4. Experiments
A number of experiments have been done to develop, test and verify the validity of this technique. This paper present some experimental results determined automatically by the proposed matching algorithm. An object surface model constructed from a neat area of a rural scene (Figure 3) is illustrated in Figure 4. The neat area of an image in Figure 5 covers an urban scene. Its object surface model is shown in Figure 6.


Figure 3: Left and right image of a rural scene

Figure 4: Surface model of the rural scene using COST

Figure 5: The neat of an urban area scene

Figure 6: Objected surface model of the urban scene using COST
The capability of the proposed DSM extraction technique can be evaluated through the accuracy of the experimental results. They were compared with the results from other approaches. The studied area from his experiment was also extracted by using the ATE procedure from a commercial photogrammetric workstation. The elevation values from sets of points derived by these two automatic terrain extraction methods are compared to the manual collections. Assuming that the data collected manually is error free, the distribution of the errors in this experiment is shown in Table 1. These results indicate that our approach, while promising, is not yielding results up to the quality of existing tools.

Table 1: Comparison of different types of error measurement

Error Type	Rural scene (m.)		Urban scene (m.)
	ATE	COST	ATE	COST
Arithmetic Mean Error	0.21	0.31	-0.12	1.56
Mean Error Magnitude	0.48	0.88	1.12	1.99
Root mean Square Error	1.33	1.86	2.33	3.82

5. Conclusions
This paper has described an integrated stereo matching technique for object surface reconstruction. The matching problem is addressed the integrating of both signal and feature based. The classic dynamic programming for feature matching is modified and enhanced to integrate signal and feature matching into a simultaneous surface determination. The initial global error statistics indicate that the methods offers promising potential in exactly the circumstances where area correlation is weak. This approach can be further developed with other features, other match criteria, and the better search method to yield even better results in the future.

References

• Apaphant, P. and Bethel, J., 1999. Integration of feature and signal matching for object surface reconstruction In; Proceedings of ISPRS-International workshop on mobile mapping mapping technology, Vol. XXXXII, Part 2W1, pp. 7A1-1 -7A1-7.
• Ayache, N. and Faverjon, B., 1987. Efficient Registration of Stereo Images by Matching Graph Description. International Journal of Computer Vision, pp. 107-131.
• Ballard, D. and Brown C., 1982. Computer vision, Prentice Hall, Inc. Englewood Cliffs, NJ.
• Bellmna, R. and Dreyfus, 1962. Applied dynamic Programming. Princeton University Press, Princeton, NJ.
• Bethel, J., 1986. The DSR11 Image Correlator. In: ACSM-ASPRS, Annual Convention, Wasthington DC., Vol. 4,pp. 44-49.
• Bethel, J., Mikhail, E., Kim, K., and Marshall, J. Intelligent map understanding -Automated feature recognition and delineation. Technical report, School of Civil Engineering, Purdue University, IN.
• Canny, J., 1986. A computational approach to edge detection. IEEE Transactions on pattern on pattern analysis and machine Intelligence, PAM1-8:429-455.
• Dhond, R.R. and Aggawal, J.K. 1989. Structure from Stereo- A Review. IEEE Transactions on Systems, Man and Cybernetics, 19(6), November -December, pp. 197-219.
• Huertas, A. and Medioni, G., 1986. Detection on intensity changes with subpixel accuracy using laplacian-Gaussian masks. IEEE Transactions on Pattern Analysis and machine Intelligence, PAM1-11(2): 121-136, July.
• Ohta, Y. and Kanade, T., 1985. Stereo by Intra- and Inter-Scanline Search using Dynamic Programming. IEEE Transactions on Pattern Analysis and Machine Intelligence, PAM1-7, March, pp. 139-154.