| In recent years,with the improvement of the road network and the increase of the number of airports,road surface maintenance and management has become an important mission in the field of transportation industry in China.Cracks are the early forms of most diseases on the pavement,so timely and accurate detection of cracks can minimize maintenance costs and improve maintenance efficiency.With the promotion of the national strategy of Outline for Building China into a Country with Strong Transportation Network,how to use machine vision technology to realize high-performance automated accurate detection of pavement cracks has become one of the most hot issues in current research.This study focus on the problem of false positive and false negative error generating caused by the interference of poor illumination condition,poor image contrast and strong noise in the actual crack detection application,the main research works in this paper are as follows:Based on the deeply analysis of the imaging principle of infrared and visible light sensors and their complementary characteristics in the crack detection application,we propose a new detection scheme based on the fusion of infrared and visible light image information.In this scheme,infrared sensor and visible light sensor are used to synchronously collect the road surface image,and the registration method based on the geometric feature extraction of salient region is used to achieve the synchronous acquisition of the road surface infrared image and visible light image.And in order to further improve the accuracy,robustness and detection speed of existing pavement crack detection algorithms based on visible image,a new pavement crack detection algorithm based on densely connected and deeply supervised network is proposed.The algorithm improves the ability of extracting crack features and the degree of information complementarity at different scales by densely connecting the network layers and deeply supervising the hidden layers from multiple scales.In addition,the class equilibrium cross entropy loss function is designed to increase theweight of crack pixel loss.The algorithm is tested on four data sets of the road surface and the experiment results show that the algorithm is superior to the comparison methods in accuracy,detection speed and robustness.The algorithm provides certain technical support for the initial fast detection based on the visible light images and the accurate determination based on multi-sensor data fusion in the later stage.And in order to solve the problem of false positive and false negative error generating caused by the interference of poor illumination condition,poor image contrast and strong noise in the actual crack detection application,a crack decision algorithm by fusing visible image and infrared image is proposed.The algorithm used the gray scale and temperature probability distribution of pixels within local region to establish a decision-level information fusion model,and obtains the crack detection results by using a variety of mathematical morphological constraints.The algorithm is tested on the data set of the road surface with strong interference,and the experiment results show that the algorithm can make full use of the comprehensive information of crack temperature and gray scale,overcome their respective shortcomings.The algorithm can provide a technical basis for the accurate detection and management of road surface cracks. |
