Method and device for extracting vehicle contour features based on point cloud bird's eye view, and medium

Abstract

The application belongs to the technical field of image processing, and provides a method and device for extracting vehicle contour features based on a point cloud bird's-eye view and a medium, the method comprising: collecting point cloud data and projecting the point cloud data into a bird's-eye view; obtaining a point cloud cluster of a target vehicle through a clustering method; selecting multiple preliminary fitting angles, each preliminary fitting angle corresponding to a minimum rectangular bounding box of a point cloud cluster; calculating statistical variance values of the point cloud cluster and each rectangular bounding box; selecting the highest value from all statistical variance values as an optimal statistical variance value; taking the preliminary fitting angle corresponding to the optimal statistical variance value as a determined fitting angle, and taking the rectangular bounding box corresponding to the determined fitting angle as a contour feature of the target vehicle. The application is used to solve the problem that the minimum rectangular bounding box obtained is easily affected by the point cloud cluster of a noise point or a convex object and deviates, thereby failing to represent the directionality of most point clouds.

Description

TECHNICAL FIELD

[0001] The application belongs to the technical field of image processing, and particularly relates to a method and device for extracting vehicle contour features based on a point cloud bird's-eye view and a medium. BACKGROUND

[0002] In an automatic driving task, the identification and tracking of a target vehicle based on vehicle point cloud is an important perception task. Vehicle point cloud is a set of discrete points, and in actual processing, the point cloud determined as a target vehicle often needs to be fitted to obtain a fitted vehicle contour, so as to determine the orientation and size of the vehicle according to the fitted vehicle, and then complete the tracking task. The methods for extracting vehicle contour features from a single frame of point cloud in the industry are mostly PCA principal component analysis method or AABB axisymmetric bounding box and minimum area rectangular box method, which do not match the actual vehicle point cloud features.

[0003] Chinese invention patent

Publication No. CN111316289A

[0004] However, for point cloud clusters with many noise points or containing convex objects such as target rearview mirrors, the minimum rectangular bounding box fitted is easily affected by noise points or convex object point cloud clusters and deviates, so as to be unable to represent the directionality of most point clouds. SUMMARY

[0005] The purpose of the present application is to provide a method and device for extracting vehicle contour features based on a point cloud bird's-eye view, which solves the problem that the minimum rectangular bounding box fitted is easily affected by noise points or convex object point cloud clusters and deviates, so as to be unable to represent the directionality of most point clouds.

[0006] To achieve the above technical purpose, the technical scheme adopted by the present application is as follows:

[0007] In a first aspect, the present application provides a method for extracting vehicle contour features based on a point cloud bird's-eye view, which comprises:

[0008] Collecting point cloud data and projecting the point cloud data into a bird's-eye view;

[0009] Obtaining a point cloud cluster of a target vehicle through a clustering method;

[0010] Selecting a plurality of preliminary fitting angles, each of which corresponds to determining a minimum rectangular bounding box of a point cloud cluster;

[0011] calculating a statistical variance value of the point cloud cluster and each of the rectangular bounding boxes;

[0012] selecting the highest value from all the statistical variance values as an optimal statistical variance value;

[0013] taking the preliminary fitting angle corresponding to the optimal statistical variance value as a determined fitting angle, and taking the rectangular bounding box corresponding to the determined fitting angle as a rectangular bounding box for extracting the contour feature of the target vehicle.

[0014] With reference to the first aspect, before the point cloud data is projected into the bird's eye view, the method further includes ground removal, in which point clouds belonging to the ground are removed by plane fitting.

[0015] With reference to the first aspect, the selecting multiple preliminary fitting angles includes selecting multiple preliminary fitting angles from a preset angle range at a preset step size.

[0016] With reference to the first aspect, the selecting multiple preliminary fitting angles from a preset angle range at a preset step size includes taking the origin of the ego vehicle coordinate system as a rotation point, taking the forward direction of the ego vehicle as the X axis, taking the X axis as the initial direction, and selecting multiple preliminary fitting angles from the preset angle range at the preset step size.

