Indoor environment vertical rod object automatic detection method, system and device based on three-dimensional point cloud and storage medium

Abstract

The invention belongs to the technical field of three-dimensional image processing, and particularly provides an indoor environment vertical rod object automatic detection method, system and device based on three-dimensional point clouds and a storage medium, and the method comprises the steps: carrying out the preprocessing of collected indoor three-dimensional point cloud data, carrying out the layering processing of the preprocessed point cloud data, and obtaining a three-dimensional image; projecting each layered point cloud slice to an xoy plane to generate a corresponding BEV image, performing circle detection on the BEV image of each layer by adopting Hough transform based on the known diameter of the vertical rod object, determining the circle center position of a circle in each BEV image, fitting the circle center point set by adopting a linear fitting method, determining the central axis parameter of the vertical rod object, and determining the center axis of the vertical rod object. And the position and the shape of the vertical rod object in the three-dimensional space are further obtained. According to the method, the geometrical characteristics and known diameter information of the vertical rod object are fully utilized, so that an efficient and reliable recognition effect is achieved.

Description

Technical Field

[0001] This invention belongs to the field of three-dimensional image processing technology, specifically, it relates to an automatic detection method, system, device and storage medium for vertical pole objects in indoor environments based on three-dimensional point clouds. Background Technology

[0002] In indoor environments, there are many objects of different shapes and sizes, among which cylindrical vertical objects are a common structural element. These vertical objects typically have a known diameter and may serve as supports, partitions, or other functions within the indoor space. They are used in many applications, such as facility maintenance, robot navigation, and operation. The ability to quickly and accurately identify these vertical objects is crucial for subsequent operations or tasks.

[0003] However, indoor environments are typically complex and contain numerous interfering factors. For example, indoor spaces may contain various parts and other structural elements with varying shapes, sizes, and materials, which can affect the acquisition and analysis of point cloud data. Furthermore, indoor environments may present insufficient lighting, reflective surfaces, or occlusions, all of which degrade the quality of 3D point cloud data, making the features of vertical objects less distinct. In this complex context, traditional clustering methods often struggle to effectively identify vertical objects. These methods may be affected by noise points, interference from other objects, or the incompleteness and sparsity of the point cloud data, making it difficult to accurately distinguish the target object from the background, thus significantly increasing the difficulty of identification.

[0004] Therefore, in complex indoor environments, developing a method that can automatically and accurately identify vertical objects from 3D point clouds is of great significance for improving work efficiency and ensuring the successful completion of tasks. This method needs to effectively cope with various interference factors in the environment, while fully utilizing the geometric features and known diameter information of the vertical objects to achieve efficient and reliable identification results. Summary of the Invention

[0005] The technical problem to be solved by the present invention is to provide an automatic detection method, system, device and storage medium for vertical objects in indoor environments based on three-dimensional point clouds, which makes full use of the geometric features and known diameter information of the vertical objects to achieve efficient and reliable recognition results.

[0006] To solve the above-mentioned technical problems, the technical solution adopted by the present invention is: an automatic detection method for vertical pole objects in indoor environments based on three-dimensional point clouds, comprising the following steps: S1. Preprocessing of point cloud data: Preprocess the collected indoor 3D point cloud data; S2, Layered BEV Image Conversion: The preprocessed point cloud data is layered, and each layered point cloud slice is projected onto...xoy The plane generates the corresponding BEV image; S3. Hough Transform Detection of Circles: Based on the known diameter of the vertical rod object, the Hough Transform is used to detect circles in the BEV image of each layer to determine the center position of the circle in each BEV image. S4. Statistical analysis and superposition of circles: Statistical analysis and superposition of the circles detected in all BEV images to obtain the set of circle center points in three-dimensional space; S5. The set of center points is fitted using a linear fitting method to determine the central axis parameters of the vertical rod object, thereby obtaining the position and shape of the vertical rod object in three-dimensional space.

[0007] In a preferred embodiment, the preprocessing in step S1 includes point cloud accumulation, downsampling, noise reduction, and removal of ground and ceiling point clouds.

