Damage Recognition of Road Auxiliary Facilities Based on Deep Convolution Network for Segmentation and Image Region Correction

Abstract

The damage of road auxiliary facilities poses a major hidden danger to driving safety. It is urgent to study a method that can automatically detect the damage of the road auxiliary facilities and provide help for the maintenance of traffic safety auxiliary facilities. In the method for identifying the absence of road auxiliary facilities based on deep convolutional network for image segmentation and image region correction, the PointRend model based on the deep convolutional networks (CNN) is first used to achieve the pixel-level fine segmentation of the auxiliary facilities area, and then, the multiple images in the same image are segmented. In anti-glare panel area, on the largest outer polygon estimated by the convex hull algorithm, the optimal outer quadrilateral is determined according to the distance between the vertices, and then, the anti-glare panel area correction is completed by affine transformation and finally through the image one-dimensional projection mapping and adjacent shading. The distance correlation between the boards realizes the identification and positioning of the missing light-shielding board. The highway anti-glare panel missing recognition method based on deep convolution image segmentation and correction uses the vertex distance to quickly determine the external quadrilateral, which is suitable for estimating the shape of the area in a dynamic scene. After actual testing and verification, it can accurately and efficiently identify the disease of the anti-glare plate. Compared with traditional image segmentation methods, the method using the PointRend target segmentation model has better segmentation quality for target details, and it is more robust when dealing with background interference.

Automated localization of mandibular landmarks in the construction of mandibular median sagittal plane

“…PointRend (point rendering) based on CNNs is an iterative segmentation technique proposed for efficient picture target recognition and segmentation [ 21 , 22 ]. Inspired by computer graphics image rendering, this novel image segmentation method solves pixel labeling tasks and over- and under-sampling issues by treating image segmentation as a rendering problem and generating high-resolution segmentation masks [ 21 – 23 ]. With improved segmentation efficiency and application in CBCT 3D data, forward processing was performed directly on the whole sample, and the overall features of the mandible were better recognized from 3D space.…”