WO2019214193A1

WO2019214193A1 - Point cloud sampling method, image processing device, and device having storage function

Info

Publication number: WO2019214193A1
Application number: PCT/CN2018/114433
Authority: WO
Inventors: 张智胜; 祁春超; 黄雄伟; 向志华
Original assignee: 深圳市华讯方舟太赫兹科技有限公司; 华讯方舟科技有限公司
Priority date: 2018-05-09
Filing date: 2018-11-07
Publication date: 2019-11-14
Also published as: CN108597019A

Abstract

A point cloud sampling method, an image processing device, and a device having a storage function. The point cloud sampling method comprises: acquiring point cloud data of a three-dimensional image (S10); calculating the curvatures of data points in the point cloud data (S11); configuring a sampling radius corresponding to the data points (S12); sampling the point cloud data on the basis of the sampling radius configured, thus allowing the data density of an area of a greater curvature to be greater than an area of a smaller curvature (S13). The method allows redundant data to be deleted while retaining an increased amount of feature data.

Description

Point cloud sampling method, image processing device, and device having storage function

[Technical Field]

The present application relates to the field of image processing technologies, and in particular, to a point cloud sampling method, an image processing device, and a device having a storage function.

【Background technique】

With the development of 3D scanning technology, the 3D point cloud data model has become an emerging digital media, which has been widely used in industrial manufacturing, architectural design, product display, medicine, and e-commerce. Since the accuracy of the existing three-dimensional scanning device is greatly improved, the scanned point cloud data has extremely high precision and contains a large amount of redundant data, which leads to complicated and slow data processing.

[Summary of the Invention]

The present application mainly provides a point cloud sampling method, an image processing device, and a device having a storage function, which can simplify point cloud data.

To solve the above technical problem, a technical solution adopted by the present application is to provide a point cloud sampling method, including: acquiring point cloud data of a three-dimensional image; calculating a curvature of each data point in the point cloud data; setting each data point Corresponding sampling radius, wherein the sampling radius corresponding to the data point with larger curvature is not larger than the sampling radius corresponding to the data point with smaller curvature, and the sampling radius corresponding to at least part of the data point with larger curvature is smaller than the data point corresponding to the smaller curvature The sampling radius; the point cloud data is sampled by the sampling radius, so that the data density of the region with larger curvature is larger than the region with smaller curvature.

In order to solve the above technical problem, another technical solution adopted by the present application is to provide an image processing apparatus including: a communication circuit and a processor connected to each other; a communication circuit for acquiring point cloud data of a three-dimensional image; Calculating the curvature of each data point in the point cloud data; setting a sampling radius corresponding to each data point, wherein the sampling radius corresponding to the data point having a larger curvature is not larger than the sampling radius corresponding to the data point having a smaller curvature, and at least part of the curvature is compared The sampling radius corresponding to the large data point is smaller than the sampling radius corresponding to the data point with the smaller curvature; the point cloud data is sampled based on the set sampling radius, so that the data density of the region with larger curvature is larger than the region with smaller curvature.

In order to solve the above technical problem, another technical solution adopted by the present application is to provide a device having a storage function, which stores an instruction, and when executed, implements a point cloud sampling method as described above.

The beneficial effects of the present application are: different from the prior art, in some embodiments of the present application, calculating the curvature of each data point in the point cloud data by acquiring the point cloud data of the three-dimensional image, and setting each data point correspondingly The sampling radius, wherein the sampling radius corresponding to the data point with larger curvature is not larger than the sampling radius corresponding to the data point with smaller curvature, and the sampling radius corresponding to at least part of the data point with larger curvature is smaller than the sampling corresponding to the data point with smaller curvature Radius, the point cloud data is sampled by the sampling radius, so that the data density of the region with larger curvature is larger than the region with smaller curvature, so that more data points are reserved in the region with larger curvature and remain in the region with less curvature. Less data points, in addition to deleting redundant data, retain feature data, which is beneficial to improve the speed of subsequent data processing, while reducing storage space and saving costs.