[0008] In a preferred embodiment, in step S1, the accumulation of point cloud is achieved by using a lidar odometry to accumulate sparse point cloud data collected at different times into a dense point cloud.

[0009] In the preferred embodiment, in step S1, the accumulated dense point cloud is downsampled using a voxel filtering method, and the point cloud is divided into voxel grids according to a preset resolution, with one representative point retained in each voxel grid.

[0010] In the preferred embodiment, in step S1, a statistical outlier removal method is used to remove noise points in the point cloud. The method determines whether a point is an outlier by calculating the local density of each point and the number of points in its neighborhood.

[0011] In a preferred embodiment, the method for removing point clouds from the ground and ceiling in step S1 is as follows: Set point cloud z The minimum and maximum values ​​in the direction are respectively and Then the threshold ranges for the ground and ceiling are defined as follows: and ,in and The point cloud is removed based on the thresholds set according to the actual environment, and the two threshold ranges are excluded.

[0012] In a preferred embodiment, step S2, specifically the layering process for the preprocessed point cloud data, involves dividing the point cloud into layers. z In the direction according to the height interval Δ z Divided into N There are several horizontal layers, and the point cloud data is located in... z The total height range in the direction is Each horizontal layer i ( i=1,2,…, N The height range is .

[0013] In the preferred embodiment, in step S2, each layered point cloud slice is processed... xoy The process of generating a BEV image by projecting it onto a plane is as follows: Initial size is M × M Two-dimensional image ,in M Set the resolution of the image and initialize all pixel values ​​to dark, which is 0; Let each point in the point cloud data be represented as ( x , y , z ), then for the first i The point cloud slices of the layer, and the corresponding BEV images are generated in the following way: ; ; in, and For point cloud data in x and y Minimum value in direction, Δ x and Δ y They are respectively x and y The unit of resolution in the direction; if( u , v Within the image range, i.e., 1 ≤ u , v ≤ M If so, the pixel value will be set to bright, which is 255.

[0014] In the preferred embodiment, step S3, which involves performing circle detection on the BEV image of each layer using the Hough transform, includes the following steps: Given that the diameter of the vertical rod is d Determine the radius of the vertical object as ; Initial size is M × M accumulator array H All accumulator values ​​are initialized to 0. H ( a , b ) is used to record possible center positions ( a , b The number of votes; For each generated BEV image Each point in (x , y ), calculate all possible center positions ( a , b The calculation formula is: ; ; in, It is an angle parameter, and its value range is... ; For each Calculate the corresponding ( a , b ), and the accumulator H ( a , b Voting will take place soon. H ( a , b Add 1 to the value of ).

[0015] In a preferred embodiment, step S3, determining the center position of the circle in each BEV image, is done as follows: Traversing the accumulator array H Find accumulator values ​​that exceed a preset threshold. T Location ( a , b These locations are the detected center points.

[0016] In the preferred embodiment, the specific method for obtaining the set of center points in three-dimensional space in step S4 is as follows: Let the first Layer BEV image The detected center position is , No. The height corresponding to the layer BEV image is Mark all the center positions as three-dimensional coordinates By superimposing the circles in all BEV images, a set of points in three-dimensional space is obtained. P Each point represents a center position of a circle: .

[0017] In a preferred embodiment, in step S5, the central axis of the vertical rod is a straight line. L ,straight line L The expression is: ; in, It is a straight line L The point above, It is a straight line L exist xoy Plane projection and x The included angle of the axis, It is a straight line L and xoy Angle between planes t It is a straight line L The parameter value.

[0018] In the preferred embodiment, in step S5, the linear fitting method is the least squares method, which determines the parameters of line L by minimizing the sum of the squared distances from each point in the point set P to line L. t The expression for the sum of squared distances is: .

