Global initialization positioning method, device and autonomous driving vehicle

Through Gaussian hybrid model and multi-resolution branch definition algorithm, GPU is used to perform parallel operations, which solves the problems of low global initialization positioning efficiency and high cost in the prior art, and achieves fast and high-precision positioning in special scenarios.

CN115265558BActive Publication Date: 2024-12-03JIUZHI (SUZHOU) INTELLIGENT TECH CO LTD
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202210974320.1
Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
Filing Date
2022-08-15
Publication Date
2024-12-03
Estimated Expiration
2042-08-15

AI Technical Summary

Technical Problem

The prior art has low efficiency and high cost in special scenarios. The deep learning-based method is too expensive, and the calculation time is increased when searching in large-scale areas.

Method used

Using Gaussian hybrid model and multi-resolution branch definition algorithm, parallel computing is implemented through GPU, 3D point cloud data is compressed into a two-dimensional grid structure, and parallel branch calculations are performed to quickly obtain candidate solution scores and reduce calculation time.

Benefits of technology

Realizing high-precision global initialization positioning at low cost and in a short period of time reduces the dependence on high-precision inertial navigation and deep learning, and improves positioning efficiency in special scenarios.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN115265558B_ABST
    Figure CN115265558B_ABST
Patent Text Reader

Abstract

A global initialization positioning method, device and autonomous driving vehicle, the method is used for a graphics processing unit, the method includes: obtaining at least two Gaussian mixture model maps of different resolutions, the Gaussian mixture model map is obtained by compressing three-dimensional point cloud data into a two-dimensional grid structure, each grid in the two-dimensional grid structure maintains a Gaussian mixture model, and the Gaussian mixture model is used to characterize the height distribution of the point cloud in the grid; obtaining point cloud data collected by a laser radar; based on the point cloud data, branching calculations are performed in parallel on the registration scores of multiple candidate solutions under the at least two Gaussian mixture model maps, and when the highest score is the score of the candidate solution under the Gaussian mixture model map with the highest resolution, the candidate solution corresponding to the highest score is output as the initial positioning result. The global initialization positioning method of the present invention can improve the positioning speed.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present invention relates to the field of autonomous driving vehicles, and more specifically, to a global initialization positioning method, device and autonomous driving vehicle. Background Art

[0002] There are usually several solutions to achieve high-precision global initialization positioning: one is global initialization positioning based on high-precision inertial navigation; the second is a global positioning initialization solution based on deep learning; the third is a matching solution based on high-precision maps.

[0003] The global positioning initialization scheme relying on high-precision inertial navigation will fail due to large errors in some special scenarios (such as occlusion, tall buildings, etc.), which are often encountered in practice. At the same time, high-precision inertial navigation also needs to stay in an open scene for a certain period of time to initialize and then enter the combined navigation state, which is inefficient. At the same time, the cost of high-precision inertial navigation is very expensive.

[0004] The global positioning method based on deep learning needs to collect the surrounding scene information of the vehicle at different positions and the current accurate position of the vehicle, and then train the model. This process requires a lot of cost to achieve. At the same time, due to the various obstacles that may exist in open scenes, higher requirements are placed on the accuracy of the model. For global initialization positioning, the cost of this solution is too high.

[0005] Matching solutions based on high-precision maps often use the currently perceived environmental information and high-precision maps for matching, thereby achieving global initial positioning. The general implementation methods are image-based or lidar-based methods. Image-based methods often do not work well in outdoor scenes due to changes in light; although lidar-based methods are not affected by lighting, they often require a relatively accurate initial value to search for high-precision global positioning. If a large range of search conditions are used, the calculation time will increase exponentially. Summary of the invention

[0006] A series of simplified concepts are introduced in the Summary of the Invention, which will be further described in detail in the Detailed Description of the Invention. The Summary of the Invention does not mean to attempt to define the key features and essential technical features of the claimed technical solution, nor does it mean to attempt to determine the scope of protection of the claimed technical solution.

