Grid-based point cloud clustering method, device and storage medium

Abstract

The application provides a grid-based point cloud clustering method, device and storage medium. The method comprises: acquiring point cloud data of an obstacle in a driving environment; projecting the point cloud data into a preset two-dimensional grid according to the pitch angle and the yaw angle of the point cloud data, and obtaining a projection point corresponding to the point cloud data; performing similarity clustering on the point cloud data corresponding to the projection point by using a breadth-first search method according to the grid similarity of the two-dimensional grid; and judging whether any two adjacent grids belong to the same row. If the two adjacent grids belong to the same row, a depth second derivative method is used to judge whether the two adjacent grids are similar. If the two adjacent grids are not similar, clustering is not performed. By not clustering the two adjacent grids which are not similar in the same row, the clustered point cloud can be ensured to be more consistent with the shape of the obstacle, and the situation that the point cloud occupies the driving space of the vehicle and causes the vehicle to be stuck can be avoided.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of computer vision, and more particularly to a grid-based point cloud clustering method, device and storage medium. BACKGROUND

[0002] Laser radar is an important sensor of an autonomous vehicle. Laser radar can form point cloud data, thereby helping the autonomous vehicle to perceive surrounding obstacles. In an autonomous driving scenario, in order to facilitate calculation, it is often necessary to model the obstacles as convex polygons, and accordingly, it is also necessary to convert the clustered point cloud clusters into convex polygons. However, the convex polygon formed by the traditional point cloud cluster may not be consistent with the shape of the convex polygon of the obstacle, which may cause the convex polygon formed by the point cloud cluster to be not fitted to the point cloud, to encroach on the driving space, and to cause the autonomous vehicle to be stalled, thereby threatening traffic safety. SUMMARY

[0003] The present application is proposed to solve the above problems. According to an aspect of the present application, a grid-based point cloud clustering method is provided, the method comprising:

[0004] obtaining point cloud data of an obstacle in a driving environment;

[0005] projecting the point cloud data into a preset two-dimensional grid according to the pitch angle and the yaw angle of the point cloud data, and obtaining a projection point corresponding to the point cloud data;

[0006] performing similarity clustering on the point cloud data corresponding to the projection point according to the grid similarity of the two-dimensional grid by using a breadth-first search method;

[0007] determining whether any two adjacent grids belong to the same row, and if so, determining whether the any two adjacent grids are similar by using a second-order derivative method, and if not, not performing clustering.

[0008] In an embodiment of the present application, performing similarity clustering on the point cloud data corresponding to the projection point according to the grid similarity of the two-dimensional grid by using a breadth-first search method comprises:

[0009] labeling each grid in the preset two-dimensional grid by using a breadth-first search method;

[0010] performing clustering on the point cloud data corresponding to the projection point in each grid of the same category to form a point cloud cluster.

[0011] In an embodiment of the present application, labeling each grid in the preset two-dimensional grid by using a breadth-first search method comprises:

[0012] The general criterion method is used to determine whether any two grids are similar, and when the any two grids are similar, the any two grids are marked as the same category.

[0013] In an embodiment of the present application, the general criterion method is used to determine whether any two grids are similar, comprising:

[0014] A first distance between representative points in the any two grids is calculated, and if the first distance is less than a first preset distance threshold, the any two grids are similar; and / or

[0015] A second distance and a third distance from the representative points in the any two grids to the radar center are respectively calculated, and if an absolute value of a difference between the second distance and the third distance is less than a second preset distance threshold, the any two grids are similar; and / or

[0016] An angle-based similarity determination: a farther one of the representative points in the any two grids is selected, a line connecting the farther one of the representative points and another representative point is taken as one side, and a line connecting the farther one of the representative points and the radar center is taken as another side to form an included angle, and if the included angle is greater than a preset angle threshold, the any two grids are similar.

[0017] In an embodiment of the present application, when the any two grids are similar, the any two grids are marked as the same category, comprising:

[0018] A first label is used to mark a category of any one grid in the preset two-dimensional grid;

[0019] The any one grid is taken as a center to diffuse to surrounding grids, and all grids similar to the any one grid are marked as the first label;

[0020] A second label is used to mark a category of any one grid in the preset two-dimensional grid which is not marked, and the any one grid not marked is taken as a center to diffuse to surrounding grids, and all grids similar to the any one grid not marked are marked as the second label;

[0021] The above steps are repeated until the grids are all marked.