[0019] This invention also provides an automatic detection system for vertical poles in indoor environments based on three-dimensional point clouds, used to execute the above-described automatic detection method for vertical poles in indoor environments based on three-dimensional point clouds, including: The point cloud preprocessing module is used to perform preprocessing operations such as accumulation, downsampling, noise reduction, and removal of ground and ceiling point clouds on the collected indoor 3D point cloud data. The layered BEV image generation module performs layered processing on the preprocessed point cloud data, projecting each layered point cloud slice onto... xoy The plane generates the corresponding BEV image; The Hough transform circle detection module is used to perform circle detection on the BEV image of each layer based on the known diameter of the vertical rod object, and to determine the center position of the circle in each BEV image. The circle center point set construction module is used to statistically analyze and overlay the circles detected in all BEV images to obtain the circle center point set in three-dimensional space. The axis fitting and result output module is used to fit the set of center points using a linear fitting method to determine the central axis parameters of the vertical rod object, thereby obtaining the position and shape of the vertical rod object in three-dimensional space.

[0020] The present invention also provides an automatic detection device for vertical objects in indoor environments based on three-dimensional point clouds, including a processor and a memory. The memory stores a computer program, and when the processor executes the computer program, it implements the above-mentioned automatic detection method for vertical objects in indoor environments based on three-dimensional point clouds.

[0021] The present invention also provides a storage medium for automatic detection of vertical objects in indoor environments based on three-dimensional point clouds. The storage medium is a computer-readable storage medium that stores a computer program. When the computer program is executed by a processor, it implements the above-mentioned method for automatic detection of vertical objects in indoor environments based on three-dimensional point clouds.

[0022] The present invention provides an automatic detection method, system, device, and storage medium for vertical pole objects in indoor environments based on three-dimensional point clouds, which has the following beneficial effects: 1. This invention addresses the problem of automatic detection of vertical objects in complex indoor environments. It achieves recognition of vertical objects by converting point cloud layers into a bird's-eye view image, using Hough transform to detect circles, and employing statistical and superimposed methods for circles. This effectively addresses various environmental interference factors while fully utilizing the geometric features and known diameter information of the vertical object to achieve efficient and reliable recognition.

[0023] 2. By using lidar odometers to accumulate sparse point clouds into dense point clouds, the problem of insufficient point cloud information at a single moment is compensated; voxel filtering downsampling is adopted to reduce computational complexity while retaining key geometric features and avoid redundant data consuming computing power; statistical outlier removal accurately filters noise points to ensure the purity of point cloud data; ground and ceiling point clouds are specifically removed to eliminate interference from non-target areas and focus on point clouds related to vertical objects, solving the problem of complex indoor environmental backgrounds.

[0024] 3. Divide the point cloud into layers along the z-axis at height intervals. Project each layer of point cloud slices onto the z-axis. xoy The planar generation of BEV images transforms complex 3D point cloud processing into mature 2D image processing, significantly reducing algorithm complexity and computational load, improving detection efficiency, and solving the problems of time-consuming and difficult traditional direct processing of 3D point clouds.

[0025] Each BEV image corresponds to a specific height range. The three-dimensional coordinates can be directly restored by superimposing the images at the center. This simplifies the detection difficulty of a single layer and preserves the spatial distribution characteristics of the vertical pole object, avoiding the loss of height information caused by two-dimensional processing.

[0026] BEV images at different height levels detect circles independently, avoiding interference between object features at different heights and improving the accuracy of circle center detection at each height level.

[0027] 4. The radius is determined based on the known diameter of the vertical rod object. The Hough transform only detects circles with this fixed radius, without having to traverse all possible radii, reducing invalid calculations and avoiding misclassifying circular objects with non-target diameters as vertical rods.

[0028] The Hough transform uses an accumulator to vote on possible center positions, and only positions with more than a threshold of votes are identified as centers, filtering out the influence of isolated interference points. Even if the point cloud has local missing points or slight noise, it can still accurately locate the center, improving the robustness of detection.

[0029] 5. Map the center of each BEV image layer to three-dimensional coordinates to form a set of center points covering the entire height of the vertical pole. This fully preserves the center distribution characteristics of the vertical pole object along the height direction, providing data support for restoring the three-dimensional shape of the vertical pole and avoiding the limitation that single height detection cannot reflect the overall structure of the vertical pole.

