Semantic segmentation of city road elements from point cloud based on semi-supervised graph convolution
Keywords: Semantic Segmentation， deep learning
Abstract: As unstructured three-dimensional data, point cloud describe objects in a more accurate, flexible and diverse manner compared to data formats such as voxels and grids. 3D data has a wide range of applications in the field of smart cities. For example, in planning urban construction, point clouds are used to generate digital traffic maps to assist in the planning of traffic lines, urban construction, etc.To improve planning efficiency and accuracy in environmental monitoring and analysis, point cloud data can be used to conduct three-dimensional analysis of actual scenes. Modeling is used to analyze landforms, hydrogeology, and building damage to facilitate urban management and maintenance. One difficulty is that feature extraction and semantic segmentation require a large amount of annotated data for model training. In terms of feature extraction and semantic segmentation, the traditional method is to manually design feature descriptors to extract features or use deep learning methods to automatically extract features using neural networks to achieve semantic segmentation respectively. However, the training process of these methods is usually supervised learning, which requires a large amount of labeled data for model training. However, the point cloud of urban scenes is huge, and if all points are manually labeled, the process will be tedious and expensive.
To address the above challenges, we proposed an urban scene semantic segmentation method based on graph convolution and semi-supervised learning network. This method uses pre-training graph convolution to obtain initialization parameters, then fully utilizes the local and color characteristics of objects in urban scene to learn first Fine-tune the network parameters using the empirical distribution and calculate the feature vector of each point through the neighborhood. Finally, a small amount of labeled data is used to assign pseudo labels to unlabeled data, perform semantic segmentation on the assigned pseudo labels and predict the category of each point. The experimental results shows that our proposed method achieve an sota road elements semantic segmentation performance.
Submission Number: 27
Loading