[0022] In an embodiment of the present application, the any one grid is taken as a center to diffuse to surrounding grids, and all grids similar to the any one grid are marked as the first label, comprising:

[0023] The any one grid is marked as the first label;

[0024] The any one grid is added to a queue;

[0025] taking out a head grid of the queue, checking whether grids adjacent to the head grid in upper, lower, left and right directions are similar to the head grid respectively, if similar, marking the similar grids as the first label and adding the similar grids to the queue.

[0026] In an embodiment of the present application, the values of the first label and the second label are the same as the number of traversals.

[0027] In an embodiment of the present application, based on a depth second derivative method, it is determined whether the two adjacent grids are similar, and if not, clustering is not performed, including:

[0028] calculating a distance from a representative point of the projection point in each grid of the preset two-dimensional grid to the center of the laser radar to obtain a first matrix; wherein the first matrix records depth information of the each grid;

[0029] According to the first matrix, calculating a depth difference of the two adjacent grids to obtain a second matrix;

[0030] According to the second matrix, calculating a depth second derivative of the two adjacent grids to obtain a third matrix;

[0031] According to the coordinates of the two adjacent grids, determining that the two adjacent grids correspond to the depth second derivative in the third matrix, and when the depth second derivative corresponding to the two adjacent grids is less than zero, the two adjacent grids are not similar.

[0032] In an embodiment of the present application, it is determined whether the two adjacent grids belong to the same row, including:

[0033] determining whether the longitudinal coordinates of the two adjacent grids are the same, if the same, the two adjacent grids belong to the same row.

[0034] In an embodiment of the present application, the point cloud data is generated by a vehicle-mounted laser radar of an autonomous vehicle.

[0035] According to another aspect of the present application, a grid-based point cloud clustering device is provided, the device comprising:

[0036] a memory and a processor, the memory storing a computer program run by the processor, when the computer program is run by the processor, the computer program causes the processor to execute the aforementioned grid-based point cloud clustering method.

[0037] According to another aspect of the present application, a storage medium is provided, and the storage medium has a computer program stored thereon, which, when executed by a processor, causes the processor to perform the above-mentioned grid-based point cloud clustering method.

[0038] According to the grid-based point cloud clustering method, device and storage medium of the present application, by projecting the point cloud data into a preset two-dimensional grid, the point cloud data corresponding to the projected points is subjected to similarity clustering, and two adjacent grids in the same row that are not similar are not subjected to clustering, so that the clustered point cloud is more in line with the shape of the obstacle, and the situation that the point cloud occupies the vehicle driving space and causes the vehicle to be stuck is avoided. BRIEF DESCRIPTION OF DRAWINGS

[0039] The above and other objects, features and advantages of the present application will become more apparent from the following detailed description of embodiments of the present application taken in conjunction with the accompanying drawings. The accompanying drawings are provided to aid in understanding the embodiments of the present application, and constitute a part of the specification. The accompanying drawings together with the specification serve to explain the present application, and do not constitute a limitation on the present application. In the drawings, the same reference numerals generally represent the same components or steps.

[0040] Figure 1 A schematic diagram of an obstacle point cloud with ground point cloud generated according to a laser radar in a conventional technology is shown;

[0041] Figure 2 A schematic diagram of a convex polygon formed by a point cloud cluster generated according to a conventional technology is shown;

[0042] Figure 3 A schematic flowchart of a grid-based point cloud clustering method according to an embodiment of the present application is shown;

[0043] Figure 4 A schematic diagram of a two-dimensional grid of MxN dimensions according to an embodiment of the present application is shown;

[0044] Figure 5 A schematic diagram of a pitch angle and a yaw angle in a projection process according to an embodiment of the present application is shown;

[0045] Figure 6 A schematic diagram of a position relationship between adjacent networks and a radar center according to an embodiment of the present application is shown;

[0046] Figure 7 A schematic diagram of a convex polygon formed by a point cloud cluster generated according to an embodiment of the present application is shown;

[0047] Figure 8 A schematic block diagram of a grid-based point cloud clustering device according to an embodiment of the present application is shown. DETAILED DESCRIPTION

[0048] In order to make the purposes, technical solutions and advantages of the present application more obvious, the example embodiments according to the present application will be described in detail below with reference to the drawings. Obviously, the described embodiments are only a part of the embodiments of the present application, rather than all the embodiments of the present application, and it should be understood that the present application is not limited to the example embodiments described herein. Based on the embodiments of the present application described in the present application, all other embodiments obtained by those skilled in the art without creative labor should fall within the protection scope of the present application.