[0017] With reference to the first aspect, the determining a minimum rectangular bounding box of a point cloud cluster corresponding to each of the preliminary fitting angles includes:

[0018] According to the preliminary fitting angle, calculating the farthest point cloud and the nearest point cloud of the point cloud cluster in the direction of the preliminary fitting angle;

[0019] increasing the preliminary fitting angle by 90° to obtain a framing angle, and calculating the nearest point cloud and the farthest point cloud of the point cloud cluster in the direction of the framing angle;

[0020] According to the farthest point cloud and the nearest point cloud of the point cloud cluster in the direction of the preliminary fitting angle, and the nearest point cloud and the farthest point cloud of the point cloud cluster in the direction of the framing angle, generating the minimum rectangular bounding box of a point cloud cluster corresponding to the preliminary fitting angle.

[0021] With reference to the first aspect, the calculating a statistical variance value of the point cloud cluster and each of the rectangular bounding boxes includes calculating a statistical variance value of the point cloud cluster and each of the rectangular bounding boxes according to a preset objective function

[0022] F = -f v (d(p i | i∈1…n )-μ(p))| θ∈[0,π / 2) ,

[0023] where p represents a point cloud, d(p i) represents the point cloud p i the minimum distance of the rectangular bounding box determined by the preliminary fitting angle θ; μ(p) represents the average value of the nearest distance of all point clouds in the point cloud cluster to the rectangular bounding box determined by the preliminary fitting angle θ; the function f v () represents a function representing statistical variance;

[0024] The statistical variance value of the point cloud cluster and each rectangular bounding box is calculated.

[0025] In a second aspect, the application provides a device for extracting vehicle contour features based on a point cloud bird's eye view, comprising:

[0026] The acquisition module is configured to acquire the point cloud data.

[0027] The projection module is configured to project the point cloud data into a bird's eye view.

[0028] The clustering module is configured to obtain a point cloud cluster of a target vehicle.

[0029] The preliminary fitting angle selection module is configured to select a plurality of preliminary fitting angles from a preset angle range with a preset step size, and determine the rectangular bounding box determined by the point cloud cluster for each preliminary fitting angle.

[0030] The statistical variance calculation module is configured to calculate the statistical variance value of the point cloud cluster and each rectangular bounding box.

[0031] The contour feature determination module is configured to select the highest value from all the statistical variance values as the optimal statistical variance value, select the preliminary fitting angle corresponding to the optimal statistical variance value as the determined fitting angle, and select the rectangular bounding box corresponding to the determined fitting angle as the contour feature of the target vehicle.

[0032] In a third aspect, the application provides a computer-readable storage medium comprising a program or instructions, when the program or instructions are executed on a computer, the method of the first aspect is executed.

[0033] The application with the above technical solution has the following advantages:

[0034] The method for extracting vehicle contour features based on the point cloud bird's eye view of the application calculates the statistical variance values of the point cloud clusters and each of the rectangular surrounding frames, selects the highest value from all the statistical variance values as the optimal statistical variance value, takes the preliminary fitting angle corresponding to the optimal statistical variance value as the determined fitting angle, and takes the rectangular surrounding frame corresponding to the determined fitting angle as the contour feature of the target vehicle. The rectangular surrounding frame can represent the direction of more point clouds based on the statistical variance value, effectively improves the accuracy of the direction of the rectangular surrounding frame, obtains more accurate direction features of the fitted contour, and effectively avoids the influence of the point cloud clusters of noise points or convex objects on the fitting frame. BRIEF DESCRIPTION OF DRAWINGS

[0035] The application can be further illustrated by the non-limiting embodiments shown in the drawings.

[0036] Figure 1 The flowchart of the method for extracting vehicle contour features based on the point cloud bird's eye view in an embodiment of the application;

[0037] Figure 2 The point cloud cluster of two targets obtained by the clustering method under the bird's eye view in the method for extracting vehicle contour features based on the point cloud bird's eye view in an embodiment of the application;

[0038] Figure 3 The schematic diagram of taking the preliminary fitting angle according to the preset step length in the method for extracting vehicle contour features based on the point cloud bird's eye view in an embodiment of the application;

[0039] Figure 4 The rectangular surrounding frame corresponding to different preliminary fitting angles in the method for extracting vehicle contour features based on the point cloud bird's eye view in an embodiment of the application;

