Obstacle filtering method, device and equipment for autonomous vehicle and medium

Abstract

The embodiment of the invention discloses an obstacle filtering method and device for an autonomous vehicle, equipment and a medium. The method comprises the steps of obtaining current point cloud data and a reference point cloud frame, and obtaining candidate point cloud data according to a preset first filtering range and the current point cloud data; inputting the candidate point cloud data into an obstacle detection model, outputting a candidate obstacle and a candidate obstacle probability, and determining a first reference obstacle and the ground according to the candidate obstacle probability and a preset probability threshold; determining reference point cloud data according to the candidate point cloud data, the ground and the first reference obstacle, and determining a second reference obstacle according to the reference point cloud data; selecting a target obstacle according to a preset second filtering range and an obstacle recognition strategy; and determining an actual obstacle in the reference point cloud frame, and updating the target obstacle to obtain an actual obstacle in the current point cloud frame. According to the scheme, through layer-by-layer filtering, the accuracy of obstacle recognition is improved.

Description

Technical Field

[0001] Embodiments of the present invention relate to the field of autonomous driving technology, and in particular to an obstacle filtering method, device, equipment and medium for an autonomous driving vehicle. Background Art

[0002] The perception module plays a vital role in the autonomous driving system. The functions of each perception sub-module are realized through three sensors: lidar, radar and camera. Each sensor performs the task of target recognition and tracking. Finally, the sensor data is fused through fusion. Each sensor realizes the function of object (obstacle) recognition.

[0003] In the existing technology, whether it is relying on the detection model to predict obstacles or detecting obstacles through traditional geometric methods, false detections will inevitably occur, such as detecting vegetation as cars or trucks. The increase in false detections will have a great impact on the planning and control of the autonomous driving route, affecting the generation of effective driving routes. Many false detections will inevitably cause emergency braking, affecting the driving experience of autonomous driving. Therefore, it is crucial to improve the accuracy of obstacle recognition. Summary of the invention

[0004] The present invention provides an obstacle filtering method, device, equipment and medium for an autonomous driving vehicle to improve the accuracy of obstacle recognition.

[0005] According to one aspect of an embodiment of the present invention, a method for filtering obstacles of an autonomous driving vehicle is provided, comprising:

[0006] Acquire current point cloud data in a current point cloud frame collected by the laser radar, and a reference point cloud frame associated with the current point cloud frame, and perform a primary filtering on the current point cloud data according to a preset first filtering range to obtain candidate point cloud data;

[0007] Inputting the candidate point cloud data into a trained obstacle detection model, outputting the candidate obstacle and the candidate obstacle probability of the obstacle type to which the candidate obstacle belongs, and determining a first reference obstacle and the ground from the candidate obstacles according to the candidate obstacle probability and a preset probability threshold; wherein the candidate obstacle includes at least one candidate point cloud data;

[0008] Determine reference point cloud data according to the candidate point cloud data, the ground and the first reference obstacle, and determine a second reference obstacle in the current point cloud frame according to the reference point cloud data;

[0009] Selecting a target obstacle from reference obstacles according to a preset second filtering range and a preset obstacle identification strategy; wherein the reference obstacles include the first reference obstacle and the second reference obstacle;

[0010] An actual obstacle in the reference point cloud frame is determined, and the target obstacle is updated according to the actual obstacle in the reference point cloud frame to obtain the actual obstacle in the current point cloud frame.

[0011] According to another aspect of the present invention, there is provided an obstacle filtering device for an autonomous driving vehicle, comprising:

[0012] A point cloud data acquisition module, used to acquire current point cloud data in a current point cloud frame collected by the laser radar, and a reference point cloud frame associated with the current point cloud frame, and perform a primary filtering on the current point cloud data according to a preset first filtering range to obtain candidate point cloud data;

[0013] a first reference obstacle determination module, configured to input the candidate point cloud data into a trained obstacle detection model, output a candidate obstacle and a candidate obstacle probability of an obstacle type to which the candidate obstacle belongs, and determine a first reference obstacle and a ground surface from the candidate obstacles according to the candidate obstacle probability and a preset probability threshold; wherein the candidate obstacle includes at least one candidate point cloud data;

[0014] a second reference obstacle determination module, configured to determine reference point cloud data according to the candidate point cloud data, the ground and the first reference obstacle, and determine a second reference obstacle in the current point cloud frame according to the reference point cloud data;

[0015] a target obstacle determination module, configured to select a target obstacle from reference obstacles according to a preset second filtering range and a preset obstacle identification strategy; wherein the reference obstacles include the first reference obstacle and the second reference obstacle;

[0016] The actual obstacle determination module is used to determine the actual obstacle in the reference point cloud frame, and update the target obstacle according to the actual obstacle in the reference point cloud frame to obtain the actual obstacle in the current point cloud frame.