[0049] In an autonomous vehicle, a laser radar is usually provided. The laser radar is a very important sensor of the autonomous vehicle. The laser radar can form point cloud data, thereby helping the autonomous vehicle to perceive the surrounding obstacles. There are mainly two ways to generate obstacles using the laser radar: one is to identify the obstacles using a deep neural network model; the other is to first remove the ground point cloud to obtain non-ground point cloud, and then perform a clustering algorithm based on geometric rules on the non-ground point cloud, and take the clustering result as an obstacle (such as Figure 1 The autonomous vehicle often uses both methods at the same time to ensure the safety of driving by comprehensive consideration.

[0050] In addition, in the autonomous driving scene, in order to facilitate calculation, it is necessary to model the obstacle as a convex polygon. In specific applications, it is also necessary to convert the clustered point cloud cluster into a convex polygon. As shown in Figure 2 When the autonomous vehicle encounters a large-angle turn, the point cloud will hit the walls on both sides. At this time, using the clustering algorithm, the point cloud images corresponding to the walls on the left and right sides can be clustered into two groups of point clouds, and two convex polygon point clouds are generated. As can be seen, the internal point cloud of the convex polygon on the left side of the vehicle is a convex polygon, while the internal point cloud of the convex polygon on the right side of the vehicle does not fit the convex polygon, but encroaches on the driving space area, which may cause the autonomous vehicle to stall in some cases and cause traffic safety problems.

[0051] Based on the foregoing technical problems, the application provides a grid-based point cloud clustering method, which comprises the following steps: acquiring point cloud data of an obstacle in a driving environment; projecting the point cloud data into a preset two-dimensional grid according to the pitch angle and the yaw angle of the point cloud data, and obtaining a projection point corresponding to the point cloud data; performing similarity clustering on the point cloud data corresponding to the projection point by using a breadth-first search method according to the grid similarity of the two-dimensional grid; and judging whether any two adjacent grids belong to the same row, if yes, then judging whether the two adjacent grids are similar by using a depth second derivative method, and if not, then not performing clustering. By projecting the point cloud data into the preset two-dimensional grid, the point cloud data corresponding to the projection point is subjected to similarity clustering, and the two adjacent grids in the same row which are not similar are not subjected to clustering, so that the clustered point cloud can be ensured to be more in line with the shape of the obstacle, and the situation that the point cloud occupies the driving space of the vehicle and causes the vehicle to be stuck can be avoided.

[0052] The grid-based point cloud clustering method according to the embodiments of the application will be described in detail below with reference to the accompanying drawings. The features of various embodiments of the application can be combined with each other without conflict.

[0053] Figure 3 A schematic flowchart of the grid-based point cloud clustering method according to the embodiments of the application is shown. As shown in Figure 3 The grid-based point cloud clustering method 300 according to the embodiments of the application can comprise the following steps S301, S302, S303 and S304:

[0054] In step S301, the point cloud data of an obstacle in a driving environment is acquired.

[0055] The point cloud data is generated by a vehicle-mounted laser radar of an autonomous vehicle.

[0056] In step S302, the point cloud data is projected into a preset two-dimensional grid according to the pitch angle and the yaw angle of the point cloud data, and a projection point corresponding to the point cloud data is obtained.

[0057] For example, the point cloud data is projected into a two-dimensional grid with MxN dimensions according to the pitch angle and the yaw angle as shown in Figure 4 Each grid records all the projection points projected into the grid, so as to determine the similarity of the grid according to the projection points.

[0058] As shown in Figure 5 The pitch angle and the yaw angle are defined as follows: for a point p in the point cloud, the pitch angle is defined as θ p , and the yaw angle is defined as φ Let ∠xop' be the angle between the projection of p onto the xoy plane and the x-axis. The pitch angle ranges from -90° to 90°, and the yaw angle ranges from -180° to 180°.

