Towards robust and optimal image stitching for pavement crack inspection and mapping

摘要

Crack detection on pavement is a vital task for road maintenance. To create a composite global view of a large pavement span, an image sequence is aligned computationally to create a continuous mosaic in this paper. The pavement image stitching is customarily solved by estimating a projective homography model that is justified when the scene is planar. However, on one hand, such condition is easily violated in practice since the pavement is often uneven and with potholes. On the other hand, the basic on which these approaches rely is sufficient matched features between images, that may be unable to meet. To this end, we propose a robust and optimal image stitching algorithm for pavement crack inspection and mapping. Firstly, point and crack pixel featuers are extracted, and a crack region alignment method is proposed to make up the situation where the matched point features are less. Then, a multilayer and multi-scale homography model is developed to deal with the problem that the scene is not a complete planar. Finally, an optimization step based on local bundle adjustment is adopted to optimally fulfil the image sequence stitching. We present convincing results to show that our method can achieve accurate stitching results from complex input pavement images. Furthermore, the proposed method can improve the image stitching performance compared with existing popular technique.