A method for extracting cylindrical objects from 3D point clouds and automatically generating vectors

Abstract

The present invention belongs to the technical field of point cloud data processing, and discloses a method for extracting cylindrical objects from three-dimensional point clouds and automatically generating vectors. First, a radar scan is performed to obtain a cylindrical point cloud of a traffic light pole. Then, a Gaussian mixture model is used to perform binary classification on the cylindrical point cloud, and the points with a smaller number of categories are noise points, and the noise points are removed. The remaining points are then divided into equal distances on the Z axis and projected onto a two-dimensional plane, and circular fitting is performed on the projection points to obtain a set of circle center coordinates. Then, outlier detection is performed on the obtained circle center point set according to the local outlier factor detection method to remove abnormal circle center points. Finally, three-dimensional straight line fitting is performed on the remaining circle center points to obtain the final cylindrical object vector. The present invention makes full use of the image features of the three-dimensional point cloud projection map, and realizes the automatic extraction of the central axis of the three-dimensional point cloud cylindrical object through a variety of machine learning fitting algorithms, and automatically generates the cylindrical object vector required for the high-precision map, saving time and labor costs.

Description

Technical Field

[0001] The present invention belongs to the technical field of point cloud data processing, and in particular relates to a method for extracting cylindrical objects from three-dimensional point clouds and automatically generating vectors. Background Art

[0002] Three-dimensional high-precision maps are widely recognized by industry and academia as the primary development direction for next-generation digital maps. They are a prerequisite for autonomous and assisted driving, providing the primary basis for precise positioning and accurate decision-making by autonomous vehicles. Assisted parking is a key component of assisted driving, and the construction of high-precision three-dimensional road maps relies on extracting vector information, such as cylindrical objects, from road point cloud data.

[0003] Traditional manual extraction of cylindrical object vector information is inefficient and its accuracy is affected by the worker's status. How to remove the influence of other objects from complex point cloud information and extract accurate cylindrical object vectors to save time and labor costs is a problem of great academic and engineering value. Summary of the Invention

[0004] To solve the problems mentioned in the background technology, the present invention proposes a method for extracting cylindrical objects from three-dimensional point clouds and automatically vectoring them, which adopts the following technical solutions:

[0005] A method for extracting cylindrical objects from a three-dimensional point cloud and automatically generating vectors includes the following steps:

[0006] Step 1: 3D point cloud Gaussian mixture model denoising: Radar scanning obtains a cylindrical point cloud of the traffic light pole. A Gaussian mixture model (GMM) is used to classify the cylindrical point cloud into two categories. Points with a small number of categories are considered noise points, and noise points outside the main body of the column are removed.

[0007] Step 2: Projection of the point cloud top view: Divide the remaining points into equal distances on the Z axis, project the points in each equal distance block onto a two-dimensional plane, and use a circular fitting algorithm to perform circular fitting on the two-dimensional plane projection points to obtain the set of fitting circle centers;

[0008] Step 3: Outlier detection of the circle center point set: perform outlier detection on the obtained circle center point set according to the local outlier factor detection method to remove abnormal circle center points;

[0009] Step 4: Generate three-dimensional cylindrical object vector by linear fitting: Project the remaining center point set into three dimensions, and use a three-dimensional linear fitting algorithm to perform linear fitting on the three-dimensional coordinate set of the center points to obtain the final cylindrical object vector.

[0010] Furthermore, the step 1 includes the following specific steps:

[0011] Radar scanning obtains a cylindrical point cloud of a traffic light pole. This cylindrical point cloud is then classified into two categories using a Gaussian mixture model (GMM). The Gaussian mixture model consists of two probability distribution models, and the maximum number of iterations is set to 100. Point clouds with a smaller number of categories deviate from the main point cloud of the cylindrical object, while point clouds with a larger number of categories represent the main point cloud of the cylindrical object.

[0012] Furthermore, the step 2 includes the following specific steps:

[0013] The main point cloud of the cylindrical object is divided into 20 sub-point clouds with equal distance on the Z axis, and the point clouds of the sub-point clouds are projected from a top view onto a two-dimensional plane (X, Y). The average Z-axis coordinate of the sub-point clouds is recorded as the Z-axis coordinate of the projection point set, and the projection points are fitted with a circle using the least squares method to obtain the center point set of the sub-point clouds.

