Attempt for Automated Detection of Damaged Buildings Using Aerial HDTV Images
Edge elements for collapsed buildings ( C-1, 2,3), nondamaged buildings (N-1, 2, 30 and bare ground (G-1) were detected. The intensities of edge elements were calculated by a general gradient filter with a 3x3 pixel window and were allocated a 1 byte value. The relationship between the edge intensity value and the cumulative percentage of relative frequency is shown in figure 4. on observing he data for nondamaged building 9 N-1) and ground ( G_1), it is found that their edge intensity values are almost all distributed in the range of 0-32. for collapsed buildings ( C-1, 2,3 ), it is found that the intensity values are almost all ( 95%) distributed in the range of 0-90.
| Area no. | Degree of building damage | Number of pixels | Remarks |
| C-1 | Collapse (rubble) | 2758 | A wooden building |
| C -2 | Collapse (rubble ) | 7951 | A wooden building |
| C -3 | Collapse (rubble ) | 15321 | A wooden building |
| N-1 | No damage | 10000 | The roof of a gymnasium with dark green color |
| N-2 | No damage | 14695 | The roof and wall of a wooden building with white color |
| N-3 | No damage | 7649 | A building of complicated shape with brown roofs. |
| G-1 | Ground | 5400 | A bare ground |
Table 1 The Characteristics of selected buildings

Figure 3 The selected buildings and areas

Figure 4 Relationship between edge intensity value and cumulative percentage of relative frequency

Figure 5 Relationship between variance value of edge intensity and cumulative percentage of relative frequency