[0059] In step S303, based on the mesh similarity of the two-dimensional mesh, the point cloud data corresponding to the projection points are clustered using the breadth-first search method.

[0060] In one embodiment of this application, based on the grid similarity of the two-dimensional grid, the point cloud data corresponding to the projection points are clustered using a breadth-first search method, including: A1, labeling each grid in the preset two-dimensional grid with a category using a breadth-first search method; A2, clustering the point cloud data corresponding to the projection points in each grid of the same category to form point cloud clusters.

[0061] In one embodiment of this application, the breadth-first search method is used to classify each grid in the preset two-dimensional grid, including: using a general criterion to determine whether any two grids are similar, and when any two grids are similar, then the any two grids are marked as the same category.

[0062] In one example, a general criterion is used to determine whether any two meshes are similar, including:

[0063] B1, calculate the first distance between representative points in any two grids; if the first distance is less than a first preset distance threshold, then the two grids are similar; and / or

[0064] B2, calculate the second and third distances from the representative points in any two grids to the radar center, respectively. If the absolute value of the difference between the second and third distances is less than a second preset distance threshold, then the two grids are similar; and / or

[0065] B3, similarity judgment based on angle: Select the more distant representative point among the representative points in any two grids, take the line connecting the more distant representative point and the other representative point as one side, and take the line connecting the more distant representative point and the radar center as the other side to form an angle. If the angle is greater than a preset angle threshold, then the two grids are similar.

[0066] For example, such as Figure 6 As shown, in step B1, similarity can be determined based on distance. First, the distance AB between representative points A and B of any two grids is calculated. If the distance AB is less than a first preset distance threshold, the two grids are considered similar.

[0067] Continue to combine Figure 6In step B2, the similarity can be determined based on the depth. The distances OA and OB of the representative points A and B of any two grids to the radar center are calculated respectively, and the difference between OA and OB is compared. If |OA-OB| is less than a second preset distance threshold, it is considered that the two grids are similar.

[0068] Continue to combine Figure 6 In step B3, the similarity can be determined based on the angle. The farther one of the representative points A and B of any two grids is selected, and the included angle formed by this point and the other two points is calculated, which corresponds to the angle β of ∠OAB in the above figure. If the included angle β is greater than a preset angle threshold, it is considered that the two grids are similar.

[0069] In the specific implementation process, one or a combination of multiple of B1, B2 and B3 can be selected for judgment.

[0070] In the above embodiment, the center point of the projection point in the grid can be used as the representative point, or the mean value of all projection points in the grid can be used as the representative point, and other feasible methods can also be used to determine the representative point.

[0071] In one example, when the two grids are similar, the two grids are marked as the same category, including: C1, using a first label to mark the category of any one grid in the preset two-dimensional grid; C2, using the any one grid as the center, diffusing to the surrounding grids, and marking all grids similar to the any one grid as the first label; C3, using a second label to mark the category of any one grid in the preset two-dimensional grid which has not been marked, and expanding to the surrounding grids from the any one grid which has not been marked, and marking all grids similar to the any one grid which has not been marked as the second label; C4, repeating the above steps until the grids are all marked.

[0072] The values of the first label and the second label are the same as the number of traversals. This is convenient for marking the grids while traversing.

[0073] In one specific embodiment, the label (maintaining state quantity) is used to represent the category of each grid. The initial value of the label of the grid can be set to 0, indicating that the grid has not been classified. The number of traversals is represented by index. When traversing for the first time, index = 1, and then for each grid in the two-dimensional grid, the following operations are performed in turn:

[0074] a) Take any one grid, if the label of the grid is label=0, it means that the grid is not classified. Let the label of the grid be label=index, use the breadth-first traversal method to spread the grid to the surrounding grid, and mark the label of all grids similar to the grid as index, to get a cluster result of label=index. Then perform the next traversal, let index=index+1, access the next grid, and repeat the above operation.

[0075] b) If the label of the next grid is label>0, it means that the grid has been classified, so skip the grid and access the next grid.

[0076] Repeat the above steps until all grids are labeled with labels. For each grid in the two-dimensional grid, perform the following operations in turn: add the point cloud data corresponding to the projection point contained in the grid to the point cloud cluster corresponding to the label of the grid, and finally generate the point cloud cluster as the clustering result of the point cloud data.