[0014] Furthermore, the step 3 includes the following specific steps:

[0015] The local outlier factor detection method uses K-nearest neighbor distance, with the K value set to 3 and the distance using Euclidean distance to determine the center point of the abnormal circle, and the center point offset of the cylindrical object is controlled between 10 and 30 cm.

[0016] Furthermore, step 4 includes the following specific steps:

[0017] The center point set after removing abnormal center points is combined with the Z-axis average value in the sub-point cloud to obtain the center point set of the cylindrical object. The least squares method is used to fit the center point set to obtain the XZ plane and YZ plane. The intersection of the XZ plane and the YZ plane is the central axis of the cylindrical object, which is the cylindrical object vector.

[0018] Furthermore, the cylindrical object vector is represented as:

[0019] Formula 1

[0020] In Formula 1, a and b are the slope and intercept of the straight line in the XZ plane, and c and d are the slope and intercept of the straight line in the YZ plane.

[0021] The method for extracting and automatically generating vectors of three-dimensional point cloud cylindrical objects of the present invention has the following advantages: the present invention makes full use of the image features of the three-dimensional point cloud projection image, and realizes the automatic extraction of the central axis of the three-dimensional point cloud cylindrical objects through a variety of machine learning fitting algorithms, and automatically generates the cylindrical object vectors required for high-precision maps, saving time and labor costs. BRIEF DESCRIPTION OF THE DRAWINGS

[0022] Figure 1 This is a flow chart of the method for extracting cylindrical objects from three-dimensional point clouds and automatically generating vectors according to the present invention;

[0023] Figure 2 Schematic diagram of denoising processing of 3D point cloud Gaussian mixture model;

[0024] Figure 3 It is a schematic diagram of circular fitting of projection points;

[0025] Figure 4 Schematic diagram of circle center outlier detection;

[0026] Figure 5 Schematic diagram of three-dimensional straight line fitting. DETAILED DESCRIPTION

[0027] The present invention is further described in detail below with reference to the accompanying drawings and examples.

[0028] In order to verify the effectiveness of this method, we collected 3D point cloud data of a road in Deqing County, Huzhou City, Zhejiang Province. The flowchart of the whole set of 3D point cloud cylindrical object extraction and vector automatic generation method is as follows: Figure 1 shown.

[0029] Step 1: 3D point cloud Gaussian mixture model denoising.

[0030] Radar scanning obtains a cylindrical point cloud of a traffic light pole. This cylindrical point cloud is then classified using a Gaussian mixture model (GMM), which consists of two probability distribution models and has a maximum number of iterations of 100. The Gaussian mixture model predicts that point clouds with a small number of categories deviate from the main cylindrical object, achieving noise reduction. Point clouds with a large number of categories represent the main cylindrical point cloud.

[0031] Step 2: Project the point cloud from top view.

[0032] The main point cloud of the cylindrical object is divided into 20 sub-point clouds at equal distances on the Z axis, and the point clouds of the sub-point clouds are projected from a top view onto a two-dimensional plane (X, Y). The average Z-axis coordinate of the sub-point clouds is recorded as the Z-axis coordinate of the projected point set, and the least squares method is used to perform circular fitting on the projected points to obtain the center point set of the sub-point clouds.

[0033] Step 3: Detect outliers in the circle center point set.

[0034] The local outlier factor detection method uses K-nearest neighbor distance, with the K value set to 3 and the distance using Euclidean distance to determine the center point of the abnormal circle, and the center point offset of the cylindrical object is controlled between 10 and 30 cm.

[0035] Step 4: Generate vector straight line fitting of three-dimensional cylindrical objects.

[0036] The center point set after removing abnormal center points is combined with the Z-axis average value in the sub-point cloud to obtain the center point set of the cylindrical object, as shown in Formula 1. The least squares method is used to fit the center point set to obtain the XZ plane and YZ plane. The intersection of the XZ plane and the YZ plane is the central axis of the cylindrical object, which is the cylindrical object vector.

[0037] Formula 1

[0038] In Formula 1, a and b are the slope and intercept of the straight line in the XZ plane, and c and d are the slope and intercept of the straight line in the YZ plane.

