3. The Test Fields
3.1 The NCKU Test Field
The NCKU test field is established indoors. The targets in the field are designed for close-range applications. There are three depth levels, as shown in Figure 1. The first and the second level each consist of several hanging metal strips. The targets are adhered to the strips. The targets of the third level are directly stuck onto the wall. The three dimensional coordinates of each target are determined with both total station and photogrammetric means. In the current experiment, several photos were taken from different angles with Kodak Advantix 2100 AUTO. Four of them are selected for further measurement and analysis.
Figure 1: The NCKU (left) and NCTU (right) Test Field
3.2 The NCTU Test Field
An outdoor test field is established on the campus of National Chiao-Tung University, Hsin-Chu. A building as shown in Figure 1 is used as the object. Natural points, such as the corner of windows, are selected as targets. The object coordinates are measured with both the conventional surveying method with a total station and the photogrammetric method with a Wild P32 metric camera. In this study, four projective stations are established, from the left to right, namely, STA1, STA2, STA3, and STA4.
4. The Photogrammetric Evaluation
Based on collinearity equation, the relationship between image and object space can be described. The correction terms implemented to model the deviation between the ideal and the real optical systems, and named as the additional parameters. According to Brown (1971), the physical distortion can be described with radial, decentering, and affine distortions. The program UNBASC1 (Moniwa, 1972), which includes the interior parameters (x0, y0, c) and distortion parameters (K1, K2, K3, P1, P2, A, B) is used in this study. This program does not have the gross error detection scheme implemented. Therefore, three times the amount of the RMSE is used as the threshold for screening the residuals. In order to avoid the spreading effect of gross errors, the gross errors are removed one at a time.
4.1 The NCKU Test Field
The image coordinates are measured with WINDIG program on an Intel-based personal computer. Each point is measured three times and the root mean square errors (RMSE) are listed in Table 6.
Table 6: The Measuring Repeatibility of Image Coordinates, NCKU Test Field
| Photo |
# of repetitions |
# of Points |
RMSE x |
RMSE y |
| mm |
pixel |
mm |
pixel |
| 2 |
3 |
82 |
0.0369 |
0.436 |
0.0495 |
0.585 |
| 3 |
3 |
84 |
0.0326 |
0.385 |
0.0373 |
0.441 |
| 5 |
3 |
115 |
0.0304 |
0.360 |
0.0325 |
0.384 |
| 7 |
3 |
106 |
0.0283 |
0.334 |
0.0266 |
0.314 |
| Sum |
3 |
387 |
0.0318 |
0.375 |
0.0364 |
0.430 |
From Table 6, the RMSEs of the image coordinate measurement are all less than one pixel (about 0.0847mm). With the space resection, some gross errors are found. After cross-examination (cross-examination) between photos, it is suspected that the locations of these points have been displaced after determination of reference coordinates. The RMSE values also change with additional parameter sets. When all distortions are considered, the residuals become the smallest.
In the space intersection stage, different stereopairs result in different RMSE values (Table 7, 8). Because stereopair 3-7 has the longest base length, and stereopair 2-3 has the shortest object distance, these two pairs have better base/height ratio, and, therefore, have better accuracy. Meanwhile, the farther the points are, the worse the result. The points on the wall have the lowest accuracy. This is because the larger the object distance, the smaller the photo scale, and then the larger the measurement error. The farther points have relatively worse geometry. That is, the larger the base/height ratio.
