A Fast and Robust Stitching Method for Parallax Images
When small rotary-wing unmanned aerial vehicles fly at ultra-low altitudes,images captured at different positions often suffer from significant parallax.Traditional global projection-based stitching methods tend to produce artifacts or distortions when dealing with parallax images.Recent adaptive stitching methods,such as As-Projective-As-Possible(APAP)and its optimization techniques,utilize"grid"to construct dense local projection fields to improve the alignment and naturalness of parallax image stitching,but they are computationally complex.In order to achieve this,a novel method for stitching parallax images is proposed,utilizing Deep Dense Projection(Deep-DP)as the feature matching framework.A lightweight parallel computing framework is designed to solve the smooth warping field,enabling seamless stitching.In comparative experiments,the proposed algorithm achieves an average runtime of 0.2 s,significantly outperforming other state-of-the-art algorithms,while attaining alignment accuracy and naturalness at a similar level.These results indicate that the proposed algorithm significantly enhances computational efficiency while ensuring image stitching accuracy.
image stitchingparallax imagelightweightwarping field
万方数据
维普