[0030] By minimizing the sum of squared distances from the point set to the line, the parameters of the vertical rod's central axis can be accurately determined. This not only yields the vertical rod's planar position but also provides a complete output of its position and shape in three-dimensional space, solving the problem that traditional methods struggle to accurately obtain complete three-dimensional information about the vertical rod. Attached Figure Description

[0031] The present invention will be further described below with reference to the accompanying drawings and embodiments: Figure 1 This is a flowchart of the automatic detection method for vertical pole objects in indoor environments based on three-dimensional point clouds according to the present invention; Figure 2 This is a simulated point cloud data map of a vertical rod-shaped object in a specific embodiment; Figure 3 This is a BEV image processing diagram; Figure 4 It is a diagram of the center axis of the identified object. Detailed Implementation

[0032] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are for illustrative purposes only and are not intended to limit the invention.

[0033] Example 1: To address the problem of automatically detecting vertical poles in complex indoor environments, this invention proposes a method for vertical pole object recognition based on 3D point clouds. This method achieves vertical pole object recognition through point cloud layer-by-layer conversion into a bird's-eye view (BEV) image, Hough transform detection of circles, and statistical superposition of circles.

[0034] like Figure 1 As shown, an automatic detection method for vertical pole objects in indoor environments based on 3D point clouds includes the following detailed technical steps: S1. Preprocessing of point cloud data: Preprocess the collected indoor 3D point cloud data.

[0035] Before performing vertical pole object recognition, the acquired indoor 3D point cloud data needs to be preprocessed to improve data quality and the efficiency of subsequent processing. The preprocessing steps mainly include point cloud accumulation, downsampling, noise reduction, and removal of point clouds from the ground and ceiling.

[0036] By utilizing LiDAR odometry, sparse point cloud data collected by LiDAR at different times is accumulated into a dense point cloud. This method allows point cloud data from multiple moments to be fused together, resulting in richer environmental details. The dense point cloud more accurately reflects the geometric structure of the indoor environment, providing a more reliable data foundation for subsequent feature extraction and object recognition.

[0037] To reduce computational complexity and improve processing efficiency, the accumulated dense point cloud is downsampled. A voxel filtering method is used to divide the point cloud into a voxel grid at a certain resolution (e.g., 0.2 meters), retaining a representative point in each voxel grid. To remove noise points from the point cloud, a statistical outlier removal method is employed. Specifically, the local density of each point is calculated, and whether a point is an outlier is determined by analyzing the number of points in its neighborhood.

[0038] Using elevation information from the environment, point clouds representing the ground and ceiling are removed. Assuming the ground and ceiling are in... z Points exhibit a relatively flat distribution along the z-axis, and these points can be identified and removed by analyzing the distribution of the point cloud along the z-axis. Specifically, suppose the point cloud has a relatively flat distribution along the z-axis. z The minimum and maximum values ​​in the direction are respectively and Then the threshold ranges for the ground and ceiling can be defined as follows: and ,in and These thresholds are set based on the actual environment. By removing points within these ranges, the influence of the ground and ceiling can be eliminated, retaining the point cloud data related to the vertical objects.

[0039] S2, Layered BEV Image Conversion: The preprocessed point cloud data is layered, and each layered point cloud slice is projected onto... xoy The plane generates the corresponding BEV image.

[0040] First, the collected indoor 3D point cloud data is processed into layers. Specifically, the point cloud is divided into layers... z In the direction, according to a certain height interval Δ z The data is divided into several horizontal layers, each containing point cloud data within a certain height range. Let the point cloud data be... z The total height range in the direction is Point clouds can be divided into N Each horizontal layer. i ( i =1,2,…, N The height range is In this way, complex 3D point cloud data can be decomposed into slices of multiple 2D planes, which facilitates subsequent processing.

