Pavement contour extraction and material recognition based on landscape camera

With the rapid development of autonomous driving, road monitoring and other fields, traffic management departments are actively promoting the digital transformation of roads and intelligent maintenance applications. In this context, it is particularly important to study the technology of extracting road width, road surface material and other basic road attribute information from road images automatically. Therefore, this paper proposes a multi-branch network based on convolutional neural network for pavement contour extraction and material recognition, firstly, CNN (convolutional neural network) is used as the backbone network to extract image features, then a branch for pavement material recognition is added after the backbone network to distinguish the type of pavement, and a branch for semantic segmentation is executed in parallel to extract pavement contour, to achieve the purpose of recognizing pavement material and extracting pavement contour at the same time The automatic extraction of pavement material and pavement contour is realized, which lays a solid foundation for the automatic extraction of pavement width, automatic evaluation of pavement technical condition and intelligent road maintenance. The experimental results show that the accuracy of pavement material recognition reaches 99.2% and the IoU of pavement profile extraction reaches 96.1% on the self-collected 20822 image sets, which proves the effectiveness of the proposed network in this paper.