[Description of the Drawings]

1 is a schematic flow chart of a first embodiment of a point cloud sampling method according to the present application;

2 is a schematic flow chart of a second embodiment of a point cloud sampling method according to the present application;

3 is a schematic diagram showing the spatial position of a first data point and a second data point;

4 is a schematic flow chart of a third embodiment of a point cloud sampling method according to the present application;

Figure 5 is a schematic diagram of a specific process of each step in Figure 4;

6 is a schematic flowchart of a fourth embodiment of a point cloud sampling method according to the present application;

7 is a schematic diagram of raw point cloud data of a three-dimensional image that is not sampled;

8 is a schematic diagram of three-dimensional image point cloud data processed by the point cloud sampling method of the present application;

9 is a schematic flowchart of a fifth embodiment of a point cloud sampling method according to the present application;

10 is a schematic structural diagram of an embodiment of an image processing apparatus of the present application;

FIG. 11 is a schematic structural diagram of an embodiment of an apparatus having a storage function according to the present application.

【detailed description】

The technical solutions in the embodiments of the present application are clearly and completely described in the following with reference to the drawings in the embodiments of the present application. It is obvious that the described embodiments are only a part of the embodiments of the present application, and not all of the embodiments. All other embodiments obtained by a person of ordinary skill in the art based on the embodiments of the present application without departing from the inventive scope are the scope of the present application.

As shown in FIG. 1 , the first embodiment of the point cloud sampling method of the present application includes:

S10: Acquire point cloud data of a three-dimensional image.

A point cloud is a large number of data points that express the surface of an object. Each data point can be represented by three-dimensional coordinates. The point cloud data includes at least a position coordinate of a plurality of data points that express the contour of the three-dimensional image, and the third-dimensional point cloud data of the target may be directly scanned by using the three-dimensional scanning device, or may be pre-stored from a storage device or other device. The point cloud data of the three-dimensional image is not specifically limited herein.

S11: Calculate the curvature of each data point in the point cloud data.

The data points in the point cloud usually do not have an explicit geometric topological relationship. Usually, the plane fitting method is used to obtain the curvature of the fitted plane as the curvature of the data points. Among them, the method of plane fitting can adopt the eigenvalue method or the least square method.

S12: Set the sampling radius corresponding to each data point.

The curvature of each data point can indicate the degree of curvature of the data point. The local area where the data point with larger curvature is more curved, and the local area where the data point with smaller curvature is less curved and relatively flat. Among them, the local area with a large degree of curvature contains more characteristic data points that can reflect the spatial characteristics of the three-dimensional image.

The sampling radius defines a circular sampling area centered on the data point and having a radius corresponding to the sampling radius, and the amount of sampled data is the same in each sampling area. In this embodiment, when the sampling radius corresponding to the data point is set, the sampling radius corresponding to the data point with larger curvature is not larger than the sampling radius corresponding to the data point with smaller curvature, and the sampling radius corresponding to the data point with at least partial curvature is larger. The sampling radius corresponding to the data point smaller than the curvature.

For example, set a range of sampling radii, such as [a,b], where a is the minimum sampling radius and b is the maximum sampling radius. When setting the sampling radius corresponding to each data point, the sampling radius corresponding to the data point with the largest curvature can be set to a, the sampling radius corresponding to the data point with the smallest curvature is set to b, and the number N of the data points can be obtained at the same time. Set the sampling radius to the arithmetic progression column, that is, set the sampling radius difference corresponding to each two data points to (ba)/N, and set the corresponding sampling radius according to the order of the data points from small to large, wherein the corresponding sampling radius is from Big to small changes.

Of course, in other applications, the sampling radius corresponding to multiple data points in a certain curvature range may be set to the same value, which is not specifically limited herein.

S13: The point cloud data is sampled based on the set sampling radius, so that the data density of the region with larger curvature is larger than the region with smaller curvature.

Specifically, when the point cloud data is sampled based on the set sampling radius, the data points may be selected in a preset order for sampling, wherein the preset order may be an increasing or decreasing order of curvature, or may be other preset sequences. , or randomly select data points for sampling, which is not specifically limited herein.

In one application, the data points with the smallest curvature are first selected for sampling. When sampling, the data points in the sampling region centered on the data point with the smallest curvature can be deleted. The number and location of the reserved data points can be set according to actual needs, such as only retaining the central data point, or only retaining the central data point and the data point closest to the central data point. Then, the next undeleted data point is selected in the order of increasing curvature, and the sampling process is the same as the above process until all the samples of the undeleted data points are completed. After the sampling is completed, the data points retained by each sampling radius are the same, but the sampling radius of the data points with larger curvature is not larger than the sampling radius of the data points with smaller curvature, and the sampling radius of the data points with at least partial curvature is The sampling radius of the data point smaller than the curvature is smaller. Therefore, the area of the area with larger curvature has more data points per unit area than the area with smaller curvature, that is, the data density of the area with larger curvature is larger than the area with smaller curvature. Data density. Therefore, while streamlining the data points in the point cloud, the area with a large degree of curvature can retain more data points, that is, retaining more feature points as much as possible, which is beneficial to improve the speed of subsequent data processing, and can also reduce storage. Space, saving costs.