[0041] Next, each layered point cloud slice will be processed. xoy Projecting a planar image to generate a bird's-eye view (BEV). First, initialize a two-dimensional image. Its size is M × M ,in M Let the resolution of the image be [resolution], and initialize all pixel values ​​to dark, which is 0. Let each point in the point cloud data be represented as (...). x , y , z ), then for the first i The point cloud slices of a layer, and their corresponding BEV images, can be generated in the following way: ; ; in, and For point cloud data in x and y Minimum value in direction, Δ x and Δ y They are respectively x and y The resolution unit in the direction. If ( u , v Within the image range (i.e., 1≤ u , v ≤ M If the pixel value is not specified, then the pixel value is set to bright, which is 255. Through the above steps, complex 3D point cloud data can be converted into multiple 2D images, each corresponding to a point cloud slice at a specific height level. In the BEV image, the point cloud data is mapped onto a 2D plane, where pixel values ​​corresponding to points are bright, and pixel values ​​corresponding to areas without points are dark. This 2D image representation not only simplifies the data structure but also facilitates subsequent image processing and analysis.

[0042] S3. Hough Transform for Circle Detection: Based on the known diameter of the vertical rod, the Hough transform is used to detect circles in the BEV image of each layer, and the center position of the circle in each BEV image is determined.

[0043] After completing point cloud layering and generating BEV images, this invention further utilizes the Hough transform to analyze the BEV images of each layer to detect circular features. Since vertical objects appear as circles in BEV images, and their diameters are known, the position of the circle in each BEV image can be accurately located using the Hough transform. The Hough transform is a classic circle detection algorithm that can effectively extract circular features from images, and is particularly suitable for detecting circles with known diameters.

[0044] For circle detection, the core idea of ​​the Hough transform is to vote on each edge point in the image within the parameter space, and finally determine the center and radius of the circle using the value of the accumulator. For a known diameter... d A circle, its radius r for: ; Let the center of the circle be ( a , b If ), then the equation of the circle is: ; In the Hough transform, each edge point ( x , y ) will be in the parameter space ( a , b Voting will be conducted within the given radius. r For a circle, the parameter space can be simplified to a two-dimensional space. a , b ),in a and b Representing the center of the circle x and y The coordinates.

[0045] For each generated BEV image Perform the following steps: Initialize an accumulator array H Its size is M × M (Same resolution as the BEV image), and initialize all accumulator values ​​to 0. Accumulator H ( a , b ) is used to record the position of each possible center of the circle. a , b The number of votes. Each point in ( x , y ), calculate all possible center positions ( a , b Due to the radius of the circle r It is known that the position of the center of the circle can be calculated using the following formula: ; ; in, It is an angle parameter, and its value range is... For each Calculate the corresponding ( a , b ), and the accumulator H ( a , b Voting will take place soon. H ( a , b Add 1 to the value of ).

[0046] Traversing the accumulator array H Find accumulator values ​​that exceed a certain threshold. T Location ( a , b These positions are the detected center points of the circles. The object corresponding to the largest parameter in the accumulator array is the parameter of the circle to be detected.

[0047] For each BEV image The center position detected through the above steps ( a , b ) and known radius r This allows us to determine the position of the circles in each BEV image. By statistically analyzing and overlaying the circles in all BEV images, we can determine the position and shape of the vertical pole object in three-dimensional space. Specifically, by analyzing the distribution of circles at different height levels, we can reconstruct the three-dimensional structure of the vertical pole object.

[0048] S4. Statistical analysis and superposition of circles: Statistical analysis and superposition of the circles detected in all BEV images to obtain the set of circle center points in three-dimensional space.

[0049] After detecting circles in each BEV image, this invention further statistically analyzes and superimposes the circles in all BEV images to determine the position and shape of the vertical rod object in three-dimensional space. The specific steps are as follows: For each BEV image Let the detected center position be... Given that the radius of the circle is... r The circles in all BEV images are superimposed, that is, the center position of each circle is superimposed in three-dimensional space. Mark, where For the first i The height corresponding to the layer BEV image. In this way, a set of points in three-dimensional space can be obtained. P Each point represents a center position of a circle: .

[0050] By analyzing point sets P The distribution of points can reconstruct the three-dimensional structure of the vertical object. Specifically, since the vertical object appears as a series of concentric circles in BEV images at different height levels, the centers of these circles should be distributed along the axis of the vertical object in three-dimensional space. Let the axis of the vertical object be a straight line. L Then the straight line L It can be represented as: ; in, It is a straight line L The point above, It is a straight line L exist xoy Plane projection and x The included angle of the axis, It is a straight line L and xoy Angle between planes t It is a straight line L The parameter value.