[0077] In an embodiment of the present application, the spreading of the grid to the surrounding grid includes: D1, marking the arbitrary grid as the first label; D2, adding the arbitrary grid to the queue; D3, taking out the head grid of the queue, checking whether the grids adjacent to the head grid in the upper, lower, left and right sides are similar to the head grid, and if similar, marking the similar grid as the first label and adding the similar grid to the queue.

[0078] In a specific example, for grid cell_p, the breadth-first traversal is used to spread its label, which is implemented by maintaining a queue queue. Through the above steps, it can be known that the category label of cell_p is label=index. Add cell_p to the queue queue, and when the queue queue is not empty, repeat the following operations: take out the head grid of the queue, denoted as cell_q; for the four adjacent grids (upper, lower, left and right) of cell_q, perform the following operations respectively:

[0079] a) If the label of the adjacent grid cell_k is label>0, it means that the adjacent grid cell_k has been classified, so skip the adjacent grid;

[0080] b) If label of the adjacent grid cell_k = 0, it means that the adjacent grid cell_k has not been classified, check whether the adjacent grid cell_k and the cell_q satisfy the similarity condition. If the similarity condition is satisfied, set the label of the cell_k = index, and add the cell_k to the queue.

[0081] In step S304, it is judged whether any two adjacent grids belong to the same row. If they belong to the same row, the deep second derivative method is used to judge whether the any two adjacent grids are similar. If they are not similar, clustering is not performed.

[0082] In one example, judging whether any two adjacent grids belong to the same row includes: judging whether the longitudinal coordinates of the any two adjacent grids are the same. If they are the same, the any two adjacent grids belong to the same row.

[0083] For example, when the grid cell_k and the grid cell_q belong to the same row of the grid matrix, the step of using the deep second derivative method to judge whether the any two adjacent grids are similar is forcibly performed.

[0084] In another example, judging whether the any two adjacent grids are similar based on the deep second derivative method, if they are not similar, clustering is not performed, includes: E1, calculating the distance from the representative point of the projection point in each grid of the preset two-dimensional grid to the center of the laser radar to obtain a first matrix; wherein the first matrix records the depth information of the each grid; E2, calculating the depth difference of the any two adjacent grids according to the first matrix to obtain a second matrix; E3, calculating the deep second derivative of the any two adjacent networks according to the second matrix to obtain a third matrix; E4, determining the deep second derivative corresponding to the any two adjacent grids in the third matrix according to the coordinates of the any two adjacent grids, when the deep second derivative corresponding to the any two adjacent grids is less than zero, the any two adjacent grids are not similar.

[0085] In a specific example, first, the distance from all representative points of the preset two-dimensional grid to the center of the radar is calculated. For example, an MxN two-dimensional grid can obtain an MxN first matrix D0, D0 records the depth information of each grid.

[0086] Secondly, the depth difference of adjacent grids is calculated based on D0, which can obtain an MxN matrix second D1, D1[i,j] = D0[i,(j+1) % N] - D0[i,j]. Here % represents the modulo operation, which is used to handle the boundary of the matrix.

[0087] Then, based on D1, the depth second derivative of adjacent grids is calculated to obtain the MxN dimensional third matrix D2, D2[i,j]=D1[i,(j+1)%N]–D1[i,j]. Here, % represents the modulo operation, which is still used to handle the matrix boundaries.

[0088] Specifically, the process for determining the second derivative of depth is as follows: For two neighboring grids cell_k and cell_q, assume their coordinates in the grid matrix are [ki,kj] and [qi,qj], respectively. Let i = ki,j = min(kj,qj). If D2[i,j] < 0, then cell_k and cell_q are considered dissimilar.

[0089] This application's embodiments introduce a depth second derivative to ensure that the clustered point cloud clusters are convex relative to the radar center, thereby ensuring that the convex polygons formed by the point cloud clusters do not encroach on the driving space.

[0090] In other words, when cell_k and cell_q belong to the same column of the grid matrix, if the above steps B1 and / or B2 and / or B3 are judged individually or in combination, then cell_k and cell_q are considered similar; otherwise, they are considered dissimilar.

