|
Hazard Mitigation
|
Automated Detection of Building Damage Due to Recent Earthquakes using Aerial Television Images
A 31 x 31 pixel window as the minimum size for the Kobe method was used for the texture analysis. Then, to determine the Rpx for the texture analysis, cumulative relative frequencies were plotted against Rpx for each training data, as shown in Figure 4. The Rpx values for the Turkey and the Taiwan images were set to 43% and 30%, respectively, because damaged and undamaged training data could be roughly distinguished by these values. Figure 5 shows distributions of the extracted damage area for these values. The result for the Turkey image is much better than that using the Kobe method, because more pixels representing building debris and collapsed buildings were extracted by this study. In addition, the result for the Taiwan image is slightly better than that using the Kobe method.
Figure 5: Building damage distribution obtained by the automated detection using the threshold values in Table 1
4. Conclusions
Automated detection of damaged buildings was carried out by the image processing of aerial images taken after the 1999 Kocaeli, Turkey and Chi-Chi, Taiwan earthquakes. In the proposed Kobe method, characteristics of the damage to wooden buildings selected as training data were defined on the basis of hue, saturation, intensity, edge intensity and variance of edge intensity, and their threshold values were established. A threshold value for a texture analysis was also determined for estimating the distribution of damaged buildings, after the damaged pixels were extracted. The damage distribution obtained by the Kobe method for the Turkey image was not in particularly good agreement with the actual damage scale, although the result of the Taiwan image was not so bad. We found that it was difficult to use the same threshold values for all the images because of the difference in some factors, such as the influence of sunshine, built environment, and so on. Therefore, the different threshold values for the image characteristics and the texture analysis were estimated for each image. The distribution results for the extracted areas were better than those obtained by the Kobe method.
The distribution of severely damage buildings was estimated by the image characteristics and texture analysis, whose threshold values were empirically determined. In order to establish more general methodologies and threshold values, the authors intend to perform further case studies in the near future.
Acknowledgments
The digital video images and the HDTV images taken in Turkey and Taiwan were provided by Dr. N. Yoshida of Sato Kogyo Co., Ltd., and the Japan Broadcasting Corporation, respectively.
References
- Earthquake Disaster Mitigation Research Center, 2000a. Report on the Kocaeli, Turkey Earthquake of August 17, 1999, EDM Technical Report, No.6.
- Earthquake Disaster Mitigation Research Center, 2000b. Report on the Chi-Chi, Taiwan Earthquake of September 21, 1999, EDM Technical Report, No.7.
- Hasegawa, H., Yamazaki, F., Matsuoka, M. and Sekimoto, I., 2000a. Extraction of Building Damage due to Earthquakes Using Aerial Television Images, Proceedings of the 12th World Conference on Earthquake Engineering, No. 1722, CD-ROM.
- Hasegawa, H., Aoki, H., Yamazaki, F. Matsuoka, M. and Sekimoto, I., 2000b. Automated Detection of Damaged Buildings Using Aerial HDTV Images, Proceedings of the IEEE 2000 International Geoscience and Remote Sensing Symposium, IEEE, CD-ROM.
- Koga, K., Miura, F. and Nozawa, M., 1998. Image Analysis for Damage Estimation of Building due to Earthquake, Proceedings of the 3rd Symposium on the Mitigation of Urban Disasters by Near-Field Earthquakes, pp. 417-418 (in Japanese).
- Ogawa, N. and Yamazaki, F., 2000. Photo-Interpretation of Building Damage due to Earthquakes Using Aerial Photographs, Proceedings of the 12th World Conference on Earthquake Engineering , No. 1906, CD-ROM.
|
|
|
|
|