[0017] According to another aspect of the present invention, there is provided an electronic device, comprising:

[0018] one or more processors;

[0019] A memory for storing one or more programs;

[0020] When one or more programs are executed by one or more processors, the one or more processors can execute any obstacle filtering method for an autonomous driving vehicle provided in an embodiment of the present invention.

[0021] According to another aspect of the present invention, a computer-readable storage medium is provided, which stores computer instructions, and the computer instructions are used to enable a processor to implement any obstacle filtering method for an autonomous driving vehicle provided in an embodiment of the present invention when executed.

[0022] An embodiment of the present invention provides an obstacle filtering solution for an autonomous driving vehicle, which obtains current point cloud data in a current point cloud frame collected by a laser radar and a reference point cloud frame associated with the current point cloud frame, and performs a primary filtering on the current point cloud data according to a preset first filtering range to obtain candidate point cloud data; inputs the candidate point cloud data into a trained obstacle detection model, outputs candidate obstacles and candidate obstacle probabilities of obstacle types to which the candidate obstacles belong, and determines a first reference obstacle and a ground from the candidate obstacles according to the candidate obstacle probabilities and a preset probability threshold; wherein the candidate obstacles include at least one candidate point cloud data; determines reference point cloud data according to the candidate point cloud data, the ground and the first reference obstacle, and determines a second reference obstacle in the current point cloud frame according to the reference point cloud data; selects a target obstacle from the reference obstacles according to a preset second filtering range and a preset obstacle recognition strategy; wherein the reference obstacles include a first reference obstacle and a second reference obstacle; determines an actual obstacle in the reference point cloud frame, and updates the target obstacle according to the actual obstacle in the reference point cloud frame to obtain an actual obstacle in the current point cloud frame. The above scheme, by introducing a preset first filtering range, an obstacle detection model, a preset probability threshold, a preset second filtering range, an obstacle recognition strategy and actual obstacles in the reference point cloud frame, realizes layer-by-layer filtering of the current point cloud data, thereby improving the accuracy of determining the actual obstacles in the current point cloud frame, that is, improving the accuracy of obstacle recognition.

[0023] It should be understood that the contents described in this section are not intended to identify the key or important features of the embodiments of the present invention, nor are they intended to limit the scope of the present invention. Other features of the present invention will become easily understood through the following description. BRIEF DESCRIPTION OF THE DRAWINGS

[0024] In order to more clearly illustrate the technical solutions in the embodiments of the present invention, the following briefly introduces the drawings required for use in the description of the embodiments. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work.

[0025] Figure 1 is a flow chart of an obstacle filtering method for an autonomous driving vehicle provided in Embodiment 1 of the present invention;

[0026] Figure 2 is a flow chart of an obstacle filtering method for an autonomous driving vehicle provided in Embodiment 2 of the present invention;

[0027] Figure 3 is a schematic structural diagram of an obstacle filtering device for an autonomous driving vehicle provided in a third embodiment of the present invention;

[0028] Figure 4 It is a structural schematic diagram of an electronic device for implementing an obstacle filtering method for an autonomous driving vehicle provided in a fourth embodiment of the present invention. DETAILED DESCRIPTION

[0029] The present invention will be further described in detail below in conjunction with the accompanying drawings and embodiments. It is to be understood that the specific embodiments described herein are only used to explain the present invention, rather than to limit the present invention. It should also be noted that, for ease of description, only parts related to the present invention, rather than all structures, are shown in the accompanying drawings.

[0030] Embodiment 1

[0031] Figure 1 This is a flow chart of an obstacle filtering method for an autonomous driving vehicle provided in Embodiment 1 of the present invention. This embodiment can be applied to filtering obstacles on the road during the operation of the autonomous driving vehicle to determine the actual obstacles on the road. The method can be executed by an obstacle filtering device of the autonomous driving vehicle, which can be implemented in software and / or hardware and can be configured in an electronic device that carries the obstacle filtering function of the autonomous driving vehicle.

[0032] See also Figure 1 The obstacle filtering method of the autonomous driving vehicle shown includes:

[0033] S110, obtaining current point cloud data in a current point cloud frame collected by the laser radar, and a reference point cloud frame associated with the current point cloud frame, and performing an initial filtering on the current point cloud data according to a preset first filtering range to obtain candidate point cloud data.

