CN-119295414-B - Pavement detection method and device, electronic equipment and storage medium

CN119295414BCN 119295414 BCN119295414 BCN 119295414BCN-119295414-B

Abstract

The application provides a pavement detection method, a pavement detection device, electronic equipment and a storage medium, wherein the method comprises the steps of obtaining a pavement image; the method comprises the steps of extracting features of a pavement image to obtain feature data, respectively carrying out center point detection, offset detection and semantic segmentation on the feature data, determining a center point area of a crack and/or a seam in the pavement image through the center point detection, determining an edge offset of the crack and/or the seam in the pavement image through the offset detection, carrying out semantic recognition on the crack and/or the seam in the pavement image through the semantic segmentation to obtain a recognition result, and generating a detection result of the pavement image by combining the center point area, the edge offset and the recognition result.

Inventors

CHENG NING
GUO YUANHAO
PAN ZONGJUN
HUO YANQIANG
GE XIAOMING
CAO JIANKUN
SUN HAOYU

Assignees

中公高科养护科技股份有限公司

Dates

Publication Date: 20260512
Application Date: 20241014

Claims (7)

1. A pavement detection method, comprising: acquiring a pavement image; Extracting features of the pavement image to obtain feature data; Respectively carrying out center point detection, offset detection and semantic segmentation on the characteristic data, wherein the center point detection is used for determining the center point area of a crack and/or a seam in the pavement image, the offset detection is used for determining the edge offset of the crack and/or the seam in the pavement image, and the semantic segmentation is used for carrying out semantic recognition on the crack and/or the seam in the pavement image to obtain a recognition result; Combining the central point area, the edge offset and the recognition result to generate a detection result of the pavement image, wherein for joints, the predicted central point and the offset of two end points are fused to obtain joint positions and lengths; The detecting the center point of the characteristic data comprises the following steps: performing central point prediction on the characteristic data by using a first convolution branch which completes training to obtain a prediction central point of any line segment in the pavement image and a confidence coefficient of whether the any line segment is a straight line; according to preset rules corresponding to the cracks and/or joints, distinguishing the cracks or the joints on the basis of the confidence coefficient, and respectively expanding the area by taking the predicted central point as the center to obtain a central area, thereby completing the detection of the central point; the detecting the offset of the characteristic data includes: The method comprises the steps of obtaining a predicted central point corresponding to any line segment and a distinguishing result determined according to the confidence coefficient, and detecting the offset of the characteristic data by utilizing a second convolution branch which completes training, wherein the offset detection is specifically that the offset of two end points of any line segment to the predicted central point is determined or the offset of an external frame body where any line segment is positioned to the predicted central point is determined according to an offset calculation mode corresponding to the distinguishing result on the basis of the predicted central point; The semantic segmentation of the feature data comprises: And acquiring the prediction center point corresponding to any line segment and a distinguishing result determined according to the confidence coefficient, and performing semantic segmentation on the feature data by using a third convolution branch which completes training, wherein the semantic segmentation is specifically that each point in the pavement image belongs to a crack or a seam through the third convolution branch based on the prediction center point and the distinguishing result, so that the semantic segmentation is completed.
2. The method of claim 1, wherein the feature extraction of the road surface image comprises: And extracting the characteristics of the pavement image by using an adjusted EFFICIENTNET convolutional neural network, wherein the adjusted EFFICIENTNET convolutional neural network specifically comprises the step of adjusting the EFFICIENTNET convolutional neural network by using hole space pyramid pooling.
3. The method of claim 2, wherein the adjusting EFFICIENTNET the convolutional neural network using hole space pyramid pooling comprises: Adding a hole space pyramid pooling layer after the EFFICIENTNET convolutional neural network; Determining the image size of a first feature image output by each convolution layer in the EFFICIENTNET convolution neural network, and taking the first feature image of the last convolution layer in each image size as a feature image to be fused; and carrying out feature fusion on the second feature map output by the cavity space pyramid pooling layer and the feature map to be fused so as to extract the features, wherein the feature map to be fused is sequentially subjected to feature fusion with the second feature map according to the reverse order of the image size.
4. The method of claim 1, wherein the generating the detection result of the road surface image comprises: performing non-maximum suppression detection on any two line segments in the detection result, and respectively determining the vertical distance from one line segment to the other line segment; and responding to the minimum value in the vertical distance to meet a preset condition, acquiring the confidence coefficient of any two line segments determined in the center point detection, and deleting the line segment with smaller confidence coefficient in any two line segments.
5. A pavement detection apparatus, comprising: the acquisition module is used for acquiring the road surface image; the extraction module is used for extracting the characteristics of the pavement image to obtain characteristic data; the detection module is used for respectively carrying out center point detection, offset detection and semantic segmentation on the characteristic data, wherein the center point detection is used for determining the center point area of the crack and/or seam in the pavement image, the offset detection is used for determining the edge offset of the crack and/or seam in the pavement image, and the semantic segmentation is used for carrying out semantic recognition on the crack and/or seam in the pavement image to obtain a recognition result; The output module is used for combining the central point area, the edge offset and the identification result to generate a detection result of the pavement image, wherein for a joint, the predicted central point and the offset of two end points are fused to obtain the joint position and the length; The detecting the center point of the characteristic data comprises the following steps: performing central point prediction on the characteristic data by using a first convolution branch which completes training to obtain a prediction central point of any line segment in the pavement image and a confidence coefficient of whether the any line segment is a straight line; according to preset rules corresponding to the cracks and/or joints, distinguishing the cracks or the joints on the basis of the confidence coefficient, and respectively expanding the area by taking the predicted central point as the center to obtain a central area, thereby completing the detection of the central point; the detecting the offset of the characteristic data includes: The method comprises the steps of obtaining a predicted central point corresponding to any line segment and a distinguishing result determined according to the confidence coefficient, and detecting the offset of the characteristic data by utilizing a second convolution branch which completes training, wherein the offset detection is specifically that the offset of two end points of any line segment to the predicted central point is determined or the offset of an external frame body where any line segment is positioned to the predicted central point is determined according to an offset calculation mode corresponding to the distinguishing result on the basis of the predicted central point; The semantic segmentation of the feature data comprises: And acquiring the prediction center point corresponding to any line segment and a distinguishing result determined according to the confidence coefficient, and performing semantic segmentation on the feature data by using a third convolution branch which completes training, wherein the semantic segmentation is specifically that each point in the pavement image belongs to a crack or a seam through the third convolution branch based on the prediction center point and the distinguishing result, so that the semantic segmentation is completed.
6. An electronic device comprising a memory, a processor and a computer program stored on the memory and executable on the processor, the processor implementing the method of any one of claims 1 to 4 when the program is executed.
7. A non-transitory computer readable storage medium storing computer instructions for causing a computer to implement the method of any one of claims 1 to 4.