[0007] In view of the deficiencies of the prior art, a first aspect of an embodiment of the present invention proposes a global initialization positioning method, which is used for a graphics processing unit and includes:

[0008] Acquire at least two Gaussian mixture model maps of different resolutions, wherein the Gaussian mixture model map is obtained by compressing three-dimensional point cloud data into a two-dimensional grid structure, wherein each grid in the two-dimensional grid structure maintains a Gaussian mixture model, and the Gaussian mixture model is used to characterize the height distribution of the point cloud in the grid;

[0009] Get point cloud data collected by LiDAR;

[0010] Based on the point cloud data, branch calculations are performed in parallel on the alignment scores of multiple candidate solutions under the at least two Gaussian mixture model maps, and when the highest score is the score of the candidate solution under the Gaussian mixture model map with the highest resolution, the candidate solution corresponding to the highest score is output as the initial positioning result.

[0011] In some embodiments, the branching calculation of the registration scores of the plurality of candidate solutions under the at least two Gaussian mixture model maps in parallel comprises:

[0012] By means of the graphics processing unit, the scores of the plurality of candidate solutions under the Gaussian mixture model map with the lowest resolution are calculated in parallel, and the candidate solutions are sorted according to the scores of the candidate solutions to obtain a candidate solution score queue;

[0013] Selecting N candidate solutions with the highest scores from the candidate solution score queue, calculating their scores under the next layer of Gaussian mixture model map, and adding them to the score queue according to the scores;

[0014] The scores of the N candidate solutions with the highest scores in the score queue under the Gaussian mixture model map of the next layer are repeatedly calculated until the highest score in the score queue is the score of the candidate solution under the Gaussian mixture model map with the highest resolution.

[0015] In some embodiments, the candidate solutions include the range of the horizontal and vertical coordinates and the range of the yaw angle of the laser radar.

[0016] In some embodiments, the method further comprises:

[0017] Get the 3D point cloud map collected by LiDAR;

[0018] Compressing the three-dimensional point cloud map into a two-dimensional grid structure in parallel to obtain the Gaussian mixture model map with the highest resolution, wherein each grid in the two-dimensional grid structure maintains a Gaussian mixture model, and the Gaussian mixture model is used to characterize the height distribution of the point cloud in the grid;

[0019] Calculating the upper bound of the joint distribution of each adjacent plurality of grids in the two-dimensional grid structure to obtain a Gaussian mixture model map with reduced resolution;

[0020] Repeat the calculation of the joint distribution upper bound of each of the plurality of adjacent grids in each of the Gaussian mixture model maps until a Gaussian mixture model map with a predetermined number of layers is obtained.

[0021] In some embodiments, compressing the three-dimensional point cloud map into a two-dimensional grid structure in parallel includes:

[0022] The mean and variance of the multi-core Gaussian distribution in each grid in the two-dimensional grid structure are estimated in parallel by using the multiple parallel computing units of the graphics processing unit.

[0023] In some embodiments, the calculating the upper bound of the joint distribution of each adjacent plurality of grids in the two-dimensional grid structure comprises:

[0024] The two-dimensional grid structure is subjected to sliding processing through a sliding window of a preset size, and the upper bound of the joint distribution of all grids currently covered by the sliding window is calculated.

[0025] In some embodiments, the method is used in an autonomous driving vehicle, and the graphics processing unit is an on-board graphics processing unit.

[0026] A second aspect of an embodiment of the present invention provides a global initialization positioning device, which includes a memory and a graphics processing unit. The memory stores a computer program run by the graphics processing unit, and the computer program executes the global initialization positioning method as described above when run by the graphics processing unit.

[0027] A third aspect of an embodiment of the present invention provides an autonomous driving vehicle, the autonomous driving vehicle comprising:

[0028] Car body;

[0029] A laser radar mounted on the vehicle body, the laser radar being used to collect point cloud data;

[0030] The global initialization positioning device as described above is connected to the laser radar and is used to execute the global initialization positioning method as described above.

[0031] A fourth aspect of an embodiment of the present invention provides a storage medium, on which a computer program is stored, and the computer program executes the global initialization positioning method as described above when running.

[0032] The global initialization positioning method, device, autonomous driving vehicle and storage medium of the embodiments of the present invention implement parallel computing based on a graphics processing unit during the positioning process, greatly improving the speed of initialization positioning. BRIEF DESCRIPTION OF THE DRAWINGS

[0033] The above and other purposes, features and advantages of the present invention will become more apparent by describing the embodiments of the present invention in more detail in conjunction with the accompanying drawings. The accompanying drawings are used to provide a further understanding of the embodiments of the present invention and constitute a part of the specification. Together with the embodiments of the present invention, they are used to explain the present invention and do not constitute a limitation of the present invention. In the accompanying drawings, the same reference numerals generally represent the same components or steps.