[0040] Figure 5 The system block diagram of the device for extracting vehicle contour features based on the point cloud bird's eye view in an embodiment of the application;

[0041] Among them, the acquisition module 1, the projection module 2, the clustering module 3, the preliminary fitting angle selection module 4, the statistical variance calculation module 5, and the contour feature determination module 6. DETAILED DESCRIPTION

[0042] The application will be described in detail below with reference to the drawings and specific embodiments. It should be noted that similar or identical parts are denoted by the same reference numerals in the drawings or description, and the implementation modes not shown or described in the drawings are known to those skilled in the art. In the description of the application, the terms "first", "second", etc. are only used for differentiation and description, and cannot be understood as indicating or implying relative importance.

[0043] The embodiment of the present application provides a method for extracting vehicle contour features based on a point cloud bird's eye view, as shown in the figure, the method comprises the following steps: Figure 1

[0044] S1: collecting point cloud data and projecting the point cloud data into a bird's eye view;

[0045] S2: obtaining a point cloud cluster of a target vehicle through a clustering method;

[0046] S3: selecting a plurality of preliminary fitting angles, each of the preliminary fitting angles corresponding to a minimum rectangular bounding box of a point cloud cluster;

[0047] S4: calculating statistical variance values of the point cloud cluster and each of the rectangular bounding boxes;

[0048] S5: selecting a highest value from all the statistical variance values as an optimal statistical variance value;

[0049] S6: taking the preliminary fitting angle corresponding to the optimal statistical variance value as a determined fitting angle, and taking the rectangular bounding box corresponding to the determined fitting angle as a contour feature of the target vehicle.

[0050] It should be noted that in step S1, the point cloud data is collected by a point cloud sensor, and a commonly used point cloud sensor in the prior art is a laser radar. The point cloud data of the road surface is collected by the laser radar on the ego vehicle. The point cloud data collected by the laser radar is visualized as three-dimensional point cloud data, and the three-dimensional point cloud data can be projected onto a plane along the vertical direction to obtain two-dimensional point cloud data, that is, two-dimensional point cloud under the bird's eye view is generated according to the point cloud of the target object.

[0051] It should be noted that in step S2, the clustering method is a conventional clustering method in the prior art. For example, a DBSCAN clustering algorithm or a Euclidean distance clustering algorithm is used to classify points with similar distances as a class by judging the distance between each two points, so as to realize the clustering of the point cloud of the target vehicle. As shown in the figure, two point cloud clusters of the target vehicles are obtained under the bird's eye view through the clustering method. Figure 2

[0052] In some optional embodiments, before the point cloud data is projected into the bird's eye view, the method further comprises ground removal, that is, removing the point cloud belonging to the ground through plane fitting. It should be noted that the ground removal is used to avoid the bird's eye view after projection being messy due to a large number of ground points, and to reduce the trouble of subsequent point cloud classification, recognition and tracking. In general, through the ground removal, the effect of clustering is avoided from being affected by the ground points.

[0053] ​​In some optional embodiments, the selecting multiple preliminary fitting angles includes: selecting multiple preliminary fitting angles from a preset angle range with a preset step size. It should be noted that the preset angle range is an angle range set according to actual conditions, and in this embodiment, the preset angle range is 0-90°, and the preset step size is 1° or 0.5°.

[0054] In some optional embodiments, the selecting multiple preliminary fitting angles from a preset angle range with a preset step size includes: taking the origin of the ego vehicle coordinate system as a rotation point, taking the forward direction of the ego vehicle as the X axis, taking the X axis as the starting direction, and selecting multiple preliminary fitting angles from the preset angle range with the preset step size. It should be noted that, as shown in Figure 3 In this embodiment, the preset step size is taken as an example of 1°, the origin of the ego vehicle coordinate system is taken as a rotation point, the forward direction of the ego vehicle is taken as the X axis, the X axis is taken as the starting direction, and 90 preliminary fitting angles are selected in the range of 0-90°.