In other embodiments, in order to maximize the deletion of redundant data when sampling based on the sampling radius, only the central data points may be retained.

As shown in FIG. 2, the second embodiment of the point cloud sampling method of the present application is based on the first embodiment of the point cloud sampling method of the present application, and further defining step S13 includes:

S131: Acquire a position coordinate of the first data point.

The first data point is a center point of a sampling area that is randomly or in a preset order. The position coordinates of the first data point are three-dimensional coordinates, which can be represented by three-dimensional coordinates of the European space, such as (x ₁ , y ₁ , z ₁ ).

S132: Calculate the distance between the first data point and the second data point.

Wherein the second data point is at least part of the data point except the first data point. The second data point may be other data points than the first data point, or may be part of the data points except the first data point, for example, when the data point is divided into multiple parts and simultaneously sampled, the second data point It is the other data points in the sampled data points except the first data point.

S133: Determine whether the distance is smaller than a sampling radius corresponding to the first data point.

If the result of the determination is yes, then step S134 is performed, otherwise step S135 is performed.

S134: Delete the second data point.

S135: The second data point is reserved.

S136: Select another data point that is not deleted as the first data point according to the preset sampling order, and repeat the above steps until the traversal completes all the undeleted data points.

The preset sampling sequence is a preset sequence, which may be a sequence of curvature changes, a sequence of spatial positions, or a random storage order, and is not specifically limited herein.

Specifically, in an application example, the second data point is a data point other than the first data point, and when sampling, the position coordinates of the first data point and the second data point need to be acquired first, for example, FIG. 3 Point A (x ₁ , y ₁ , z ₁ ) and point B (x ₂ , y ₂ , z ₂ ), and then calculate the distance between the first data point and the second data point, such as calculating the first data Euclidean distance between point A and second data point B

After the distance d is obtained, the distance d is compared with the sampling radius R1 corresponding to the first data point A. If the distance d is smaller than the sampling radius R1, it indicates that the second data point B is located at the first data point A. For the center, in the circular area with the radius of the sampling radius R1 as the radius, at this time, the second data point B is deleted. When the second data point is not located in a circular area centered on the first data point A and having a radius of the sampling radius R1, as shown by the point C (x ₃ , y ₃ , z ₃ ) in FIG. 3, The distance between C and point A is greater than the sampling radius R1, at which time the second data point C is retained. Then, another data point that has not been deleted may be continuously selected as the first data point according to a preset sampling sequence, and the above sampling step is repeatedly performed until the traversal completes all the undeleted data points, that is, all the undeleted data points are taken as The first data point performs the above sampling step, and the sampling process is completed. After the above sampling process, after each data point is used as the first data point, the sampling area formed by the corresponding sampling radius retains only itself, and other data points in the sampling area are deleted, thereby deleting redundant data. The streamlined point cloud also reduces the storage space, which is beneficial to improve the speed of subsequent data processing and improve efficiency.

In other embodiments, the point cloud data may be divided into a plurality of sub-point clouds and the sampling radius is set for sampling.

As shown in FIG. 4, the third embodiment of the point cloud sampling method of the present application is based on the first embodiment of the point cloud sampling method of the present application, and further defining step S12 includes:

S121: The data points are arranged in a preset curvature order and then divided into a plurality of sub-point clouds.

The preset curvature order may be an increasing or decreasing order of curvature, or may be a decreasing order after the curvature is incremented, or may be an increasing order after the decreasing of the curvature, which is determined according to actual needs, and is not specifically limited herein.

The number of sub-point clouds divided is at least two, and the specific number can be set according to actual needs, for example, five sub-point clouds. The number of data points in each sub-point cloud may be the same or different, and is not specifically limited herein.

Of course, in other embodiments, when dividing the sub-point cloud, a curvature range may be set for each sub-point cloud, and data points falling within the corresponding curvature range are divided into corresponding sub-point clouds, and then arranged.

Optionally, as shown in FIG. 5, step S121 includes:

S1211: Arrange the data points in the order of increasing/decrementing the curvature.