[0034] Figure 1 is a schematic flow chart of a global initialization positioning method according to an embodiment of the present invention;

[0035] Figure 2 is a schematic block diagram of a global initialization positioning device according to an embodiment of the present invention;

[0036] Figure 3 A schematic block diagram of an autonomous driving vehicle according to an embodiment of the present invention. DETAILED DESCRIPTION

[0037] In order to make the purpose, technical solutions and advantages of the present application more obvious, the example embodiments according to the present application will be described in detail with reference to the accompanying drawings. Obviously, the described embodiments are only 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 work should fall within the scope of protection of the present application.

[0038] In the following description, a large number of specific details are provided to provide a more thorough understanding of the present application. However, it is apparent to those skilled in the art that the present application can be implemented without one or more of these details. In other examples, in order to avoid confusion with the present application, some technical features well known in the art are not described.

[0039] It should be understood that the present application can be implemented in different forms and should not be interpreted as being limited to the embodiments set forth herein. On the contrary, providing these embodiments will make the disclosure thorough and complete, and will fully convey the scope of the present application to those skilled in the art.

[0040] The purpose of the terms used herein is only to describe specific embodiments and is not intended to be limiting of the present application. When used herein, the singular forms "one", "an" and "said / the" are also intended to include plural forms, unless the context clearly indicates otherwise. It should also be understood that the terms "consisting of" and / or "comprising", when used in this specification, determine the presence of the features, integers, steps, operations, elements and / or parts, but do not exclude the presence or addition of one or more other features, integers, steps, operations, elements, parts and / or groups. When used herein, the term "and / or" includes any and all combinations of the relevant listed items.

[0041] In order to thoroughly understand the present application, a detailed structure will be proposed in the following description to illustrate the technical solution proposed by the present application. The optional embodiments of the present application are described in detail as follows, but in addition to these detailed descriptions, the present application may also have other implementation methods.

[0042] The main purpose of the embodiment of the present invention is to achieve global high-precision initialization positioning in special scenarios at a relatively low cost (computing resources, time cost). Although the time consumption of point cloud matching based on laser radar and high-precision map will increase exponentially when searching over a large range, GPU high parallel acceleration can be used to reduce the time consumption. Therefore, the embodiment of the present invention proposes a large-scale search global initialization positioning method based on GPU to achieve the matching of point cloud collected by laser radar with high-precision point cloud map. The embodiment of the present invention mainly adopts Gaussian mixture model and multi-resolution branch and bound algorithm, and realizes map compression and high-concurrency acceleration of multi-resolution branch and bound algorithm through GPU, thereby realizing fast calculation under large-scale search.

[0043] See first Figure 1 , Figure 1 The schematic flowchart of the global initialization positioning method 100 according to an embodiment of the present invention is shown. The global initialization positioning method 100 of the embodiment of the present invention can be used for an autonomous driving vehicle, which can also be called an unmanned vehicle. It is an intelligent vehicle that does not require a driver to perform driving operations and can automatically complete vehicle driving tasks instead of the driver. The autonomous driving vehicle is equipped with a laser radar for collecting point cloud data to locate the autonomous driving vehicle. Figure 1 As shown, the global initialization positioning method 100 of the embodiment of the present invention includes the following steps:

[0044] In step S110, at least two Gaussian mixture model maps of different resolutions are obtained, wherein the Gaussian mixture model map is obtained by compressing the three-dimensional point cloud data into a two-dimensional grid structure, wherein each grid in the two-dimensional grid structure maintains a Gaussian mixture model, and the Gaussian mixture model is used to characterize the height distribution of the point cloud in the grid;

[0045] In step S120, point cloud data collected by the laser radar is obtained;

[0046] In step S130, based on the point cloud data, branch calculations are performed in parallel on the registration scores of multiple candidate solutions under the at least two Gaussian mixture model maps. When the highest score is the score of the candidate solution under the Gaussian mixture model map with the highest resolution, the candidate solution corresponding to the highest score is output as the initial positioning result.