[0055] In some optional embodiments, the corresponding determination of each preliminary fitting angle to a minimum rectangular bounding box of a point cloud cluster includes:

[0056] According to the preliminary fitting angle, the farthest point cloud and the nearest point cloud of the point cloud cluster in the direction of the preliminary fitting angle are calculated;

[0057] The preliminary fitting angle is increased by 90° to obtain a framing angle, and the nearest point cloud and the farthest point cloud of the point cloud cluster in the direction of the framing angle are calculated;

[0058] According to the farthest point cloud and the nearest point cloud of the point cloud cluster in the direction of the preliminary fitting angle, and the nearest point cloud and the farthest point cloud of the point cloud cluster in the direction of the framing angle, the minimum rectangular bounding box of the point cloud cluster corresponding to the preliminary fitting angle is generated.

[0059] It should be noted that the line connecting the farthest point cloud and the nearest point cloud of the point cloud cluster in the direction of the preliminary fitting angle, the line connecting the nearest point cloud and the farthest point cloud of the point cloud cluster in the direction of the framing angle, the line connecting the farthest point cloud in the direction of the preliminary fitting angle and the farthest point cloud in the direction of the framing angle, and the line connecting the nearest point cloud in the direction of the preliminary fitting angle and the nearest point cloud in the direction of the framing angle generate the rectangular bounding box, and in this embodiment, 90 rectangular bounding boxes are generated corresponding to 90 preliminary fitting angles, as shown in Figure 4 .

[0060] In some optional embodiments, the calculation of the statistical variance value of the point cloud cluster and each rectangular bounding box includes: according to a preset target function

[0061] F = -f v (d(p i | i∈1…n )-μ(p))| θ∈[0,π / 2) ,

[0062] wherein, p represents a point cloud, d(p i ) represents a minimum distance of the point cloud p i to the rectangular bounding box determined by the preliminary fitting angle θ; μ(p) represents an average value of the nearest distance of all point clouds in the point cloud cluster to the rectangular bounding box determined by the preliminary fitting angle θ; the function f v () represents a function representing statistical variance;

[0063] The statistical variance value of the point cloud cluster and each rectangular bounding box is calculated.

[0064] It should be noted that the statistical variance value represents the stable value of the point cloud cluster in the rectangular bounding box, and the greater the statistical variance value, the more stable the point cloud in the rectangular bounding box, and the more capable of representing the direction of the point cloud cluster.

[0065] The embodiment of the application also provides a device for extracting vehicle contour features based on a point cloud bird's eye view, as shown in Figure 5 , comprising:

[0066] The acquisition module 1 is used to acquire the point cloud data.

[0067] The projection module 2 is used to project the point cloud data into a bird's eye view.

[0068] The clustering module 3 is used to obtain a point cloud cluster of a target vehicle.

[0069] The preliminary fitting angle selection module 4 is used to select a plurality of preliminary fitting angles from a preset angle range with a preset step size, and determine the rectangular bounding box corresponding to each preliminary fitting angle determined by the point cloud cluster.

[0070] The statistical variance calculation module 5 is used to calculate the statistical variance value of the point cloud cluster and each rectangular bounding box.

[0071] The contour feature determination module 6 is used to select the highest value from all the statistical variance values as the optimal statistical variance value, select the preliminary fitting angle corresponding to the optimal statistical variance value as a determined fitting angle, and select the rectangular bounding box corresponding to the determined fitting angle as the contour feature of the target vehicle.

[0072] The embodiment of the present application further provides a computer readable storage medium, comprising a program or instructions, when the program or instructions are executed on a computer, the above-mentioned method for extracting vehicle contour features based on a point cloud bird's eye view is executed.

[0073] The above describes the method, device and medium for extracting vehicle contour features based on a point cloud bird's eye view provided by the present application in detail. The description of the specific embodiments is only used to help understand the method of the present application and its core idea. It should be pointed out that, for those skilled in the art, without departing from the principle of the present application, the present application can be improved and modified in several ways, and these improvements and modifications also fall within the protection scope of the claims of the present application.