[0051] S5. The set of center points is fitted using a linear fitting method to determine the central axis parameters of the vertical rod object, thereby obtaining the position and shape of the vertical rod object in three-dimensional space.

[0052] To determine the straight line L The parameters can be obtained using linear fitting methods such as the least squares method for the point set. P Perform a fitting. Specifically, minimize the set of points. P From each point in the line L The sum of the squares of the distances, i.e.: .

[0053] Through the above fitting, the straight line of the axis of the vertical rod can be obtained. L The parameters. Based on the straight line. L The parameters determine the position and shape of the vertical rod in three-dimensional space. Specifically, the central axis of the vertical rod is a straight line. L The height is .

[0054] Example 2: To verify the feasibility of this invention, scientific demonstration was conducted through simulation experiments.

[0055] The experimental parameters were set as follows: In the ROS (Robot Operating System) system, C++ program was used to generate three-dimensional simulated point cloud data of indoor vertical pole objects. The computer configuration was: I7-12500, 16G of RAM.

[0056] like Figure 2 The figure shown is a simulated point cloud data diagram of a vertical pole-shaped object in a specific embodiment. The diagram shows the simulated data after preprocessing, that is, the data accumulation and filtering have been completed, and the point clouds of the ground and ceiling have been removed, while the data of the outer surface of the vertical pole object is retained.

[0057] like Figure 3 The image shown is a processing diagram based on a bird's-eye view (BEV) view. Figure 2 The simulated point cloud data in the figure is processed by the S2 layered BEV image transformation method to divide the point cloud data into layers. In the simulation experiment, the data was divided into 20 layers, and the S3 Hough transform was used to detect the circles in each layer. The blue plane in the figure is the circle detected in each layer, and the circular outline can be clearly presented in each layer.

[0058] like Figure 4 The image shown is a diagram of the center axis of the identified object. S4 will... Figure 3 The center of all circles is detected as a point on a straight line in space, resulting in a set of center points in three-dimensional space. The center of the circles is then fitted using the straight line fitting method in S5 to generate the central axis of the vertical rod object. The green line segment in the figure is the fitted central axis, and the axis is completely consistent with the actual spatial posture of the vertical rod, thus allowing the calculation of relevant parameters of the vertical rod object.

[0059] This embodiment can completely reconstruct the position and shape of a vertical object in three-dimensional space, not only locating its planar coordinates but also accurately estimating its spatial orientation. It successfully completed the layered processing of 20 point clouds, circle detection, and axis fitting, demonstrating the automatic, accurate, and robust detection capability of the method of this invention for vertical objects in complex indoor scenes.

[0060] Example 3: This embodiment provides an automatic detection system for vertical poles in indoor environments based on 3D point clouds, used to execute the automatic detection method for vertical poles in indoor environments based on 3D point clouds described in Embodiment 1, including: The point cloud preprocessing module is used to perform preprocessing operations such as accumulation, downsampling, noise reduction, and removal of ground and ceiling point clouds on the collected indoor 3D point cloud data. The layered BEV image generation module performs layered processing on the preprocessed point cloud data, projecting each layered point cloud slice onto... xoy The plane generates the corresponding BEV image; The Hough transform circle detection module is used to perform circle detection on the BEV image of each layer based on the known diameter of the vertical rod object, and to determine the center position of the circle in each BEV image. The circle center point set construction module is used to statistically analyze and overlay the circles detected in all BEV images to obtain the circle center point set in three-dimensional space. The axis fitting and result output module is used to fit the set of center points using a linear fitting method to determine the central axis parameters of the vertical rod object, thereby obtaining the position and shape of the vertical rod object in three-dimensional space.

[0061] Example 4: This embodiment provides an electronic device for automatic detection of vertical objects in indoor environments based on three-dimensional point clouds, including a processor and a memory. The memory stores a computer program, and when the processor executes the computer program, it implements the automatic detection method for vertical objects in indoor environments based on three-dimensional point clouds described in Embodiment 1.