[0047] The global initialization positioning method 100 of the embodiment of the present invention is implemented in a graphics processing unit (GPU). Further, the global initialization positioning method 100 of the embodiment of the present invention can be implemented in a graphics processing unit on board an unmanned vehicle. The on-board graphics processing unit is connected to the laser radar of the vehicle by wire or wirelessly to obtain the point cloud data collected by the laser radar. The unique hardware structure of the GPU enables it to support large-scale parallel computing and can quickly achieve global initialization positioning.

[0048] Specifically, in step S110, at least two Gaussian mixture model maps of different resolutions are obtained, and the at least two Gaussian mixture model maps of different resolutions are referred to as multi-resolution Gaussian mixture model maps. The Gaussian mixture model map is obtained by compressing the three-dimensional point cloud data into a two-dimensional grid structure, and each grid in the two-dimensional grid structure maintains a Gaussian mixture model, and the Gaussian mixture model is used to characterize the height distribution of the point cloud in the grid.

[0049] The multi-resolution Gaussian mixture model map can be pre-built offline. When building the multi-resolution Gaussian mixture model map, first obtain a high-precision three-dimensional point cloud map collected by the laser radar. The three-dimensional point cloud map has richer and more accurate spatial information and can achieve accurate positioning of the vehicle.

[0050] After acquiring the three-dimensional point cloud map, the three-dimensional point cloud map is compressed into a two-dimensional grid structure at a certain resolution in parallel through the graphics processing unit to obtain a Gaussian mixture model map with the highest resolution. Each grid in the two-dimensional grid structure maintains a Gaussian mixture model, and the Gaussian mixture model in each grid is used to characterize the height distribution of the point cloud in the grid; the Gaussian mixture model uses the Gaussian probability density function (normal distribution curve) to accurately quantify things and decompose things into several models based on the Gaussian probability density function (normal distribution curve). Exemplarily, the maximum expectation algorithm (EM algorithm) can be used to estimate the mean and variance of the multi-core Gaussian distribution of the point cloud in each grid.

[0051] Since each grid is calculated independently, if the time consumption of using the CPU to serially calculate the Gaussian mixture model of each grid is t, then for all n grids, the total calculation time consumption is about n*t. In comparison, the embodiment of the present invention uses the GPU to perform high-concurrency calculations, and estimates the mean and variance of the multi-core Gaussian distribution in each grid in the two-dimensional grid structure in parallel through multiple parallel computing units of the graphics processing unit. Assuming that the number of concurrent threads that can be used at the same time is m, each thread can independently calculate the Gaussian mixture model for the grid, so the overall time consumption is about n*t / m. Due to the advantages of the GPU under high concurrency, the value of m is usually large, so the compression time consumption of the Gaussian mixture model map is significantly reduced.

[0052] After the above compression steps, a Gaussian mixture model map with a resolution of r is obtained, which is the Gaussian mixture model map with the highest resolution. In order to implement the subsequent branch and bound algorithm and speed up the search speed, on the basis of the Gaussian mixture model map with a resolution of r, the upper bound of the joint distribution of each adjacent multiple grids in the two-dimensional grid structure is calculated to obtain a Gaussian mixture model map with reduced resolution. Specifically, the two-dimensional grid structure can be slid by a sliding window of a preset size (for example, a 2*2 sliding window), and the upper bound of the joint distribution of all grids currently covered by the sliding window is calculated. The sliding window traverses the current Gaussian mixture model map with a step size of resolution r to generate a Gaussian mixture model map with reduced resolution.

[0053] According to the above idea, the upper bounds of the joint distribution of all grids currently covered by the sliding window are calculated independently, so the GPU can be used for high-concurrency acceleration. The sliding window operation is performed simultaneously by different computing units of the GPU, which greatly shortens the calculation time. After that, the upper bounds of the joint distribution of each adjacent multiple grids in each Gaussian mixture model map are repeatedly calculated, that is, the sliding window operation is performed on the basis of each Gaussian mixture model map until a Gaussian mixture model map with a predetermined number of layers is obtained. The embodiment of the present invention does not limit the size of the sliding window and the specific number of layers of the map.