Claims

1. A method for extracting vehicle contour features based on a point cloud bird's eye view, characterized in that, The method comprises: Collecting point cloud data and projecting the point cloud data into an aerial view; Obtaining a point cloud cluster of a target vehicle through a clustering method; Selecting multiple preliminary fitting angles, each of which corresponds to a minimum rectangular bounding box of a point cloud cluster; Each of the preliminary fitting angles corresponds to a minimum rectangular bounding box of a point cloud cluster, comprising: According to the preliminary fitting angle, the farthest point cloud and the nearest point cloud of the point cloud cluster in the direction of the preliminary fitting angle are calculated; The preliminary fitting angle is increased by 90° to obtain a framing angle, and the nearest point cloud and the farthest point cloud of the point cloud cluster in the direction of the framing angle are calculated; According to the farthest point cloud and the nearest point cloud of the point cloud cluster in the direction of the preliminary fitting angle, and the nearest point cloud and the farthest point cloud of the point cloud cluster in the direction of the framing angle, the minimum rectangular bounding box of the preliminary fitting angle corresponding to a point cloud cluster is generated; Calculate the statistical variance value of the point cloud cluster and each of the rectangular bounding boxes; Select the highest value from all the statistical variance values as the optimal statistical variance value; The preliminary fitting angle corresponding to the optimal statistical variance value is taken as the determination fitting angle, and the rectangular bounding box corresponding to the determination fitting angle is taken as the extracted contour feature of the target vehicle.

2. The method for extracting vehicle contour features based on a point cloud bird's eye view according to claim 1, characterized in that, Before the point cloud data is projected into the aerial view, it further comprises: Ground removal, removing point clouds belonging to the ground by plane fitting.

3. The method of claim 1, wherein, The multiple preliminary fitting angles are selected, comprising: Select multiple preliminary fitting angles from a preset angle range with a preset step size.

4. The method for extracting vehicle contour features based on a point cloud bird's eye view according to claim 3, characterized in that, The multiple preliminary fitting angles are selected from the preset angle range with the preset step size, comprising: Taking the origin of the ego vehicle coordinate system as the rotation point, taking the forward direction of the ego vehicle as the X axis, and taking the X axis as the starting direction, multiple preliminary fitting angles are selected from the preset angle range with the preset step size.

5. The method for extracting vehicle contour features based on a point cloud bird's eye view according to claim 1, characterized in that, The calculation of the statistical variance value of the point cloud cluster and each of the rectangular bounding boxes comprises: According to a preset target function ， wherein, representing a point cloud, ) representing a point cloud to the preliminary fitting angle the minimum distance of the determined rectangular bounding box; representing all point clouds in the point cloud cluster, to the preliminary fitting angle the average of the nearest distances of the determined rectangular bounding box, function representing a function characterizing the statistical variance; Calculate the statistical variance value of the point cloud cluster and each of the rectangular bounding boxes.

6. The method for extracting vehicle contour features based on a point cloud bird's eye view according to claim 1, characterized in that, The point cloud cluster of the target vehicle is obtained through the clustering method, comprising: Based on the DBSCAN algorithm or the Euclidean clustering method, by judging the distance between every two points, the point clouds with similar distances are classified into the same point cloud cluster.

7. An apparatus for extracting vehicle contour features based on a point cloud bird's eye view, configured to perform the method for extracting vehicle contour features based on a point cloud bird's eye view according to any one of claims 1 to 6, characterized in that Comprise: The acquisition module is used for collecting the point cloud data; The projection module is used for projecting the point cloud data into an aerial view; The clustering module is used for obtaining a point cloud cluster of a target vehicle; The preliminary fitting angle selection module is used for selecting multiple preliminary fitting angles from a preset angle range with a preset step size, and determining the rectangular bounding box corresponding to each of the preliminary fitting angles of the point cloud cluster; The statistical variance calculation module is used for calculating the statistical variance value of the point cloud cluster and each of the rectangular bounding boxes; The contour feature determination module is used for selecting the highest value from all the statistical variance values as the optimal statistical variance value, taking the preliminary fitting angle corresponding to the optimal statistical variance value as the determination fitting angle, and taking the rectangular bounding box corresponding to the determination fitting angle as the extracted contour feature of the target vehicle.

8. A computer-readable storage medium, characterized in that, comprising a program or instructions, which when run on a computer, cause the method of any one of claims 1 to 6 to be performed.

Images