[0091] When cell_k and cell_q belong to the same row of the grid matrix, if they simultaneously satisfy the individual or combined judgments of steps B1 and / or B2 and / or B3, and also satisfy the judgment that D2[i,j]>0 in steps E1 to E4, then cell_k and cell_q are considered similar; otherwise, they are considered dissimilar.

[0092] The grid-based point cloud clustering method according to the embodiments of this application can obtain, as follows: Figure 6 The point cloud clusters shown, such as Figure 7 As shown, the convex polygons formed by the point cloud clusters fit the obstacles well and do not encroach on the driving space. The similarity clustering algorithm involved in this application adds an additional judgment condition for judging geometric similarity on the basis of traditional similarity clustering algorithms. This can ensure that the clustered point cloud clusters are convex relative to the radar center, thereby ensuring that the convex polygons formed by the point cloud clusters do not encroach on the driving space.

[0093] This application embodiment projects point cloud data onto a preset two-dimensional grid, and performs similarity clustering on the point cloud data corresponding to the projection points. Two dissimilar adjacent grids in the same row are not clustered. This ensures that the clustered point cloud fits the shape of the obstacle better and avoids the point cloud encroaching on the vehicle's driving space and causing the vehicle to get stuck.

[0094] The following is combined with Figure 8 The grid-based point cloud clustering device of this application is described, wherein, Figure 8A schematic block diagram of a grid-based point cloud clustering apparatus according to an embodiment of the present application is shown.

[0095] As shown in Figure 8 , the grid-based point cloud clustering apparatus 800 comprises one or more memories 801 and one or more processors 802, the memories 801 have computer programs stored thereon which are run by the processors 802, and the computer programs, when run by the processors 802, cause the processors 802 to perform the grid-based point cloud clustering method described above.

[0096] The apparatus 800 can be part or all of a computer device which can implement the grid-based point cloud clustering method by software, hardware or a combination of software and hardware.

[0097] As shown in Figure 8 , the apparatus 800 comprises one or more memories 801, one or more processors 802, a display (not shown) and a communication interface, etc., and these components are interconnected by a bus system and / or other forms of connection mechanism (not shown). It should be noted that Figure 8 The components and structures of the apparatus 800 shown are only exemplary and are not limiting, and the apparatus 800 can also have other components and structures as needed.

[0098] The memory 801 is used to store various data generated during the operation of the train and executable program instructions, for example, to store various application programs or algorithms for implementing various specific functions. It can include one or more computer program products, which can include various forms of computer readable storage media, such as volatile memory and / or non-volatile memory. The volatile memory may, for example, include random access memory (RAM) and / or cache memory, etc. The non-volatile memory may, for example, include read-only memory (ROM), hard disk, flash memory, etc.

[0099] The processor 802 can be a central processing unit (CPU), a graphics processing unit (GPU), an application-specific integrated circuit (ASIC), a field-programmable gate array (FPGA), or other forms of processing units having data processing and / or instruction execution capabilities, and can be connected with other components in the apparatus 800 to perform desired functions.

[0100] In one example, the apparatus 800 further comprises an output apparatus which can output various information (such as images or sounds) to the outside (such as a user), and can include one or more of a display device, a speaker, etc.

[0101] The communication interface can be an interface of any communication protocol known at present, such as a wired interface or a wireless interface, wherein the communication interface can include one or more serial ports, USB interfaces, Ethernet ports, WiFi, wired networks, DVI interfaces, device integrated interconnection modules or other suitable various ports, interfaces or connections.

[0102] In addition, according to the embodiments of the present application, a storage medium is also provided, in which program instructions are stored, and when the program instructions are run by a computer or a processor, the corresponding steps of the grid-based point cloud clustering method according to the embodiments of the present application are executed. The storage medium can include, for example, a memory card of a smart phone, a storage component of a tablet computer, a hard disk of a personal computer, a read-only memory (ROM), an erasable programmable read-only memory (EPROM), a portable compact disc read-only memory (CD-ROM), a USB memory, or any combination of the above storage media.

[0103] The grid-based point cloud clustering device and the storage medium according to the embodiments of the present application have the same advantages as the aforementioned grid-based point cloud clustering method, because the grid-based point cloud clustering method can be implemented.