S1212: The ranked data points are equally divided into a plurality of sub-point clouds.

Specifically, in an application example, after calculating the curvature of each data point, the data points may be arranged in the order of increasing/decreasing curvature, for example, 1000 data are arranged in increasing order of curvature, wherein the data points having the same curvature They can be arranged in the original storage order or randomly. When dividing the sub-point cloud, the arranged data points may be equally divided into N sub-point clouds in order, for example, 1000 arranged data points are equally divided into 5 sub-point clouds, and each sub-point cloud includes 200 data points.

S122: Set a sampling radius for each sub-point cloud, wherein a sampling radius of the sub-point cloud with a larger curvature of the data point is smaller than a sampling radius of the sub-point cloud with a smaller curvature of the data point.

The specific value of the sampling radius may be determined according to the actual situation of the spatial range of the three-dimensional image, and is not specifically limited herein.

In one application, the curvature of a data point in a sub-point cloud is [0.01, 0.03), and the curvature of a data point in another sub-point cloud is (0.03, 0.04), and the data point in the previous sub-point cloud. The curvature of the data point is smaller than the curvature of the data point in the next sub-point cloud. At this time, when the sampling radius is set for the sub-point cloud, the sampling radius of the previous sub-point cloud should be larger than the sampling radius of the latter sub-point cloud, for example, the previous sub-point. The sampling radius of the cloud is set to 0.5, and the sampling radius of the latter sub-point cloud is set to 0.4.

Optionally, referring to FIG. 5, step S122 includes:

S1221: The sampling radius of the sub-point cloud is sequentially set to a decreasing arithmetic sequence with the first radius as an initial value in the order of increasing curvature; or the sampling radius of the sub-point cloud is sequentially set to the first in the order of decreasing curvature The second radius is an incremental progression of the initial value.

The first radius is a preset maximum sampling radius, and the second radius is a preset minimum sampling radius, and the specific value may be determined according to actual needs, and is not specifically limited herein.

Specifically, in an application example, when data points are sequentially divided into m sub-point clouds in order of increasing curvature, the curvature of the data points in the first sub-point cloud is the smallest, and the sampling of the first sub-point cloud can be performed at this time. The radius is set to the first radius b, and the sampling radius of the other sub-point clouds can be set according to the decreasing arithmetic progression column, that is, according to the following formula (1):

Where R is the sampling radius of the sub-point cloud, b is the first radius, a is the second radius, m is the number of sub-point clouds, and I is the sequence number of the sub-point cloud. The sequence number of the first sub-point cloud is 0, the sequence number of the second sub-point cloud is 2, and so on, and the sequence number of the m-th sub-point cloud is m-1. Where a>0, b>0, a<b.

When the data points are sequentially divided into m sub-point clouds in the order of decreasing curvature, the curvature of the data points in the first sub-point cloud is the largest, and the sampling radius of the first sub-point cloud can be set to the second radius a. The sampling radius of other sub-point clouds can be set according to the incremental difference series, that is, according to the following formula (2):

This embodiment can also be combined with the second embodiment of the point cloud sampling method of the present application.

In other embodiments, after dividing a plurality of sub-point clouds, each sub-point cloud may be sampled and then the sampled data points are combined, and finally the sampled point cloud may be obtained.

Specifically, as shown in FIG. 6 , the fourth embodiment of the point cloud sampling method of the present application is based on the second embodiment of the point cloud sampling method of the present application, and further defines that the preset sampling order is an increasing or decreasing order of curvature, and The various sub-steps of step S13 are further defined.

Wherein, step S131 includes:

S1311: Select a curvature minimum/maximum data point in the first sub-point cloud as an initial first data point, and obtain a position coordinate of the first data point.

The first sub-point cloud is a sub-point cloud currently being sampled, and when sampling, the data points may be selected in the order of increasing/decreasing curvature. First, the data point with the largest/minimum curvature in the first sub-point cloud may be selected as the initial first data point, and the position coordinates thereof may be obtained, and the position coordinates may be three-dimensional European space coordinates.

As shown in FIG. 6, the second data point is other data points in the first sub-point cloud except the first data point, and step S132 includes:

S1321: Calculate a first distance between the other data points in the first sub-point cloud and the first data point.

The other data points in the first sub-point cloud are other data points in the first sub-point cloud except the current first data point. When calculating the first distance between the other data points and the first data point, the position coordinates of the first data point and other data points may be obtained, and the Euclidean distance between the two data points is calculated by using a formula, that is, the first distance.