[0054] In step S120, the point cloud data collected by the laser radar is obtained. For example, the laser radar can be a laser radar with regular repeated scanning or a laser radar with complex scanning trajectory and non-repetitive scanning characteristics. The radar is used to sense environmental information outside the vehicle, such as distance information, orientation information, reflection intensity information, speed information, etc. of environmental targets. The point cloud data includes at least distance information and orientation information.

[0055] In one implementation, the laser radar can detect the distance from the detection object to the laser radar by measuring the time of light propagation between the laser radar and the detection object, that is, the light flight time (TOF). Alternatively, the laser radar can also detect the distance from the detection object to the laser radar through other technologies, such as a ranging method based on phase shift measurement, or a ranging method based on frequency shift measurement, etc., which is not limited in the embodiment of the present invention.

[0056] In step S130, based on the point cloud data, branch calculations are performed in parallel on the registration scores of multiple candidate solutions under the multi-resolution Gaussian mixture model map. The candidate solutions come from a preset solution space within a certain range, which is a set of candidate solutions k within a certain range near the predicted value, expressed as S = [xi, yi, yawi], i = 0, ..., k. For example, a single-point GPS with a large error can be used as the predicted value to reduce the reliance on high-precision inertial navigation and reduce hardware costs.

[0057] In addition, since the tilt and pitch generated by the autonomous vehicle are small, the candidate solutions include the range of the horizontal and vertical coordinates and the range of the yaw angle of the lidar, which represent the range of the horizontal and vertical coordinates and the range of the yaw angle of the autonomous vehicle. In the point cloud matching process, the yaw angle can be first applied to the point cloud points, and then translated according to the range of the horizontal and vertical coordinates. The final global initialization positioning result is the global three-dimensional pose [x, y, yaw], where x and y represent translation, and yaw represents the yaw angle.

[0058] During the search process, in order to evaluate the current candidate solution, it is necessary to evaluate the log-likelihood of each point in the point cloud data, and then sum the log-likelihoods of all points to obtain the total log-likelihood. Specifically, the candidate solution can be applied to each point in the point cloud data and indexed into the Gaussian mixture model map to obtain the Gaussian mixture of each grid. On this basis, each point is regarded as an independent Gaussian observation, and the total log-likelihood is calculated by summing the log-likelihood of each point as the score of the current candidate solution. Since each grid in the Gaussian mixture model map is very similar in space, this means that the inference of adjacent candidate solutions will produce similar log-likelihoods. If an exhaustive search method is used, unnecessary time will usually be consumed in low-likelihood areas. To this end, the embodiment of the present invention adopts a branch-bounded search method, performs an exhaustive search on the Gaussian mixture model map with the lowest resolution, and performs a layer-by-layer branch search on the candidate solutions with higher scores, thereby shortening the search time.

[0059] Specifically, the GPU first calculates the scores of multiple candidate solutions in the Gaussian mixture model map with the lowest resolution (i.e., the k-th layer map) in parallel, and sorts the candidate solutions according to their scores to obtain a candidate solution score queue. The higher the score, the higher the position in the queue. Since the GPU can calculate the scores of different candidate solutions in parallel, the speed of score calculation is greatly shortened.

[0060] After that, the scores are further branched, and the N candidate solutions with the highest scores in the candidate solution score queue are selected, and their scores in the next layer of Gaussian mixture model map (i.e., the k-1 layer map) are calculated. They are added to the score queue according to the scores, sorted together with the remaining candidate solutions in the score queue, and the N candidate solutions with the highest scores are selected again in the candidate solution score queue to calculate their scores in the k-2 layer map. The scores of the N candidate solutions with the highest scores in the score queue under the next layer of Gaussian mixture model map are calculated repeatedly in this way until the highest score in the score queue is the score of the candidate solution under the Gaussian mixture model map with the highest resolution. At this time, the candidate solution with the highest score is the high-precision global initialization positioning result.

[0061] Based on the above description, it can be seen that as the scope of the search expands, the number of candidate solutions increases exponentially. If the CPU is used for serial calculation, the time consumption will also increase exponentially. Therefore, in the step of calculating the scores of the candidate solutions, the embodiment of the present invention uses the GPU for high concurrency acceleration, and the scores of all candidate solutions can be calculated in a very short time and sorted. At the same time, when selecting the N candidate solutions with the highest scores for the next calculation, the multi-concurrency of the GPU can also be relied on to select the N candidate solutions with the highest scores at one time for further calculation, so as to quickly solve to the optimal solution.