[0104] Although the example embodiments have been described herein with reference to the accompanying drawings, it is to be understood that the example embodiments are merely exemplary and are not intended to limit the scope of the present application. Those of ordinary skill in the art can make various changes and modifications without departing from the scope and spirit of the present application. All such changes and modifications are intended to be included within the scope of the present application as claimed in the appended claims.

[0105] Those of ordinary skill in the art can realize that the units and algorithm steps of each example described in connection with the embodiments disclosed herein can be realized by electronic hardware, or a combination of computer software and electronic hardware. Whether the functions are executed in hardware or software depends on the specific application and design constraints of the technical solution. Those of ordinary skill in the art can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of the present application.

[0106] In several embodiments provided in the present application, it should be understood that the disclosed devices and methods can be implemented in other ways. For example, the device embodiments described above are merely illustrative, for example, the division of the units is only a logical function division, and actual implementation can have another division manner, for example, multiple units or components can be combined or integrated into another device, or some features can be ignored or not executed.

[0107] In the description provided herein, numerous specific details are set forth. However, it is understood that embodiments of the application can be practiced without these specific details. In some instances, well-known methods, structures and techniques have not been shown in detail in order not to obscure an understanding of this description.

[0108] Similarly, it is to be understood that the various features of the application can sometimes be used to advantage together, but each of the features can be used independently of one another. Furthermore, various embodiments of the present application have been described herein in terms of processes, procedures, and techniques. It should be understood that the embodiments of the application can be implemented in software, firmware, hardware, or any combination thereof. It should also be understood that the various features of the application can be used in any combination or sub-combination thereof.

[0109] Those skilled in the art will appreciate that all features described herein (including all accompanying claims, abstract and drawings), and steps of any method or procedure so disclosed, can be combined in any combination, except where features are mutually exclusive (to the extent such features are mutually exclusive), as understood by those of ordinary skill in the art. Each of the features of the application disclosed in the specification (including the claims, abstract and drawings) can be replaced by alternative features serving the same, equivalent or similar purpose, unless expressly stated otherwise.

[0110] Furthermore, those skilled in the art will recognize that references in the specification to "one embodiment", "an embodiment", "an example embodiment", etc., mean that a particular feature, structure, or characteristic described in connection with the embodiment is included in at least one embodiment of the application. The appearances of such phrases in various places in the specification are not necessarily all referring to the same embodiment. Further, where a particular feature, structure, or characteristic is stated to have been added in an embodiment, it will be understood that to the extent possible, this particular feature, structure or characteristic has been added to that embodiment but not necessarily to all other embodiments.

[0111] Embodiments of various components of the application can be implemented in hardware, software, or a combination thereof. Those skilled in the art will appreciate that some or all of the functionality of some of the modules according to embodiments of the application can be implemented using a microprocessor or a digital signal processor (DSP) in practice. The application can also be implemented as a program (e.g., a computer program and computer program product) for performing part or all of the methods described herein. Such program(s) can be stored on a computer readable medium, or can be transmitted over a network. Such a network can be the Internet, or any other form of network.

[0112] It should be noted that the above-mentioned embodiments illustrate rather than limit the application, and that one skilled in the art will be able to design many alternative embodiments without departing from the scope of the appended claims. In the claims, any reference signs placed between parentheses shall not be construed as limiting the claim. The word "comprising" does not exclude the presence of elements or steps other than those listed in a claim. The word "a" or "an" preceding an element does not exclude the presence of a plurality of such elements. The application can be implemented by means of both hardware and software, and any combination thereof. In a unitary claim, several devices or sub-claims can be joined by means of the expression "and / or". The use of the term "at least" followed by a list of one or more items should be interpreted as including at least one of the items but it does not exclude the presence of others not listed. The use of the term "one" followed by a list of one or more items should be interpreted as including at least one of the items but it does not exclude the presence of others not listed. It is emphasized that the terms "comprises / comprising" when used in this specification are taken to specify the presence of stated features, integers, steps or components but do not preclude the presence or addition of one or more other features, integers, steps, components or groups thereof.

[0113] The above description is only specific embodiments of the present application or specific explanations of specific embodiments. The protection scope of the present application is not limited to this, and any person skilled in the art can easily think of changes or replacements within the technical range disclosed in the present application, which should be covered in the protection scope of the present application. The protection scope of the present application should be subject to the protection scope of the claims.