As shown in FIG. 6, step S133 includes:

S1331: Determine whether the first distance is smaller than a sampling radius corresponding to the first sub-point cloud.

Specifically, each sub-point cloud has set a corresponding sampling radius, and after calculating the first distance between the first data point and the other data points, the first distance and the sampling radius corresponding to the first sub-point cloud may be For comparison, if the first distance is smaller than the sampling radius corresponding to the first sub-point cloud, if the determination result is time, step S1341 is performed.

S1341: Determine that the distance between the first data point and the second data point is smaller than a sampling radius corresponding to the first data point.

After determining that the distance between the first data point and the second data point is smaller than the sampling radius corresponding to the first data point, step S134 may be continued to delete the second data point.

If the result of the determination in step S1331 is negative, it is determined that the distance between the first data point and the second data point is greater than or equal to the sampling radius corresponding to the first data point, and the step S135 may be continued to retain the second data. point.

With continued reference to Figure 6, step S136 includes:

S1361: Select another data point that is not deleted in the first sub-point cloud as the first data point in the order of increasing/decreasing the curvature, and return to performing the step of acquiring the position coordinates of the first data point until the traversing completes the first sub-point cloud. Undeleted data points.

Specifically, if the first distance is smaller than the sampling radius corresponding to the first sub-point cloud, if the determination result is yes, it may be determined that the distance between the first data point and the second data point is smaller than the corresponding one of the first data points. The sampling radius can be continued at step S1342, that is, the second data point is deleted. Otherwise, step S1351 is performed, that is, the second data point is reserved. Then, another data point that is not deleted in the first sub-point cloud may be selected as the first data point in the order of increasing or decreasing the curvature, and steps S131-S135 are repeatedly performed until the traversal is completed without deleting the first sub-point cloud. The data points, that is, the data points that are not deleted in the first sub-point cloud, have been sampled as the first data points. Wherein, if the initially selected first data point is the data point with the smallest curvature, the next data point is selected in the order of increasing curvature, and if the first selected data point is the data point with the largest curvature, the order of decreasing curvature is performed. Select the next data point.

Optionally, after step S1361, the method further includes:

S1371: Select another sub-point cloud that is not sampled, and return to perform the step of acquiring the position coordinates of the first data point until the traversal completes all the sub-point clouds.

Specifically, each sub-point cloud may be sampled according to the step of the first sub-point cloud described above, until the sampling of all the sub-point clouds is completed, and the sampling process of the three-dimensional image point cloud data is completed. At this time, each sub-point cloud after the sampling is completed may be combined to obtain the sampled point cloud data, and the point cloud data may be output or displayed, and the sampled three-dimensional image may be obtained.

For example, as shown in FIG. 7 and FIG. 8 , the original point cloud data of the unsampled three-dimensional image is as shown in FIG. 7 , and the spatial distribution of the original point cloud data is relatively uniform, the data is dense, the amount of data is large, and the subsequent data processing process is cumbersome and consumes. The duration of the three-dimensional image point cloud data processed by the point cloud sampling method of the present application is as shown in FIG. 8 , and the data density of the region where the curvature of the sampled point cloud data is small is small, the data points are sparse, and the curvature is large. Regional data density is large and data points are dense. It can be seen from the comparison between FIG. 7 and FIG. 8 that the point cloud sampling method of the present application can delete redundant data and streamline the point cloud while retaining feature points as much as possible, which is beneficial to improve the rate and efficiency of subsequent data processing.

This embodiment can also be combined with the third embodiment of the point cloud sampling method of the present application.

In the point cloud sampling method of the present application, when calculating the curvature of the data point, the curvature of the data point can be calculated by using the eigenvalue method after fitting the neighborhood point plane.

As shown in FIG. 9 , the fifth embodiment of the point cloud sampling method of the present application is based on the first embodiment of the point cloud sampling method of the present application, and further defining step S11 includes:

S111: Acquire a neighborhood point where the first data point is within a preset distance.

The preset distance may be a preset range of the neighborhood, and the specific value may be determined according to actual requirements, and is not specifically limited herein.

Specifically, in an application example, a binary tree, such as a KD-tree (KD-number), may be used to acquire all points within a preset distance R from the first data point, that is, neighbor points. Of course, in other embodiments, an octree or other method may also be used to obtain the neighborhood point of the first data point.

