A method for extracting rods from vehicle point cloud based on plane selection

Abstract

The invention proposes a plane selection-based method for extracting rod-like objects from a vehicle-mounted point cloud. According to the idea of ​​octree splitting, the present invention iteratively divides the unsequenced point cloud into point sets of different scales according to the spatial position, and selects a planar slice as the initial plane; utilizes local geometric features and plane isolation detection Select and filter the initial plane to obtain the candidate plane of the rod; realize the extraction of the missing part of the rod based on the region growing algorithm in the vertical direction; represented by the candidate plane; the final filter deletes the candidate plane whose height from the lowest point of the candidate plane to the ground does not satisfy the pole-shaped object; the present invention effectively converts the extraction problem of the pole-shaped object in the vehicle-mounted point cloud data into the problem of plane selection and filtering, and finally Accurate and robust extraction of rod-like objects in vehicle-mounted point clouds.

Description

Technical field: [0001] The invention belongs to the technical field of vehicle point cloud data processing, and in particular relates to a method for extracting vehicle point cloud rods based on plane selection. Background technique: [0002] In urban-level scenarios, the vehicle-mounted laser scanning system can accurately and quickly obtain surface information of urban buildings, road traffic facilities, tunnels and other objects, and is very suitable for rapid and accurate acquisition and model reconstruction of 3D spatial information of urban objects. Extracting rod-like objects from vehicle-mounted 3D laser scanning data is conducive to creating a city geospatial information database, updating high-precision city maps in time, and promoting urban management and the construction of intelligent transportation systems. However, 3D laser scanning data has the characteristics of unorganized structure, uneven density and huge amount of data. How to efficiently and robustly e...

AI Technical Summary

Problems solved by technology

From the perspective of classification methods, it can be roughly summarized as model fitting-based rod extraction methods, geometric feature-based rod extraction methods, and learning-based rod extraction methods: model fitting-based rod extraction methods The double-cylindrical model is the most widely used. This type of method can extract most rod-shaped objects, but there is a high probability of missed detection for objects that are partially blocked by other objects or are too close to other objects; based on the characteristic rod-shaped object extraction method Rods are extracted from point cloud data by analyzing the local geometric features of points in the point cloud. Shape features such as height and projected area and contextual features are commonly used features. This method can well realize rods. The detection of rod-shaped objects, but more geometric feature constraints mean the increase of parameters, and the universality of the method needs to be improved; 3) The learning-based method is currently used more and more in rod-shaped object extraction, and has a high completeness and correctness of

However, the learning-based rod extraction method requires the assistance of prior knowledge and manual labeling of learning samples. There is a problem that the training data from one data set may not be suitable for the model of other data sets.

[0004] The types of objects that may exist in the urban street environment are complex and diverse, such as buildings, pedestrians, vehicles, street trees, road markings, traffic lights, etc. There may also be mutual entanglement and mutual occlusion among various objects. Automatically extracting questions gets more complicated

The robustness of most existing methods depends on good data sets and empirical parameter control after a large number of tests, and is not universal

Images