[0062] At the same time, considering that the low-resolution Gaussian mixture model map is the joint upper bound of the high-resolution Gaussian mixture model map, when calculating the scores of the top N candidate solutions, the Nth i The candidate solution has reached the highest resolution Gaussian mixture model map, so the score of all subsequent candidate solutions cannot exceed that of the Nth candidate solution. i candidate solutions, this large-scale pruning method can further reduce the amount of calculation and improve the calculation efficiency.

[0063] In summary, the global initialization positioning method 100 of the embodiment of the present invention adopts a Gaussian mixture model and a multi-resolution branch and bound algorithm, and achieves high concurrency acceleration through a GPU, thereby achieving fast calculation under a large-scale search.

[0064] The embodiment of the present invention also provides a global initialization positioning device, see Figure 2The global initialization positioning device 200 includes a memory 210 and a graphics processing unit 220. The memory 210 stores a computer program executed by the graphics processing unit 220. When the computer program is executed by the graphics processing unit 220, the global initialization positioning method 100 is executed.

[0065] Exemplarily, the memory 210 may include one or more computer program products, and the computer program product may include various forms of computer-readable storage media, such as volatile memory and / or non-volatile memory. The volatile memory may include, for example, random access memory (RAM) and / or cache memory (cache), etc. The non-volatile memory may include, for example, read-only memory (ROM), hard disk, flash memory, etc. The graphics processing unit 220 may run the program instructions stored in the memory 210 to implement the functions (implemented by the graphics processing unit) in the embodiments of the present invention described herein and / or other desired functions. Since the graphics processing unit 220 includes multiple parallel computing units, the operations of each step in the above method can be performed in parallel, and the global initialization positioning of the vehicle can be achieved in a very short time.

[0066] The embodiment of the present invention also provides an autonomous driving vehicle, which can be used to implement the global initialization positioning method 100 described above. The autonomous driving vehicle is an intelligent vehicle that does not require a driver to perform driving operations and can automatically complete vehicle driving tasks instead of the driver; the autonomous driving vehicle may also have a manual driving function. Figure 3 , Figure 3 A schematic block diagram of an autonomous driving vehicle according to an embodiment of the present invention is shown.

[0067] like Figure 3 As shown, the autonomous driving vehicle includes a vehicle body 300, a laser radar 310, and a global initialization positioning device 320. The laser radar 310 is used to collect point cloud data, and the autonomous driving vehicle may include one or more laser radars. The global initialization positioning device 320 is connected to the laser radar 310 by wire or wirelessly to receive the point cloud data collected by the laser radar 310.

[0068] Wherein, as an example, the laser radar may include a transmitting module, a receiving module, a sampling module and an operation module. Wherein, the transmitting module may transmit a light pulse sequence (e.g., a laser pulse sequence). The receiving module may receive a light pulse sequence reflected by the detected object, and perform photoelectric conversion on the light pulse sequence to obtain an electrical signal, which may be output to the sampling module after processing the electrical signal. The sampling module may sample the electrical signal to obtain a sampling result. The operation module may determine the distance between the laser radar and the detected object based on the sampling result of the sampling module.

[0069] In addition to the above modules, the laser radar may also include a scanning module for changing the propagation direction of at least one laser pulse sequence emitted by the transmitting module. The scanning module may include multiple optical elements for changing the propagation path of the light beam, wherein the optical element may change the propagation path of the light beam by reflecting, refracting, diffracting, etc. the light beam. Exemplarily, each optical element in the scanning module may rotate to project light in different directions, thereby scanning the space around the laser radar.

[0070] The global initialization positioning device 320 may be as described above with reference to Figure 2 The described global initialization positioning device is used to execute the global initialization positioning method 100 as described above, so as to perform global initialization positioning of the autonomous driving vehicle based on the point cloud data collected by the laser radar. It should be noted that the autonomous driving vehicle also includes other components, which are not limited in the embodiment of the present invention. The global initialization positioning method 100 executed by the global initialization positioning device 320 can refer to the above and will not be described in detail here.