S112: Using a eigenvalue method to fit a neighborhood point to a plane equation to obtain a feature matrix of a best fit plane.

Specifically, first, the spatial plane equation is set to ax+by+cz=d, where a, b, and c are unit normal vectors of the plane, that is, a ² +b ² +c ² =1, and d is the coordinate origin to the plane. Distance, d ≥ 0. The distance from any one of the neighborhood points (x _i , y _i , z _i ) to the plane is

d _i =|ax _i +by _i +cz _i -d|; (3)

In order to obtain the best fit plane, it should be satisfied under the condition a ² + b ² + c ² =1:

Using the Lagrangian multiplier method to find the extremum of the function, you can get the following function:

Deriving the above formula (5) for d and making the derivative zero can be calculated.

among them:

Let the above formula (5) be derived for a, and let the derivative be zero.

among them:

Similarly, formula (5) is derived for b and c, respectively, and the derivative is zero.

Then the formula (7-9) can constitute the following eigenvalue equation

With the above formula (10), a real symmetric matrix with a feature matrix of 3*3 of the best fit plane can be obtained.

S113: Calculate a feature value of the feature matrix.

S114: Taking the ratio of the smallest feature value of the feature values to the sum of all the feature values as the curvature of the first data point.

Specifically, by solving the feature values and feature vectors of the feature matrix A, the unit normal vectors a, b, and c of the best fit plane can be obtained.

The eigenvalues of the real symmetric matrix A can be calculated by the following formula (11).

|A-λI|=0; (11)

Where I is the identity matrix. Since A is a real symmetric matrix of 3*3, matrix A has at most three real eigenvalues, respectively λ ₀ , λ ₁ , λ ₂ , where λ ₀ is the minimum eigenvalue, and the curvature σ of the first data point can be Calculated using the following formula (12)

That is, the curvature σ of the first data point is the ratio of the minimum eigenvalue of the feature matrix of the best fit plane to the sum of all the feature values. The curvature of other data points can also be calculated by the above method, and will not be repeated here.

In this embodiment, the curvature of each data point is calculated by using the eigenvalue method. Of course, in other embodiments, the curvature of the data point may be calculated by other methods such as least squares method, which is not specifically limited herein.

This embodiment can also be combined with any of the second to fourth embodiments of the point cloud sampling method of the present application or a combination thereof that does not conflict.

As shown in FIG. 10, an embodiment of the image processing apparatus 20 of the present application includes: a communication circuit 201 and a processor 202 that are connected to each other.

The communication circuit 201 is configured to transmit and receive data, and is an interface that the image processing device 20 communicates with other devices. Specifically, the communication circuit 201 is configured to acquire point cloud data of a three-dimensional image.

The processor 202 controls the operation of the communication device, and the processor 202 may also be referred to as a CPU (Central Processing Unit). Processor 202 may be an integrated circuit chip with signal processing capabilities. The processor 202 can also be a general purpose processor, a digital signal processor (DSP), an application specific integrated circuit (ASIC), an off-the-shelf programmable gate array (FPGA) or other programmable logic device, discrete gate or transistor logic device, discrete hardware component. . The general purpose processor may be a microprocessor or the processor or any conventional processor or the like.

In particular, the processor 202 is operative to execute a program to implement a method as provided by any one of the first to fifth embodiments of the point cloud sampling method of the present application or a non-conflicting combination thereof.

Of course, in other embodiments, the image processing device 20 may further include other components such as a memory (not shown), which are not specifically limited herein.

The image processing device in this embodiment may be a mobile terminal, a fixed terminal, a server, or the like, or may be an integrated independent component such as an image processing chip.

In the image processing device of this embodiment, after acquiring the point cloud data of the three-dimensional image through the communication circuit, the processor calculates the curvature of each data point in the point cloud data, and sets a sampling radius corresponding to each data point, wherein the curvature is larger. The sampling radius corresponding to the data point is not larger than the sampling radius corresponding to the data point with the smaller curvature, and the sampling radius corresponding to the data point with at least part of the curvature is smaller than the sampling radius corresponding to the data point with the smaller curvature, and the sampling radius is used for the point cloud. The data is sampled so that the data density of the region with larger curvature is larger than the region with smaller curvature, so that more data points are reserved in the region with larger curvature, and fewer data points are reserved in the region with less curvature, and then deleted. At the same time of redundant data, retaining feature data is beneficial to improve the speed of subsequent data processing, while reducing storage space and saving costs.

