Traditional concrete cracks detection methods have some disadvantages such as a low efficiency,concerns of workers safety as well as Inconsistent standards.In order to overcome these problems,this paper proposes a deep learning-based binocular stereo vision method to detect the cracks and estimate its width for the first time.Considering the uneven distribution of positive and negative samples in the crack image,as well as the unclear features,the proposed method firstly applies a trained Attention_Unet model to achieve rapid crack recognition and segmentation of binocular images.Then,the edge stereo matching algorithm is used to process the cracks extracted from the left and right pairs of images to calculate the depth map of the crack pixels.After that,a binocular stereo distance-measurement algorithm will be performed to estimate the physic width of the concrete cracks.The experimental results show that the proposed method achieves a segmentation accuracy of 89.2％,and the average relative error of width estimation is 7.4％.By comparing with traditional methods,the proposed method achieves high-precision concrete crack detection and width estimation for the first time under non-contact and non-measurement conditions,greatly reducing the complexity of engineering operations.
维普
万方数据