Feature Extraction and Classification of Point Cloud in Urban Intersections Environment

Due to the complexity of the urban traffic scene, the changeable state of dynamic targets, and the incomplete environmental information acquired by equipment, a point cloud feature extraction and classification method based on viewpoint feature histogram (VFH) and support vector machine (SVM) is proposed according to the perception requirements of objects in a complex intersection environment. The point cloud information of typical urban intersections is preprocessed by pass-through data filtering, and the scene point cloud data is segmented by a conditional Euclidean clustering algorithm. The three-dimensional (3D) point cloud features of different objects are extracted based on the VFH, and the feature data sets of different traffic participants and facilities are established. Through the cross-validation method, the linear kernel, polynomial kernel, hyperbolic tangent kernel, Gaussian radial basis function, and other different kernel functions are used to evaluate the quality of the model. The linear kernel with a high cross-check rate is selected as the best kernel function of the SVM algorithm. The classification and recognition experiments of point cloud data in the traffic scene are carried out with logic regression, decision tree, and neural network algorithm set of urban intersection point cloud data for real-time processing. The experimental results show that the proposed method is robust to the classification and recognition of typical objects in the urban intersection environment.