As shown in FIG. 11 , in an embodiment of the present application having a storage function, the device 30 having a storage function internally stores an instruction 301. When the instruction 301 is executed, the first to fifth applications of the point cloud sampling method of the present application are implemented. A sampling method provided by an embodiment or a combination thereof that does not conflict.

The device 30 having a storage function may be a portable storage medium such as a USB flash drive or an optical disk, or may be a terminal, a server, or an integrated independent component such as an image processing chip or the like.

In this embodiment, when the instruction stored in the device with the storage function is executed, the curvature of each data point in the point cloud data is calculated by acquiring the point cloud data of the three-dimensional image, and the sampling radius corresponding to each data point is set. The sampling radius corresponding to the data point with larger curvature is not larger than the sampling radius corresponding to the data point with smaller curvature, and the sampling radius corresponding to at least part of the data point with larger curvature is smaller than the sampling radius corresponding to the data point with smaller curvature, The sampling radius samples the point cloud data so that the data density of the region with larger curvature is larger than the region with smaller curvature, so that more data points are reserved in the region with larger curvature, and less data is retained in the region with smaller curvature. Points, in addition to deleting redundant data, retain feature data, which is beneficial to improve the speed of subsequent data processing, while reducing storage space and saving costs.

The above description is only the embodiment of the present application, and thus does not limit the scope of the patent application, and the equivalent structure or equivalent process transformation of the specification and the drawings of the present application, or directly or indirectly applied to other related technologies. The fields are all included in the scope of patent protection of this application.

Claims

A point cloud sampling method, which includes:

Obtain point cloud data of a three-dimensional image;

Calculating a curvature of each data point in the point cloud data;

Setting a sampling radius corresponding to each data point, wherein a sampling radius corresponding to a data point having a larger curvature is not larger than a sampling radius corresponding to a data point having a smaller curvature, and a sampling radius corresponding to at least a portion of the data point having a larger curvature is smaller than a smaller curvature The sampling radius corresponding to the data point;

The point cloud data is sampled based on the set sampling radius such that a region having a larger curvature has a data density greater than a region having a smaller curvature.
The method of claim 1, wherein the sampling the point cloud data based on the set sampling radius comprises:

Obtaining the position coordinates of the first data point;

Calculating a distance between the first data point and a second data point, the second data point being at least part of a data point other than the first data point;

Determining whether the distance is smaller than a sampling radius corresponding to the first data point;

If the judgment result is yes, deleting the second data point;

Otherwise, the second data point is reserved, another data point that is not deleted is selected as the first data point according to a preset sampling order, and the above steps are repeatedly performed until the traversal completes all the undeleted data points.
The method of claim 2 wherein said setting a sampling radius corresponding to each data point comprises:

The data points are arranged in a preset curvature order and then divided into a plurality of sub-point clouds;

A sampling radius is set for each of the sub-point clouds, wherein a sampling radius of the sub-point cloud having a larger curvature of the data point is smaller than a sampling radius of the sub-point cloud having a smaller curvature of the data point.
The method according to claim 3, wherein the dividing the data points into a plurality of sub-point clouds after being arranged in a preset order of curvature comprises:

Arranging the data points in increasing or decreasing order of curvature;

The aligned data points are equally divided into a plurality of the sub-point clouds.
The method of claim 3 wherein said setting a sampling radius for each of said sub-point clouds comprises:

The sampling radii of the sub-point cloud are sequentially set to a decreasing arithmetic progression with the first radius as an initial value in an order of increasing curvature; or

The sampling radii of the sub-point cloud are sequentially set to an increasing number of arithmetic progressions with the second radius being the initial value in the order of decreasing curvature.
The method of claim 3, wherein the predetermined sampling order is an increasing or decreasing order of curvature;

The obtaining location coordinates of the first data point includes:

Selecting a curvature minimum/maximum data point in the first sub-point cloud as the initial first data point, and acquiring position coordinates of the first data point;

The calculating the distance between the first data point and the second data point includes:

Calculating a first distance between the other data points in the first sub-point cloud and the first data point;

The determining whether the distance is smaller than a sampling radius corresponding to the first data point includes:

Determining whether the first distance is smaller than a sampling radius corresponding to the first sub-point cloud;

If the determination result is yes, determining that the distance is smaller than a sampling radius corresponding to the first data point;

