Page 238 - Mechatronics for Safety, Security and Dependability in a New Era

Ch45-I044963.fm Page 222 Tuesday, August 1, 2006 3:55 PM
Ch45-I044963.fm
222
222 Page 222 Tuesday, August 1, 2006 3:55 PM
performed because the number of edge pixels was reduced.

(a) (b) (c) (d)
Figure 5: Generated template images with selected edge pixels

(a) (b) (c) (d)
(d)
(a)
(b)
(c)
Figure 6: Matching results by the proposed method
TABLE 2
RECOGNITION RATE BY THE PROPOSED METHOD
Template (a) (b) (c) (A) Ave.
Proposed method 100% 96% 90% 100% 96.5%
The proposed matching method was executed by using the generated template images with selected
edge pixels. Fig. 6 shows examples of matching results by the proposed method. It is found the
inclined target image areas were sought successfully. Over 90% recognition rates were recorded as
shown in Table 2. The proposed method reduced the large number of edge pixels for matching by
using the selected template table and the method recorded the equal or higher recognition rates than
using the template image with all edge pixels.

TABLE 3
MATCHING TIME [ms]

(a) (b) (c) (d) Ave.
Method 1 75 115 296 98 146.0
Method2 71 121 223 99 128.5
Proposed method 67 110 142 89 102.0
Table 3 shows the processing time to search for the target image area. The averaged processing time of
the proposed method was 102ms. The proposed method that was used a template image with selected
edge pixels was recorded the equal or faster processing time than the method 1 and method2. The
proposed method can search a target component even if the component has a free location and an
inclination and the method is supposed to be realized the high-speed matching for industrial use.

REFERENCES

Hollond J.H. (1975). Adaptation in Natural and Artificial Systems, The Univ. Michigan Press
Saitoh F. (2003). Rotation Invariant Image Matching Based on Correlation of Curvature
Distribution. Electrical Engineering in Japan 145:4, 975-981.