Road surface crack image detection method

Abstract

The invention discloses a pavement crack image detection method. The method comprises the following steps: acquiring a pavement image to be detected; obtaining training data, wherein the training datacomprises a plurality of road surface images and crack mark images corresponding to the road surface images; obtaining a pre-trained depth model, and constructing an initial pavement crack detectionmodel based on the depth model; based on the training data, training the initial pavement crack detection model; and acquiring a crack mark image of the to-be-detected pavement image based on the trained pavement crack detection model to obtain a pavement crack image detection result. According to the method, the pavement crack detection model can be constructed according to the pre-trained depthmodel, the model training efficiency is improved, and the precision of the obtained crack mark image is improved.

Description

technical field [0001] The invention relates to the technical field of crack detection, in particular to a road surface crack image detection method. Background technique [0002] With the rapid development of our country's economy, the mileage of highways is increasing, and the maintenance tasks of highway pavement are becoming increasingly heavy. Cracks are the initial symptoms of pavement damage. In order to prevent traffic safety from being affected, pavement cracks need to be discovered and remedied in time. However, the huge traffic flow on the highway puts forward high requirements for the implementation of maintenance tasks. The traditional manual maintenance method not only has safety hazards but also is very time-consuming, which makes the automatic pavement crack detection device gradually favored by the industry. [0003] At present, the core content of the automatic pavement crack detection device mainly includes crack image detection. The goal of crack image d...

AI Technical Summary

Problems solved by technology

For example, Zhang A et al. disclosed a kind of Based on the efficient architecture (CrackNet) of Convolutional Neural Network (CNN), but CrackNet needs to be further improved to detect finer cracks, and the network structure is related to the size of the input image, making the generalization ability of this method poor; In order to overcome the interference caused by changes in the real environment (for example, changes in lighting and shadows), Young-Jin Cha et al. wrote in the article "Deep Learning-Based Crack Damage Detection Using Convolutional Neural Networks" (Cha Y J, Choi W, O.Computer-Aided Civil and Infrastructure Engineering, 2017) discloses a computer vision-based method that uses the deep architecture of convolutional neural network (CNN) to detect concrete cracks, that is, first divides the image into blocks using a sliding window, Then use CNN to predict whether the image block contains cracks. Since this method can only find block cracks without considering the pixel level, the output of the network is not a complete crack image, but an image block; Pauly L et al. In the article "DeeperNetworks for Pavement Crack Detection" (Pauly L, Peel H, Luo S, et al. 34th International Symposium in Automation and Robotics in Construction, 2017), the use of deeper networks in computer vision-based pavement crack detection is disclosed. method, but a deeper network means more parameters and longer training time; Zou Q et al. in the article "DeepCrack: Learning Hierarchical Convolutional Features for Crack Detection" (Zou Q, Zhang Z, Li Q, et al. Image Processing, 2018) discloses an end-to-end trainable deep convolutional neural network (DeepCrack), which automatically detects cracks by learning advanced features of cracks

[0007] In the process of proposing the present disclosure, the inventors found that almost every pavement crack image detection method in the prior art has a specific scope of application, so that the existing pavement crack image detection methods can be used under the condition of clear crack image and single background , the performance is good, but it is difficult to meet the actual diverse engineering needs

For example, affected by noise such as lighting and shadows, the environment where cracks are located is usually more complex. Although the detection method based on CNN has a certain effect, the segmentation result is an image block, which overestimates the width of the crack; In the finer crack images, the cracks are not completely segmented, and there is an under-segmentation phenomenon; in order to make the FCN network converge, it is necessary to train each network in stages, resulting in a complicated and time-consuming training process, which is not conducive to the realization of the real-time process

Images