For the obtained gray-level image, the microdrill has to be separated from its background. The Otsu’s auto- thresholding method [1] is used. After thresholding, two morphological operators of open and close are applied to eliminate unexpected noise. The image of microdrill after such pre-processing is shown in Figure 4.
Figure 4: Clear microdrill image obtained after pre-processing is applied on Figure 3 (b).
3.2.2 Projection
Projection method is common and useful for edge detection and has been applied widely in machine vision. By projecting the image on the X-axis, we can have the accumulative pixel values of the image along the projected direction as shown in Figure 5. Horizontal axis of Figure 5 represents the positions corresponding to the projected direction, while vertical axis represents the intensity levels. With the projection method, the range of flute length can be obtained and is shown in Figure 5. The diameter of the microdrill
can also be obtained from the projection diagram, but its precision level is still not always satisfied. The projection algorithm is position-invariant that no adjustment is need before the measurement.
Figure 5: Illustration of the projection image of microdrill on X-axis. Both length and diameter of the microdrill can be obtained from the projection diagram.
3.2.3 Edge detection with subpixel accuracy
In order to obtain accurate edge measurement, it is necessary to determine the location of an edge to a higher resolution than the spacing between the pixels of the image sensor. This technique which is accurate to the subpixel level makes it possible for us to relax the limitation of resolution imposed by the pixel size of CCD. Many subpixel edge detection techniques ([6, 9]) have been proposed. Li et al. [3] presented a two-stage approach on edge detection and had a comparison of subpixel accuracy of several edge estimation methods.
Kisworo et al. [4] proposed a new technique for 1-D and 2-D edge feature extraction to subpixel accuracy by using edge model and local energy approach. The previous subpixel approaches might ensure subpixel edge detection. But their approaches require huge amount of computation and are difficult to be applied for high-speed and on-line measurement. In this study, a fast and simple subpixel method for 1-D edge detection was proposed for
In many cases, the gray level of an edge changes gradually, and the real position of edge is difficult to be detected. Assumed the gray level change of edge of the microdrill flute can be sampled by a scan line perpendicular to it as shown in Figure 6. Then the real edge position of this flute can be estimated by processing every pixel on the scan line through the range of flute length of the microdrill by interpolation method. Such method is so fast that it can also be applied in on-line edge detection.
Figure 6: Illustration of the proposed 1-D subpixel edge detection method.
Horizontal axis represents the positions of edge pixel, while vertical axis represents the corresponding gray levels of edge scan line.
3.2.4 Adjustment of slanting by software
The fixture is designed for convenient loading and unloading of microdrill.
When a microdrill is loading on, it may not be fixed on the fixture firmly every time. The captured image of such microdrill might be slightly slanting and cause some measuring error. Hence, it has to take the slanting of microdrill into account. We use software solution for this situation. Each center point (xi
,yi) of the flute of microdrill is found by the upper and lower edge. Afterwards, the tilt angle of the microdrill can be estimated by the method of least square linear regression to fit the center line of the microdrill, as indicated in Equation (1) and shown in Figure 7. The measuring results are rectified by the estimated
tilt angle θ, as indicated in Equations (2) and (3). Such procedure makes the measurement result still can be acknowledged when a microdrill is loaded on the fixture with a larger slanting tolerance.
⎟⎟
Adjusted diameter = Original diameter × cosθ (2)
Adjusted length = Original length × secθ (3)
Figure 7: Illustration of the estimated tilt angle by least square linear regression to fit the center line of the microdrill.
3.2.5 Outlier elimination
Using the proposed measurement algorithm, the diameter of each point is acquired in the flute length of a microdrill. Generally, the maximum of these values can symbolize the diameter if the microdrill is clean and without any particle on it. The real diameter of the microdrill is found after some procedure for eliminating the outlier values, as shown in Figure 7. The procedure is necessary especially in dirty environment while measuring.
Figure 8: The procedures for eliminating the outlier values
3.2.6 ST and UC classification
Generally, microdrills can be classified into standard (ST) type and undercut (UC) type. A microdrill with undercut type costs more but can perform an excellent hole-wall quality than the one with standard type. An UC type microdrill is characterized with a slightly larger diameter on the flute tip, as shown in Figure 9.
Figure 9: The side view of a microdrill with undercut type.
The difference between ST and UC is so small that it is extremely difficult for human eyes to classify it accurately. Therefore ST and UC microdrills are likely to be mixed up in the production line. Using the proposed measurement algorithm, the diameter of each point is acquired in the flute length of a microdrill. After comparing the maximal diameter between the tip and the body in the flute length of a microdrill, the proposed vision system performs well in the recognition of ST and UC.