The selecting, according to the preset sampling order, another data point that is not deleted as the first data point, and repeatedly performing the foregoing steps until traversing all the undeleted data points includes:

And selecting another data point that is not deleted in the first sub-point cloud as the first data point according to an order of increasing/decreasing curvature, and returning to performing the step of acquiring position coordinates of the first data point until traversing The data points that are not deleted by the first sub-point cloud are completed.
The method of claim 6, wherein after the traversing the data points that are not deleted by the first sub-point cloud, the method further comprises:

Select another sub-point cloud that is not sampled, and return to perform the step of acquiring the position coordinates of the first data point until the traversal completes all the sub-point clouds.
The method of claim 1 wherein said calculating a curvature of each of said point cloud data comprises:

Obtaining a neighborhood point of the first data point within a preset distance;

The eigenvalue method is used to fit the neighborhood point to the plane equation to obtain a feature matrix of the best fit plane;

Calculating a feature value of the feature matrix;

The ratio of the smallest feature value among the feature values to the sum of all the feature values is taken as the curvature of the first data point.
An image processing device, comprising: a communication circuit and a processor connected to each other;

The communication circuit is configured to acquire point cloud data of a three-dimensional image;

The processor is configured to calculate a curvature of each data point in the point cloud data; set a sampling radius corresponding to each data point, wherein a data radius corresponding to the data point corresponding to the curvature is not greater than a data point corresponding to the smaller curvature point a sampling radius, a sampling radius corresponding to at least a portion of the data point having a larger curvature is smaller than a sampling radius corresponding to the data point having a smaller curvature; and sampling the point cloud data based on the set sampling radius to make the region with a larger curvature The data density is larger than the area with less curvature.
The image processing device according to claim 9, wherein said processor is configured to:

Obtaining a position coordinate of the first data point; calculating a distance between the first data point and the second data point, the second data point being at least part of the data point except the first data point; determining whether the distance is And less than the sampling radius corresponding to the first data point; and when the determining result is yes, deleting the second data point; otherwise, retaining the second data point, selecting another undeleted according to a preset sampling order A data point is used as the first data point, and the above steps are repeated until the traversal completes all the undeleted data points.
The image processing device according to claim 10, wherein the processor is further configured to:

Arranging the data points in a preset curvature order and dividing into a plurality of sub-point clouds; setting a sampling radius for each of the sub-point clouds, wherein a sampling radius of the sub-point cloud having a larger curvature of the data points is smaller than The sampling radius of the sub-point cloud having a small curvature of the data point.
The image processing device according to claim 11, wherein the processor is further configured to:

The data points are arranged in an order of increasing/decreasing curvature; and the arranged data points are equally divided into a plurality of the sub-point clouds.
The image processing device according to claim 11, wherein the processor is further configured to:

The sampling radii of the sub-point cloud are sequentially set to a decreasing arithmetic sequence with the first radius as an initial value in an order of increasing curvature; or the sampling radii of the sub-point cloud are sequentially set to an order of decreasing curvature The incremental difference sequence is the initial value of the second radius.
The image processing device according to claim 11, wherein said predetermined sampling order is a curvature increasing/decreasing order; said processor is further configured to:

Selecting a curvature minimum/maximum data point in the first sub-point cloud as the initial first data point, and acquiring position coordinates of the first data point; calculating other data points in the first sub-point cloud and the a first distance of the first data point; determining whether the first distance is smaller than a sampling radius corresponding to the first sub-point cloud; and determining that the distance is smaller than the first data point when the determining result is yes Corresponding sampling radius; selecting another data point that is not deleted in the first sub-point cloud as the first data point in the order of increasing/decreasing curvature, and returning to perform performing the acquiring position coordinate of the first data point The step of traversing until the data point of the first sub-point cloud is not deleted.
The image processing device according to claim 14, wherein said processor is further configured to:

Select another sub-point cloud that is not sampled, and return to perform the step of acquiring the position coordinates of the first data point until the traversal completes all the sub-point clouds.
The image processing device according to claim 9, wherein said processor is configured to:

Obtaining a neighborhood point of the first data point within a preset distance; fitting the neighborhood point to a plane equation by using an eigenvalue method to obtain a feature matrix of the best fit plane; calculating a feature value of the feature matrix; The ratio of the smallest feature value among the feature values to the sum of all the feature values is taken as the curvature of the first data point.
A device having a storage function storing instructions, wherein the instructions are executed to implement the point cloud sampling method of any one of claims 1-8.