AN IMPROVED ALGORITHM USED IN AUTOMATIC MATCHING FOR LOW-ALTITUDE AERIAL IMAGE

Abstract. In this paper, we present an improved algorithm used for low altitude aerial image automatic matching based on SIFT operator. Compared to traditional photogrammetry based on platforms such as satellites and aerospace aircrafts, the platforms of low-altitude remote sensing system have relatively lighter weight, therefore existing rotation angle and scale differences in the stereo-images. In addition, there appears fracture lines and the discontinuities of parallax in the elevation undulating area. The characteristics above make it unsuitable for the traditional photogrammetry matching method based on grey scale correlation and the matching search strategy based on continuous parallax.

In this paper an improved SIFT(Scale-invariant feature transform) operator is applied to the automatic matching of low-altitude aerial images. Several improvements were made to enhance the feature recurrence rate, matching correct rate and speed of matching. Firstly, we applied the theory of zero-crossing in SIFT feature extraction introducing the image geometry feature in scale space detection. Secondly, correlation coefficient is used as similarity measure instead of Euclidean distance in SIFT algorithm. Thirdly, we proposed a new matching strategy based on principal orientation constrain to shorten the search distance compared to the global search in SIFT algorithm.

To demonstrate the feasibility of the approach, experiments were carried with four groups low-altitude remote sensing stereo-images from different sensors, and presented different distortions. Results showed that the improved algorithm has higher feature recurrence rate, matching correct rate and speed of matching dealing with different scale, large rotation angle, affine distortion and nonlinear distortion of low-altitude remote sensing stereo-images.