Description

Pavement detection method and device, electronic equipment and storage medium Technical Field The present application relates to the field of image processing technologies, and in particular, to a road surface detection method, a device, an electronic apparatus, and a storage medium. Background For highway pavement, cracks are a typical pavement damage type, however, the detection of pavement cracks alone cannot fully realize scientific evaluation of pavement technical conditions, and in many cases, the positions and treatment means of pavement joints need to be considered. When detecting the pavement, the joints and the cracks are required to be distinguished and respectively and accurately positioned, most of the current applications are mainly technical routes of 'one disease one model', the production efficiency is low, and meanwhile, a scheme for accurately detecting the joints and the cracks at the same time is not provided. Disclosure of Invention In view of the above, the present application provides a road surface detection method, apparatus, electronic device and storage medium to solve or partially solve the above-mentioned problems. Based on the above object, the present application provides a road surface detection method, comprising: acquiring a pavement image; Extracting features of the pavement image to obtain feature data; Respectively carrying out center point detection, offset detection and semantic segmentation on the characteristic data, wherein the center point detection is used for determining the center point area of a crack and/or a seam in the pavement image, the offset detection is used for determining the edge offset of the crack and/or the seam in the pavement image, and the semantic segmentation is used for carrying out semantic recognition on the crack and/or the seam in the pavement image to obtain a recognition result; And generating a detection result of the pavement image by combining the central point area, the edge offset and the identification result. In some exemplary embodiments, the feature extraction of the road surface image includes: And extracting the characteristics of the pavement image by using an adjusted EFFICIENTNET convolutional neural network, wherein the adjusted EFFICIENTNET convolutional neural network specifically comprises the step of adjusting the EFFICIENTNET convolutional neural network by using hole space pyramid pooling. In some exemplary embodiments, the adjusting EFFICIENTNET the convolutional neural network using hole space pyramid pooling includes: Adding a hole space pyramid pooling layer after the EFFICIENTNET convolutional neural network; Determining the image size of a first feature image output by each convolution layer in the EFFICIENTNET convolution neural network, and taking the first feature image of the last convolution layer in each image size as a feature image to be fused; and carrying out feature fusion on the second feature map output by the cavity space pyramid pooling layer and the feature map to be fused so as to extract the features, wherein the feature map to be fused is sequentially subjected to feature fusion with the second feature map according to the reverse order of the image size. In some exemplary embodiments, the performing center point detection on the feature data includes: performing central point prediction on the characteristic data by using a first convolution branch which completes training to obtain a prediction central point of any line segment in the pavement image and a confidence coefficient of whether the any line segment is a straight line; And according to preset rules corresponding to the cracks and/or joints, distinguishing the cracks or the joints based on the confidence coefficient, and respectively expanding the area by taking the predicted central point as the center to obtain a central area, thereby completing the detection of the central point. In some exemplary embodiments, the performing offset detection on the feature data includes: and obtaining the predicted central point corresponding to any line segment and a distinguishing result determined according to the confidence coefficient, and carrying out offset detection on the characteristic data by utilizing a second convolution branch which completes training, wherein the offset detection is specifically to determine the offset from two end points of any line segment to the predicted central point or determine the offset from an external frame body where any line segment is positioned to the predicted central point according to an offset calculation mode corresponding to the distinguishing result on the basis of the predicted central point. In some exemplary embodiments, the semantically segmenting the feature data includes: And acquiring the prediction center point corresponding to any line segment and a distinguishing result determined according to the confidence coefficient, and performing semantic segmentation on the feature data by