D scanner based on this principle is presented. 1. THE PRINCIPLE OF STEREOSCOPIC SCANNING Due to their great depth of focus, which is in the order of a few centimeters, modern flatbed scanners can be used not only for scanning printed paper, but also for scanning three-dimensional objects. Since they use CCD-elements as light sensitive elements, which are much smaller than the glass surface, special optics are needed. Fig. 1 gives a simplified view of these optics. In the direction of motion of the scanning bar (x) objects are scanned according to a parallel projection, as is well known from technical drawings. Perpendicular to this direction (y) the projection lines are diverging, giving a point projection, as it is known for example from the human eye. Thus, flatbed scanners have a very interesting projection geometry, combining the properties of parallel Reprint, Proc. Conf. 'Quality Control by Artificial Vision', Trois-Rivières, Québec., Canada, May 1999 and of point projection. Exact measures of the scanned objects can be taken without any corrections in the direction of motion of the scanning bar. For all other directions, the exact geometry must be taken into account. At first glance one could suppose that a parallel projection in both directions would be much better. For a lot of applications this seems to be the case. But the diverging projection lines in the direction perpendicular to the motion of the scanning bar give the possibility to scan objects with slightly different viewing angles by simply putting them on different positions along y on the glass surface of the scanner. In this way one gets the seemingly paradoxical and thus not presumed possibility of scanning objects in three dimensions on a flatbed scanner, which has first been published in [1] and which will be explained on different examples in the next chapter. 1 2. RESISTORS AND ROSES Fig. 2 Biscuits in a Box scanned with a conventional flatbed scanner in top down orientation. Two neighboring boxes give a stereoscopic image when viewed with crossed eyes. The box in fig. 2 has been scanned in five different positions along y (s. fig. 1) directly beneath the glass surface of a conventional flatbed scanner in upside down orientation. The box has a width of 8 cm and thus is much bigger than the distance between the adjacent scanning positions, which had been chosen to be about 2.5 cm. Nevertheless, in the picture above one can simply imagine looking simultaneously at the box in the different scanning positions through the glass surface. Looking at the lateral walls of the box, one clearly sees the different viewing angles for the different scanning positions. Referring to fig. 1, it is clear that it is necessary to look at adjacent boxes with crossed eyes in order to get a stereoscopic impression. Once having succeeded in seeing one box stereoscopically, it is possible without any problems to move the eyes along the whole row without losing the stereoscopic impression. In this way, the whole picture with the five boxes behaves like an autostereogram, which were in the media a few years ago ('Magic Eye'). The five single pictures give four stereoscopic image pairs. One really has the impression of looking into four adjacent boxes, along the line of symmetry. As the boxes have a width of 8 cm and have been displaced only 2.5 cm between different scans, adjacent walls of neighboring boxes appear not to be parallel. As a consequence, the right and the left wall of one box also seem not to be parallel in contrast to what they really are. Cardboard boxes as used in this case are ideal for stereoscopic scanning. First, they allow movement of the objects in a defined manner, while keeping the original orientation. The rim of the box can be used to define where the single images must be cut so that the opening of the box lies in the paper plane when viewing the stereoscopic image. Stray light is prevented form entering the CCD-array although the document cover cannot be closed. Further, the boxes reduce lateral shadows caused by the scanner lamp. Finally, the boxes allow for an identical illumination in the different lateral scanning positions. Without a box, objects placed close to the lateral edges of the glass surface are less illuminated than objects placed closer to the center of the glass surface. ______________ There is a second class of flatbed scanners which use so called contact image sensors (CIS) instead of CCD-elements with the corresponding optics. They can not be used for stereoscopic scanning. QCAV99Reprintbc.doc 2/6 Reprint, Proc. Conf. 'Quality Control by Artificial Vision', Trois-Rivières, Québec., Canada, May 1999 QCAV99Reprintbc.doc 3/6 Fig. 3 Electronic Christmas-tree: The height of the printed circuit board is about 1 mm, that of bright capacitor about 2 mm. Fig. 4 gives an enlarged view of the central part. (left picture for left eye right picture for right eye) Fig. 4 Surface Mount Devices (close up of central part of electronic Christmas-tree shown in fig. 3.) Fig. 5 Electronic Alarm for CPU-fan, as used in modern PCs; Original size: 17 mm x 42 mm. Fig. 3 gives the stereoscopic image of an "electronic Christmas-tree" as used by hobbyists during Christmas time. This is a typical example of modern printed circuit boards as used in consumer electronics, computers and so on. The resisters, capacitors, LEDs, integrated circuits and so on come in the form of so called "surface mount devices", which are directly soldered to the printed circuit board, resulting in a very flat design. The board itself has a thickens of about one millimeter. The highest device mounted on it, the bright capacitor, has a height of about two millimeters. To give a better impression, the two pictures have been taken in two positions about 7,5 cm apart on the glass surface of the scanner. Thus, the devices on the board appear much higher than they really are. The enlarged view of the central part (Fig. 4), on the contrary, has been taken at two positions about 2.5 cm apart giving a more realistic impression of the height of the devices. Fig. 5 shows an alarm device for fans used in modern PCs. On the small printed circuit board (42 mm x 17 mm) the electronic devices, which are in this case conventional wired devices, are clearly seen in three dimensions. 