[0039] Experimental process

[0040] like Figure 2 As shown in the figure, after the input point cloud is denoised by the Gaussian mixture model, the appendages that do not belong to the cylindrical object are removed as noise due to the small number of points. After the cylindrical object is divided into equal distances, the point cloud in the divided space is fitted with the center of the circle. The fitting process is as follows: Figure 3 As shown. The center point set after fitting is removed after outliers are removed, as shown Figure 4 As shown in the figure, the center of the circle that deviates too much from the center of most circles is removed to ensure that the central axis of the cylinder is within a reasonable range of variation. The center point set with outliers removed is fitted with straight lines on the XZ plane and the YZ plane respectively. The intersection of the XZ plane and the YZ plane is the desired central axis. Figure 5 This is the final fitting result.

[0041] Experimental results show that the automatic extraction of cylindrical objects from three-dimensional point clouds by this method is highly consistent with the automatic generation of vectors.

[0042] The above embodiments are used to illustrate the present invention rather than to limit the present invention. Any modifications and changes made to the present invention within the spirit of the present invention and the protection scope of the claims shall fall within the protection scope of the present invention.

Claims

1. A method for extracting cylindrical objects from a three-dimensional point cloud and automatically generating vectors, characterized in that: The following steps are included: Step 1: 3D point cloud Gaussian mixture model denoising: Radar scanning obtains a cylindrical point cloud of a traffic light pole. The Gaussian mixture model (GMM) is used to classify the cylindrical point cloud into two categories. Points with a small number of categories are considered noise points, and noise points outside the main body of the column are removed. Step 2: Projection of the point cloud top view: Divide the remaining points into equal distances on the Z axis, project the points in each equal distance block onto a two-dimensional plane, and use a circular fitting algorithm to perform circular fitting on the two-dimensional plane projection points to obtain the set of fitting circle centers; Step 3: Outlier detection of the circle center point set: perform outlier detection on the obtained circle center point set according to the local outlier factor detection method to remove abnormal circle center points; Step 4: Generate three-dimensional cylindrical object vector by linear fitting: Project the remaining center point set into three dimensions, and use a three-dimensional linear fitting algorithm to perform linear fitting on the three-dimensional coordinate set of the center points to obtain the final cylindrical object vector.

2. The method for extracting cylindrical objects from three-dimensional point clouds and automatically generating vectors according to claim 1, characterized in that: The step 1 includes the following specific steps: Radar scanning obtains a cylindrical point cloud of a traffic light pole. This cylindrical point cloud is then classified into two categories using a Gaussian mixture model (GMM). The Gaussian mixture model consists of two probability distribution models, and the maximum number of iterations is set to 100. Point clouds with a smaller number of categories deviate from the main point cloud of the cylindrical object, while point clouds with a larger number of categories represent the main point cloud of the cylindrical object.

3. The method for extracting cylindrical objects from three-dimensional point clouds and automatically generating vectors according to claim 1, characterized in that: The step 2 includes the following specific steps: The main point cloud of the cylindrical object is divided into 20 sub-point clouds with equal distance on the Z axis, and the point clouds of the sub-point clouds are projected from a top view onto a two-dimensional plane (X, Y). The average Z-axis coordinate of the sub-point clouds is recorded as the Z-axis coordinate of the projection point set, and the projection points are fitted with a circle using the least squares method to obtain the center point set of the sub-point clouds.

4. The method for extracting cylindrical objects from three-dimensional point clouds and automatically generating vectors according to claim 1, characterized in that: The step 3 includes the following specific steps: The local outlier factor detection method uses K-nearest neighbor distance, with the K value set to 3 and the distance using Euclidean distance to determine the center point of the abnormal circle, and the center point offset of the cylindrical object is controlled between 10 and 30 cm.

5. The method for extracting cylindrical objects from three-dimensional point clouds and automatically generating vectors according to claim 1, characterized in that: The step 4 includes the following specific steps: The center point set after removing abnormal center points is combined with the Z-axis average value in the sub-point cloud to obtain the center point set of the cylindrical object. The least squares method is used to fit the center point set to obtain the XZ plane and YZ plane. The intersection of the XZ plane and the YZ plane is the central axis of the cylindrical object, which is the cylindrical object vector.

6. The method for extracting cylindrical objects from three-dimensional point clouds and automatically generating vectors according to claim 5, characterized in that: The cylindrical object vector is represented as: Formula 1 In Formula 1, a and b are the slope and intercept of the straight line in the XZ plane, and c and d are the slope and intercept of the straight line in the YZ plane.

Images