Key points are not available for this paper at this time.
This paper presents a novel approach for automatically finding conjugate points between video images collected by a small autonomous unmanned aircraft. Our approach introduces the idea of saving the resampled patch from successive least-squares matching epochs and using them as templates for subsequent images. Tests show that this method is superior to using the first image as a template for all subsequent matching attempts. We show how the algorithm performs in terms of retention of points on successive images, distribution of points on the images, and utility when used for bundle adjustments in comparison with the conventional method of using the first image as a template. Our proposed method is able to match points on an average of 2.7 times as many images before failure compared with using the conventional method. This leads to stronger geometrical configuration, higher redundancy, and ultimately, significantly better bundle adjustment solutions.
Wilkinson et al. (2009) studied this question.