Exploring The Collaborative Mechanism For Pavement Crack Detection

Pavement maintenance work plays an important role in maintaining pavement safety and extending the service life of pavements.Effective and accurate detection of pavement cracks is an important content of road maintenance work.This paper studies the pavement crack detection technology,and aims to locate the crack location from the input crack image at the pixel level.At present,most pavement crack detection algorithms directly draw on the image segmentation algorithm,and ignore the difference between the crack image and the natural image.For example,the proportion of crack pixels in the image is much smaller than that of the target object in the natural image.The lack of specific design causes most existing crack detection algorithms to be unable to effectively process pavement images in complex scenes.Inspired by the research on saliency and skeleton detection,this paper explores crack detection algorithms based on saliency theory and flux field theory,and proposes a collaborative crack detection algorithm on this basis.Specifically,in the salient crack detection algorithm,this paper uses aggregation interaction module and self-interaction module,together with the use of consistency enhanced loss for crack detection.In the flux field crack detection algorithm,this paper models the crack detection as a regression task of the flux field,and designs an algorithm for the representation and prediction of the flux field and the recovery of the crack.This paper finds that,on the one hand,the saliency algorithm can accurately detect coarse cracks,while the flux field algorithm can finely find fine cracks;on the other hand,the saliency algorithm may be affected by large disturbances such as stains and shedding.The flux field algorithm may detect similar pavement textures as cracks.Considering the complementary characteristics of the two algorithms,this paper proposes a collaborative crack detection algorithm to further improve the performance of crack detection.First of all,this paper proposes a confidence estimation algorithm for detection results based on inner-outer-contrast to evaluate the detection effect of the algorithm on image regions.Then,this paper proposes a confirmation collaboration mechanism,which converts wide cracks into fine cracks and uses flux field algorithms to confirm,and converts fine cracks into wide cracks and uses saliency algorithms to confirm.In addition,for uncertain areas,this paper proposes a fusion cooperation mechanism,which uses two algorithms to detect cracks in different resolution image areas.Finally,the improved refinement algorithm was used to extract the skeleton from the crack detection results,and the characteristic parameters such as crack length,area and average width were calculated to comprehensively evaluate the pavement health.This paper uses a large-scale crack detection dataset to verify the effectiveness of the algorithm.Specifically,the ablation experiment selects appropriate parameters for the algorithm,and proves the contribution of each module to the final performance;quantitative and qualitative comparison experiments prove that the collaborative crack detection algorithm is superior to the traditional methods.In the end,this paper makes pavement crack detection software based on the collaborative crack detection model to provide support for pavement maintenance.