[0071] In addition, according to an embodiment of the present invention, a computer storage medium is also provided, on which program instructions are stored, and when the program instructions are run by a computer or a processor, the corresponding steps of the global initialization positioning method 100 of the embodiment of the present invention are executed, and the specific details can be found above. The computer storage medium may 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 disk read-only memory (CD-ROM), a USB memory, or any combination of the above storage media. The computer-readable storage medium may be any combination of one or more computer-readable storage media.

[0072] In summary, the global initialization positioning method, device, autonomous driving vehicle and storage medium of the embodiments of the present invention can achieve fast and high-precision positioning under large-scale search. On the one hand, it reduces the dependence on high-precision inertial navigation and can use a single-point GPS with large error as the initial value to reduce hardware costs; on the other hand, it can expand the search range in special scenarios and quickly obtain high-precision positioning results.

[0073] Although example embodiments have been described herein with reference to the accompanying drawings, it should be understood that the above example embodiments are merely exemplary and are not intended to limit the scope of the present application thereto. Various changes and modifications may be made therein by one of ordinary skill in the art 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 required by the appended claims.

[0074] Those of ordinary skill in the art will appreciate that the units and algorithm steps of each example described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are performed in hardware or software depends on the specific application and design constraints of the technical solution. Professional and technical personnel can use different methods to implement the described functions for each specific application, but such implementation should not be considered to be beyond the scope of this application.

[0075] In the several embodiments provided in this 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 only schematic, for example, the division of the units is only a logical function division, and there may be other division methods in actual implementation, such as multiple units or components can be combined or integrated into another device, or some features can be ignored or not executed.

[0076] In the description provided herein, a large number of specific details are described. However, it is understood that the embodiments of the present application can be practiced without these specific details. In some instances, well-known methods, structures and techniques are not shown in detail so as not to obscure the understanding of this description.

[0077] Similarly, it should be understood that in order to streamline the present application and help understand one or more of the various inventive aspects, in the description of the exemplary embodiments of the present application, the various features of the present application are sometimes grouped together into a single embodiment, figure, or description thereof. However, the method of the present application should not be interpreted as reflecting the following intention: the claimed application requires more features than the features clearly stated in each claim. More specifically, as reflected in the corresponding claims, the inventive point is that the corresponding technical problem can be solved with features less than all the features of a single disclosed embodiment. Therefore, the claims following the specific embodiment are hereby explicitly incorporated into the specific embodiment, wherein each claim itself serves as a separate embodiment of the present application.

[0078] It will be understood by those skilled in the art that, except for mutually exclusive features, all features disclosed in this specification (including the accompanying claims, abstracts and drawings) and all processes or units of any method or device disclosed in this specification may be combined in any combination. Unless otherwise expressly stated, each feature disclosed in this specification (including the accompanying claims, abstracts and drawings) may be replaced by an alternative feature that provides the same, equivalent or similar purpose.

[0079] In addition, those skilled in the art will appreciate that, although some embodiments described herein include certain features included in other embodiments but not other features, the combination of features of different embodiments is meant to be within the scope of the present application and form different embodiments. For example, in the claims, any one of the claimed embodiments can be used in any combination.

[0080] The various component embodiments of the present application can be implemented in hardware, or in software modules running on one or more processors, or in a combination thereof. It should be understood by those skilled in the art that a microprocessor or digital signal processor (DSP) can be used in practice to implement some or all functions of some modules according to the embodiments of the present application. The application can also be implemented as a device program (e.g., computer program and computer program product) for executing a part or all of the methods described herein. Such a program implementing the present application can be stored on a computer-readable medium, or can have the form of one or more signals. Such a signal can be downloaded from an Internet website, or provided on a carrier signal, or provided in any other form.

[0081] It should be noted that the above embodiments illustrate rather than limit the present application, and that those skilled in the art may devise alternative embodiments without departing from the scope of the appended claims. In the claims, any reference symbol between brackets should not be constructed as a limitation to the claims. The present application may be implemented by means of hardware including several different elements and by means of a suitably programmed computer. In a unit claim that lists several devices, several of these devices may be embodied by the same hardware item. The use of the words first, second, and third, etc. does not indicate any order. These words may be interpreted as names.