[0062] Example 5: This embodiment provides a storage medium for automatic detection of vertical objects in indoor environments based on three-dimensional point clouds. The storage medium is a computer-readable storage medium that stores a computer program. When the computer program is executed by a processor, it implements the automatic detection method for vertical objects in indoor environments based on three-dimensional point clouds described in Embodiment 1.

[0063] Those skilled in the art will readily understand that the above description is merely a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

Claims

1. An automatic detection method for vertical pole objects in indoor environments based on 3D point clouds, characterized in that, Includes the following steps: S1. Preprocessing of point cloud data: Preprocess the collected indoor 3D point cloud data; S2, Layered BEV Image Conversion: The preprocessed point cloud data is layered, and each layered point cloud slice is projected onto... xoy The plane generates the corresponding BEV image; S3. Hough Transform Detection of Circles: Based on the known diameter of the vertical rod object, the Hough Transform is used to detect circles in the BEV image of each layer to determine the center position of the circle in each BEV image. S4. Statistical analysis and superposition of circles: Statistical analysis and superposition of the circles detected in all BEV images to obtain the set of circle center points in three-dimensional space; S5. The set of center points is fitted using a linear fitting method to determine the central axis parameters of the vertical rod object, thereby obtaining the position and shape of the vertical rod object in three-dimensional space.

2. The automatic detection method for vertical pole objects in indoor environments based on three-dimensional point clouds according to claim 1, characterized in that, The preprocessing in step S1 includes point cloud accumulation, downsampling, noise reduction, and removal of ground and ceiling point clouds.

3. The method for automatic detection of vertical poles in indoor environments based on three-dimensional point clouds according to claim 2, characterized in that, In step S1, the accumulation of point cloud is achieved by using a lidar odometry to accumulate sparse point cloud data collected at different times into a dense point cloud.

4. The automatic detection method for vertical pole objects in indoor environments based on three-dimensional point clouds according to claim 2, characterized in that, In step S1, the accumulated dense point cloud is downsampled using a voxel filtering method, and the point cloud is divided into voxel grids according to a preset resolution, with one representative point retained in each voxel grid.

5. The automatic detection method for vertical pole objects in indoor environments based on three-dimensional point clouds according to claim 2, characterized in that, In step S1, a statistical outlier removal method is used to remove noise points in the point cloud. The method determines whether a point is an outlier by calculating the local density of each point and the number of points in its neighborhood.

6. The automatic detection method for vertical pole objects in indoor environments based on three-dimensional point clouds according to claim 2, characterized in that, In step S1, the method for removing point clouds from the ground and ceiling is as follows: Set point cloud z The minimum and maximum values ​​in the direction are respectively and Then the threshold ranges for the ground and ceiling are defined as follows: and ,in and The point cloud is removed based on the thresholds set according to the actual environment, and the two threshold ranges are excluded.

7. The automatic detection method for vertical pole objects in indoor environments based on three-dimensional point clouds according to claim 1, characterized in that, In step S2, the specific method for performing layered processing on the preprocessed point cloud data is as follows: the point cloud is divided into layers... z In the direction according to the height interval Δ z Divided into N There are several horizontal layers, and the point cloud data is located in... z The total height range in the direction is Each horizontal layer i ( i =1,2,…, N The height range is .

8. The method for automatic detection of vertical poles in indoor environments based on three-dimensional point clouds according to claim 1, characterized in that, In step S2, each layered point cloud slice is processed... xoy The process of generating a BEV image by projecting it onto a plane is as follows: Initial size is M × M Two-dimensional image ,in M Set the resolution of the image and initialize all pixel values ​​to dark, which is 0; Let each point in the point cloud data be represented as ( x , y , z ), then for the first i The point cloud slices of the layer, and the corresponding BEV images are generated in the following way: ； ； in, and For point cloud data in x and y Minimum value in direction, Δ x and Δ y They are respectively x and y The unit of resolution in the direction; if( u , v Within the image range, i.e., 1 ≤ u , v ≤ M If so, the pixel value will be set to bright, which is 255.