3. PARTICULARITIES OF A SCANNER AS STEREOSCOPIC NEAR FIELD CAMERA Taking for example the CPU-alarm in fig. 5, it is difficult at first glance to see that the stereoscopic image has been taken with a flatbed scanner and not with a stereoscopic camera. There are nevertheless a lot of differences between the two methods worthwhile noting. First, the scanner can be used only for small, especially flat objects. The scanner does not take the whole picture at the same time, thus limiting its application in general to static objects. On the other hand, it is much cheaper and much easier to use than a stereoscopic camera. Putting the objects in a defined position on the glass surface gives very reproducible pictures without all the equipment needed for adjusting the stereo camera, the objects and the light sources. The scanner lamp moves together with the CCD-array giving homogenous illumination conditions throughout the whole scan. In some respect, this illumination gives a somewhat artificial impression. One could assume that the overall effect of the moving lamp would be the same as that of a light source covering the whole scanner area. This is however not the case, as can be seen from the streaks on the cherries (fig. 6) and chocolate balls (fig. 2), clearly resulting from the reflections of the scanner lamp. The human eye is somewhat confused when it tried to determine what the light source is and where it is placed. From the streaks in fig. 2 and 6 one would guess that there is a one-dimensional light source, as for example a fluorescent lamp a few meters away from the object. In this way however, it cannot be explained why the light Reprint, Proc. Conf. 'Quality Control by Artificial Vision', Trois-Rivières, Québec., Canada, May 1999 QCAV99Reprintbc.doc 4/6 Fig. 6 Cherries on a Dish. intensity falls off rapidly for objects a few centimeters further away. In this respect, the topmost cherry in fig. 6 is far too bright. The most striking difference however is the cylindrical optics of the flatbed scanner in contrast to the point projection of conventional cameras. If the observer knows or may guess the form of an object, he may be somewhat confused. The box in fig. 2 being rectangular, there is not point in space where one sees the lower and upper walls exactly from top and the left and right walls obliquely. The distortion of the dish in fig. 6 is also a consequence of the cylindrical geometry. On the other hand, there is no way to tell that the rose in fig. 7 is distorted. There is no reason, why a rose should not have the form given in fig. 7 even if the picture were taken in point projection with a conventional stereoscopic camera. For technical applications the cylindrical optics seems to have no disadvantages. On the contrary, it is easier to evaluate the pictures. Measures along the direction of motion of the CCD-array can be taken without any correction. Further, for small objects (<1cm) it is almost indistinguishable from point projection. For automated 3D model generation it is enough to perform a one-dimensional correlation to find the corresponding image points in the two pictures. Summarizing, one can say that the cylindrical projection is simply different from the point projection. For some applications it leads to somewhat unrealistic distortions, while in other applications it is much better than the point projection. Fig. 7 Rose on Cardboard 4. STEREOSCOPIC SCANNER IN DESIGN PHASE Xyz-table conv. scanner RS232 DLL classical TWAIN-driver stereo-module picture window conventional dialog stereo images movies (TIF, JPG, GIF, MOV ..) “3DTWAIN-driver“ Fig. 8 Block diagram of stereoscopic scanner Concept Fig. 8 gives a block diagram of a stereoscopic scanner as described in a recent patent application [2] in its current design phase. A conventional scanner is put in upside down orientation, allowing the objects to be scanned from above. It is controlled via the classical TWAINdriver. The positioning of the objects is done by an actuator beneath the scanner, which is controlled via a conventional serial interface. The whole 3D-scanner in turn is controlled via a "3D-TWAIN-driver" which has in addition to the elements of a classical TWAIN-driver a "stereo-module", which allows to define all 3D parameters of the scan, such as the number of single scans and the displacement between the scans. For applications in quality control in small series the scanner can be equipped with a multiple sample holder consisting of a box with a lot of small compartments. In analogy to computer science, one could say that the process of stereoscopic scanning can be accelerated by pipelining (length of pipe = 4) and multitasking (6 parallel processes) with the configuration depicted in fig. 9. 2 _____________ 2 Alternatively a completely new 3D Scanner not including a conventional flatbed scanner has been proposed elsewhere [3] Reprint, Proc. Conf. 'Quality Control by Artificial Vision', Trois-Rivières, Québec., Canada, May 1999 QCAV99Reprintbc.doc 5/6 step motor scanning bar scanner housing mutiple sample holder 3D Display