[0082] The above is only a specific implementation method or an explanation of a specific implementation method of the present application, and the protection scope of the present application is not limited thereto. Any changes or substitutions that can be easily thought of by a person skilled in the art within the technical scope disclosed in the present application should be included in the protection scope of the present application. The protection scope of the present application shall be based on the protection scope of the claims.

Claims

1. A global initialization positioning method, characterized in that: The method is used for a graphics processing unit, and the method comprises: Acquire at least two Gaussian mixture model maps of different resolutions, wherein the Gaussian mixture model map is obtained by compressing three-dimensional point cloud data into a two-dimensional grid structure, wherein each grid in the two-dimensional grid structure maintains a Gaussian mixture model, and the Gaussian mixture model is used to characterize the height distribution of the point cloud in the grid; Get point cloud data collected by LiDAR; Based on the point cloud data, branch calculations are performed in parallel on the registration scores of multiple candidate solutions under the at least two Gaussian mixture model maps, and when the highest score is the score of the candidate solution under the Gaussian mixture model map with the highest resolution, the candidate solution corresponding to the highest score is output as the initial positioning result; The parallel branching calculation of the registration scores of the multiple candidate solutions under the at least two Gaussian mixture model maps includes: calculating the scores of the multiple candidate solutions under the Gaussian mixture model map with the lowest resolution in parallel through the graphics processing unit, and sorting the candidate solutions according to the scores of the candidate solutions to obtain a candidate solution score queue; selecting the N candidate solutions with the highest scores in the candidate solution score queue, calculating their scores under the next layer of Gaussian mixture model map, and adding them to the score queue according to the scores; repeatedly calculating the scores of the N candidate solutions with the highest scores in the score queue under the next layer of Gaussian mixture model map, until the highest score in the score queue is the score of the candidate solution under the Gaussian mixture model map with the highest resolution; The method also includes: acquiring a three-dimensional point cloud map collected by a laser radar; compressing the three-dimensional point cloud map into a two-dimensional grid structure in parallel to obtain the Gaussian mixture model map with the highest resolution, wherein each grid in the two-dimensional grid structure maintains a Gaussian mixture model, and the Gaussian mixture model is used to characterize the height distribution of the point cloud in the grid; calculating the upper bound of the joint distribution of each adjacent plurality of grids in the two-dimensional grid structure to obtain a Gaussian mixture model map with reduced resolution; and repeatedly calculating the upper bound of the joint distribution of each adjacent plurality of grids in each Gaussian mixture model map until a Gaussian mixture model map with a predetermined number of layers is obtained.

2. The global initialization positioning method according to claim 1, characterized in that: The candidate solutions include the range of the horizontal and vertical coordinates and the range of the yaw angle of the laser radar.

3. The global initialization positioning method according to claim 1, characterized in that: The step of compressing the three-dimensional point cloud map into a two-dimensional grid structure in parallel includes: The mean and variance of the multi-core Gaussian distribution in each grid in the two-dimensional grid structure are estimated in parallel by using the multiple parallel computing units of the graphics processing unit.

4. The global initialization positioning method according to claim 1, characterized in that: The calculating the upper bound of the joint distribution of each adjacent plurality of grids in the two-dimensional grid structure comprises: The two-dimensional grid structure is subjected to sliding processing through a sliding window of a preset size, and the upper bound of the joint distribution of all grids currently covered by the sliding window is calculated.

5. The global initialization positioning method according to claim 1, characterized in that: The method is used in an autonomous driving vehicle, and the graphics processing unit is an on-board graphics processing unit.

6. A global initialization positioning device, characterized in that: The device comprises a memory and a graphics processing unit, wherein a computer program executed by the graphics processing unit is stored in the memory, and the computer program executes the global initialization positioning method according to any one of claims 1 to 5 when executed by the graphics processing unit.

7. An autonomous driving vehicle, characterized in that: The autonomous driving vehicle comprises: Car body; A laser radar mounted on the vehicle body, the laser radar being used to collect point cloud data; The global initialization positioning device as described in claim 6 is connected to the laser radar and is used to execute the global initialization positioning method as described in any one of claims 1-5.

8. A storage medium having a computer program stored thereon, wherein the computer program executes the global initialization positioning method according to any one of claims 1 to 5 when running.