Fig. 1 Overview of our algorithm More

Fig. 2 Generating ROI More

Fig. 3 Using ROI to crop input images to reduce processing time More

Fig. 4 Showing the change of matrix Ct to estimate tilt angle with H t More

Fig. 5 Translational accumulated error, so runway has too much scale shift More

Fig. 6 Solution for accumulated error More

Fig. 7 Method to create template point from video image More

Fig. 8 Runway feature detector program More

Fig. 9 Result of first experiment. On the left is the adjusted (warped) template. The result of matching is in the middle and the result of detecting the whole runway is on the right More

Fig. 10 Time consumption with (fast) and without (slow) runway coordination approximation More

Fig. 11 Comparison of accuracy of runway detection human expertise (small circles), ROI (light crosses), and warping image (dark crosses) More

Fig. 12 ( a ) Using only ROI and ( b ) using ROI as well, to estimate tilt angle More

Fig. 13 Robustness with image noise, adjusted by mixing different levels of Gaussian noise ( σ ) More

Fig. 14 Example of result when points of interest (P3 and P4) are not in the image More

Fig. 1 Power spectrum of random excitation More

Fig. 2 Shifted gamma distribution More

Fig. 3 Second-order response moments E[ X 2 ( t )] (left) and E [ X ˙ 2 ( t ) ] (right) normalized by second-order excitation moment E[ U 2 ] (upper: α = 0.01, middle: α = 0.05, and lower: α = 1) More