Claims

1. A grid-based point cloud clustering method, characterized in that, The method includes: Acquire point cloud data of obstacles in the driving environment; Based on the pitch and yaw angles of the point cloud data, the point cloud data is projected onto a preset two-dimensional grid to obtain the projection points corresponding to the point cloud data. Based on the grid similarity of the two-dimensional grid, the point cloud data corresponding to the projection points are clustered using the breadth-first search method. Determine whether any two adjacent grids belong to the same row. If they do, use the depthwise second derivative method to determine whether the two adjacent grids are similar. If they are not similar, do not perform clustering.

2. The method of claim 1, wherein, Based on the grid similarity of the two-dimensional grid, the point cloud data corresponding to the projection points are clustered using a breadth-first search method, including: Each grid in the preset two-dimensional grid is categorized using a breadth-first search method. The point cloud data corresponding to the projection points in each grid of the same category are clustered to form point cloud clusters.

3. The method of claim 2, wherein, The breadth-first search algorithm is used to classify each grid in the preset two-dimensional grid, including: A general criterion is used to determine whether any two grids are similar. If any two grids are similar, they are marked as belonging to the same category.

4. The method of claim 3, wherein, A general criterion is used to determine whether any two grids are similar, including: Calculate the first distance between representative points in any two grids; if the first distance is less than a first preset distance threshold, then the two grids are similar; and / or Calculate the second and third distances from the representative points in any two grids to the radar center. If the absolute value of the difference between the second and third distances is less than a second preset distance threshold, then the two grids are similar; and / or Similarity judgment based on angle: Select the more distant representative point among the representative points in any two grids, take the line connecting the more distant representative point and the other representative point as one side, and take the line connecting the more distant representative point and the radar center as the other side to form an angle. If the angle is greater than a preset angle threshold, then the two grids are similar.

5. The method of claim 3, wherein, When any two grids are similar, they are labeled as belonging to the same category, including: Use the first label to mark the category of any one of the preset two-dimensional grids; Taking any one of the grids as the center, the process expands outwards to the surrounding grids, and marks all grids similar to any one of the grids as the first label; The category of any unlabeled grid in the preset two-dimensional grid is marked by the second label, and the unlabeled grid is used to expand to the surrounding grids, and all grids similar to the unlabeled grid are marked with the second label. Repeat the above steps until all the grids have been marked.

6. The method of claim 5, wherein, Centered on any one of the grids, the process expands outwards to surrounding grids, marking all grids similar to that grid as the first label, including: Mark any one of the grids as the first label; Add any one of the grids to the queue; Take out the head grid of the queue, and check whether the grids above, below, to the left and to the right of the head grid are similar to the head grid. If they are similar, mark the similar grids as the first label and add the similar grids to the queue.

7. The method of claim 5, wherein, The values ​​of the first label and the second label are the same as the number of times they are traversed.

8. The method of claim 1, wherein, Based on the deep second-order derivative method, it is determined whether any two adjacent grids are similar. If they are not similar, no clustering is performed, including: The distance from the representative point of the projection point in each grid of the preset two-dimensional grid to the center of the lidar is calculated to obtain a first matrix; wherein the first matrix records the depth information of each grid. Based on the first matrix, calculate the depth difference between any two adjacent grids to obtain the second matrix; The third matrix is ​​obtained by calculating the depth-first second derivative of any two adjacent networks based on the second matrix. The second depth derivative of any two adjacent grids is determined based on their coordinates in the third matrix. If the second depth derivative of any two adjacent grids is less than zero, then the two adjacent grids are not similar.

9. The method of claim 1, wherein, Determine whether any two adjacent grid cells belong to the same row, including: Determine whether the ordinates of any two adjacent grids are the same. If they are the same, then the two adjacent grids belong to the same row.

10. The method of claim 1, wherein, in, The point cloud data is generated by the onboard LiDAR of the autonomous vehicle.

11. A mesh-based point cloud clustering apparatus, characterized by, The device includes: A memory and a processor, wherein the memory stores a computer program that is executed by the processor, the computer program, when executed by the processor, causes the processor to perform the grid-based point cloud clustering method as described in any one of claims 1 to 10.

12. A storage medium, characterized by The storage medium stores a computer program that, when executed by a processor, causes the processor to perform the grid-based point cloud clustering method as described in any one of claims 1 to 10.

Images