9. The automatic detection method for vertical pole objects in indoor environments based on three-dimensional point clouds according to claim 1, characterized in that, In step S3, the specific process of performing circle detection on the BEV image of each layer using Hough transform includes: Given that the diameter of the vertical rod is d Determine the radius of the vertical object as ; Initial size is M × M accumulator array H All accumulator values ​​are initialized to 0. H ( a , b ) is used to record possible center positions ( a , b The number of votes; For each generated BEV image Each point in ( x , y ), calculate all possible center positions ( a , b The calculation formula is: ； ； in, It is an angle parameter, and its value range is... ; For each Calculate the corresponding ( a , b ), and the accumulator H ( a , b Voting will take place soon. H ( a , b Add 1 to the value of ).

10. The automatic detection method for vertical pole objects in indoor environments based on three-dimensional point clouds according to claim 9, characterized in that, In step S3, the method for determining the center position of the circle in each BEV image is as follows: Traversing the accumulator array H Find accumulator values ​​that exceed a preset threshold. T Location ( a , b These positions are the detected center of the circle.

11. The automatic detection method for vertical pole objects in indoor environments based on three-dimensional point clouds according to claim 1, characterized in that, In step S4, the specific method for obtaining the set of center points in three-dimensional space is as follows: Let the first Layer BEV image The detected center position is , No. The height corresponding to the layer BEV image is Mark all the center positions as three-dimensional coordinates By superimposing the circles in all BEV images, a set of points in three-dimensional space is obtained. P Each point represents a center position of a circle: 。 12. The method for automatic detection of vertical pole objects in indoor environments based on three-dimensional point clouds according to claim 1, characterized in that, In step S5, the central axis of the vertical rod is a straight line. L ,straight line L The expression is: ； in, It is a straight line L The point above, It is a straight line L exist xoy Plane projection and x The included angle of the axis, It is a straight line L and xoy Angle between planes t It is a straight line L The parameter value.

13. The method for automatic detection of vertical pole objects in indoor environments based on three-dimensional point clouds according to claim 1, characterized in that, In step S5, the linear fitting method is the least squares method, which determines the parameters of line L by minimizing the sum of the squared distances from each point in the point set P to line L. t The expression for the sum of squared distances is: 。 14. An automatic detection system for vertical pole objects in indoor environments based on three-dimensional point clouds, characterized in that, The method for automatic detection of vertical pole objects in indoor environments based on three-dimensional point clouds, as described in any one of claims 1 to 13, includes: The point cloud preprocessing module is used to perform preprocessing operations such as accumulation, downsampling, noise reduction, and removal of ground and ceiling point clouds on the collected indoor 3D point cloud data. The layered BEV image generation module performs layered processing on the preprocessed point cloud data, projecting each layered point cloud slice onto... xoy The plane generates the corresponding BEV image; The Hough transform circle detection module is used to perform circle detection on the BEV image of each layer based on the known diameter of the vertical rod object, and to determine the center position of the circle in each BEV image. The circle center point set construction module is used to statistically analyze and overlay the circles detected in all BEV images to obtain the circle center point set in three-dimensional space. The axis fitting and result output module is used to fit the set of center points using a linear fitting method to determine the central axis parameters of the vertical rod object, thereby obtaining the position and shape of the vertical rod object in three-dimensional space.

15. An electronic device for automatic detection of vertical objects in indoor environments based on three-dimensional point clouds, characterized in that, The device includes a processor and a memory, the memory storing a computer program, and the processor executing the computer program to implement the automatic detection method for vertical pole objects in indoor environments based on three-dimensional point clouds as described in any one of claims 1 to 13.

16. A storage medium for automatic detection of vertical objects in indoor environments based on three-dimensional point clouds, characterized in that, The storage medium is a computer-readable storage medium storing a computer program. When the computer program is executed by a processor, it implements the automatic detection method for vertical pole objects in indoor environments based on three-dimensional point clouds as described in any one of claims 1 to 13.

Images