This paper assumed the situation communication is not conducted and execution of formation flying with only image by arranging one line measure and point-tracker. And, as the basic research for formation flying with only image, algorithm expecting attitude of line measure was studied, it was embodied by X-PLANE simulation and the performance was verified by appying algorithm.
To find the attitude with only image processing, validity of several algorithms was examined and algorithms judged to be embodied were selected. And, this paper was conducted by comparing with other algorithms.
As the algorithms drawing special features, SURF algorithm and LKT algorithm were used. Those algorithms are strong for environmental changes, so it was judged to be suitable for application of actual flying image.
And, as a method to expect an attitude, POSIT algorithm was used. It reduces errors by repeated conduction and has strongness through the repeated conduction unlike the common mathematical algorithms. The speed of calculation doesn''t show a big differences from mathematical algorithms. The calculated value of revolution was indicated by Euler angle.
As a method verifying algorithm, X-PLANE simulation program and actual flying image were used. As the result of simulation, average errors were 1.76 degree of Roll direction, 1.05 degree of Pitch direction, 1.43 degree of Yaw direction. And, standard deviation showed 0.89 of Roll direction, 0.63 of Pitch direction, 0.79 of Yaw direction. And, with actual flying image, performance was verified. Average errors were 1.3 ~ 4.9 degree of Roll direction, 1.4 ~ 5.8 degree of Pitch direction, 1.8 ~ 4.9 degree of Yaw direction. Standard deviation showed 0.6 ~ 4.1 of Roll direction, 1.1 ~ 2.6 of Pitch direction, 1.3 ~ 1.8 of Yaw direction.
More errors are shown in actual image rather than simulation because of the effects of vibration, noise, etc. and effects of installment errors, etc. caused by difficult maintenance of accurate parallel between airplane and camera during the installment of equipment.
For actual flying, tendency can be shown by about 5 degree of error, but it''s difficult to explain it''s satisfactory performance. Therefore, it''s necessary to improve performance more by attaching filter. But, in both of two cases, similar tendency is shown in actual data and image data, so it shows the result that attitude can be expected sufficiently with only image. In the future, it''s necessary to improve performance additionally through the additional study for reducing errors and study reducing computing speed.