عنوان مقاله [English]
Different methods have been employed to improve the accuracy of 3D scanning, including hardware and software solutions. The most straightforward solution to improve the accuracy of scanners is to increase the number of sensor elements of the sensing array. However, such a system will not only be expensive, but will also generate a large amount of data, resulting in the need for higher computation power and memory. In addition, other affecting factors, such as lens resolution, lens distortion and aberrations, sampling rate, noise, and environmental related errors limit the final accuracy of the system. The current paper addresses the accuracy of scanning and, consequently, presents a
new technique to improve the accuracy of a 3D model reconstruction of mechanical parts. The idea, called the Mechanical Dithering Technique (MDT), takes advantage of small movements of either the sensing array or the object
and rescanning. Movements provide the sensing array with images of the object cast in different positions on the sensing array. The sensing array, being a digital device, will sample the image every time with different quantization
errors, the average of which should normally approach zero. The idea is similar to the dithering technique, but random movements are used instead of random noise. The current work deals first with the problem theoretically. The idea is then verified by experimentation. A test rig was established and controlled tests were carried out. The rig consists of a stationary CCD camera as the sensing array, a laser beam projector as the structured light source, a
conveyor belt with controllable speed, a robot arm as a programmable stand for the fine adjustment of light source position and orientation, and, finally, a geometrically known object as the control object. A triangulation technique,
along with structured laser light with Gaussian distribution, is used to generate depth information. Prior to actual experiment, the calibration process is performed to obtain the intrinsic and extrinsic parameters of the camera.
Calibration allowed the elimination of major sources of lens and camera error. In order to compute the depth information of the object, the angle of sheet of light must also be known, the value of which should be measured mechanically or by means of optics. Here, a novel simple method was used to obtain this value.
The experiment procedure includes illumination of the object with laser structured light and capturing an image, then moving the object to a new position and repeating the capturing several times. The 2D image data for each
capture is transferred to the computer and pre-processed. During extracting the 2D data point, some curve approximation methods are also used to construct a smooth curve and remove the effect of noise during the measurement process. Using MDT, different 2D images are converted to a single image, from which a
more accurate geometric model can be reconstructed. The 3D geometric model is then generated and converted into a CAD data format. Consequently, the CAD model is generated in a CAD environment. To test the functionality of the
system and to evaluate the accuracy of the developed system, several control points measured by a CMM machine were compared to those obtained by the developed system. It is observed that the system works well and MDT improve the 3D model reconstruction accuracy. However; the extra accuracy is obtained at the cost of extra scanning time. The larger number of movements, the higher the improved accuracy. But, most improvements are obtained for the first few movements.