Pavement patching area extraction and patching type judgment method

Abstract

The invention provides a pavement patching area extraction and patching type determination method. The method comprises the following steps: step 1, patching area coarse extraction and patching type determination based on YOLOv5; 2, preprocessing the image; step 3, performing block repair candidate region extraction based on superpixel segmentation; step 4, performing false detection and removal of the block-shaped repair area; step 5, marking a block-shaped repairing area; step 6, carrying out super-pixel segmentation preprocessing on the strip-shaped repairing area; 7, judging a low-contrast area; 8, performing strip-shaped repair extraction on the high-contrast area; step 9, performing strip-shaped repair extraction of a low-contrast area; and step 10, marking the strip-shaped repair area. According to the technical scheme, block-shaped and strip-shaped repairing areas of the pavement can be extracted with high precision, and particularly, the repairing areas are extracted under a low-contrast background.

Description

technical field [0001] The present invention relates to the technical field of image processing, in particular, to a method for extracting a road repair area and determining a repair type. Background technique [0002] At present, the methods for extracting road repair areas are mainly divided into two categories: deep learning methods and traditional methods. The deep learning methods mainly include road repair area extraction methods based on convolutional neural networks; the traditional methods mainly include repair areas based on local texture binary patterns. Extraction method, patch area extraction method based on window contrast image feature extraction, patch area extraction method based on area detection The main ideas of each method are as follows: [0003] (1) The road patch extraction method based on convolutional neural network is to use multiple frames of road surface sample images as training sets to train multiple convolutional neural network models to obtai...

AI Technical Summary

Problems solved by technology

[0008] (1) The existing problems of the road patch extraction method based on deep learning are: the current road patch extraction based on deep learning adopts the semantic segmentation method, which requires pixel-level data annotation, which consumes a lot of manpower and material resources, and the training speed is slow

Without using a large amount of actual road data for training, the generalization of the model is poor, and it cannot handle the actual complex road conditions well, such as roads with ruts, stains, etc.

And when the detection results are poor, it is difficult to correct the model according to the actual situation to deal with different road conditions

[0009] (2) The problem of the road surface repair extraction method based on the local texture binary mode is that it is susceptible to noise interference and is suitable for relatively flat repair areas.

Less effective for complex situations where the contrast between the repair and surrounding pavement is low or where the pavement repair area has heavy wear, stains, cracks

The machine learning classifier used is trained by a road scene, and the practicability for new application scenarios needs to be considered

[0010] (3) The problem of the road surface repair extraction method based on window contrast is that this method cannot accurately locate and extract the repair area more completely, and it is difficult to remove ruts, water stains and large stains by false detection

[0011] (4) The problem of the repaired road surface extraction method based on area detection is that the detection effect is poor when there is a break in the repaired area or the contrast between the repaired area and the surrounding road area is low. The effect of area detection is yet to be considered

Images