[0034] Among them, the current point cloud frame is the point cloud frame collected by the laser radar at the current moment. The current point cloud data refers to the data in the current point cloud frame. Exemplarily, the current point cloud frame can be understood as an image, and the current point cloud data can be understood as the data in the image. The reference point cloud frame refers to a point cloud frame that has a temporal relationship with the current point cloud frame. Exemplarily, the reference point cloud frame may include historical point cloud frames associated with the current point cloud frame, and / or future point cloud frames associated with the current point cloud frame. For example, if the current point cloud frame is a point cloud frame collected by the laser radar at the 10th second, the reference point cloud frame may include the point cloud frame collected by the laser radar in the previous 0 to 9 seconds, and / or the point cloud frame collected by the laser radar at the 11th second.

[0035] Among them, the preset first filter range can be understood as the initial range of the identified obstacles. The embodiment of the present invention does not impose any limitation on the size of the preset first filter range, which can be set by technicians based on experience. Exemplarily, the preset first filter range can be determined based on the road boundary and road area of ​​the lane where the autonomous driving vehicle is located determined by the high-precision map. Candidate point cloud data refers to point cloud data obtained by initial filtering from the current point cloud data.

[0036] For example, since the sensor in the autonomous driving vehicle is a laser radar, the laser point cloud obtained by the laser radar is filtered from the moment it enters the module. First, the NAN (error) points, that is, the points where the laser point cloud is emitted but not returned, are filtered out, which are invalid points. Because the embodiment of the present invention only cares about the point cloud within the lane range, the current point cloud data can be initially filtered according to the set ROI (region of interst, range of interest) range, that is, the preset first filtering range, to obtain candidate point cloud data. If there is a high-precision map module, the road boundary and road area can be extracted through the high-precision map module as the preset first filtering range, and the current point cloud data can be filtered according to the road boundary and road area to obtain candidate point cloud data.

[0037] Specifically, the current point cloud data in the current point cloud frame collected by the laser radar and the reference point cloud frame associated with the current point cloud frame are obtained; the current point cloud data within the preset first filtering range is used as candidate point cloud data, and the current point cloud data not within the preset first filtering range is deleted.

[0038] S120: input the candidate point cloud data into a trained obstacle detection model, output the candidate obstacles and the candidate obstacle probabilities of the obstacle types to which the candidate obstacles belong, and determine a first reference obstacle and the ground from the candidate obstacles according to the candidate obstacle probabilities and a preset probability threshold.

[0039] The obstacle detection model refers to a model that can be used to determine the obstacles to which the candidate point cloud data belongs. The embodiment of the present invention does not impose any restrictions on the structure of the obstacle detection model, which can be set by technicians based on experience. The obstacle detection model in the embodiment of the present invention can use the bev (Bird's Eye View, a neural network that converts input data into a top view (bird's eye view) representation) method or the pillar method to extract features and implement model training.

[0040] The candidate obstacle refers to a detection frame obtained based on the obstacle detection model, that is, one detection frame corresponds to one candidate obstacle, and one detection frame includes at least one candidate point cloud data, that is, the candidate obstacle includes at least one candidate point cloud data. The obstacle type refers to the type of the candidate obstacle. The embodiment of the present invention does not impose any limitation on the obstacle type, which can be set by technicians based on experience.

[0041] The candidate obstacle probability refers to the probability that the candidate obstacle belongs to the candidate obstacle type.

[0042] The embodiment of the present invention does not impose any limitation on the size of the preset probability threshold, which can be set by a technician based on experience or needs, or determined repeatedly through a large number of experiments. The first reference obstacle refers to an obstacle in the current point cloud frame determined based on the obstacle detection model that may affect the operation of the autonomous driving vehicle.

[0043] For example, if the obstacle types in the obstacle detection model can include type A, type B and type C, the candidate point cloud data is input into the obstacle detection model, the obstacle detection model divides the candidate point cloud data into at least one detection frame, and predicts the candidate obstacle probability of the obstacle type to which each detection frame belongs.

[0044] In the embodiment of the present invention, by introducing an obstacle detection model, grids or pillars that do not meet the characteristics of target obstacles (cars, trucks, cones, cyclists, pedestrians, etc.) can be deleted.

[0045] S130: Determine reference point cloud data according to the candidate point cloud data, the ground, and the first reference obstacle, and determine a second reference obstacle in the current point cloud frame according to the reference point cloud data.

[0046] The reference point cloud data refers to point cloud data obtained by screening the candidate point cloud data based on the ground and the first reference obstacle. Exemplarily, the point cloud data corresponding to the ground and the point cloud data corresponding to the first reference obstacle are eliminated from the candidate point cloud data to obtain the reference point cloud data.

[0047] The second reference point cloud data refers to obstacles in the current point cloud frame determined based on the reference point cloud data that may affect the operation of the autonomous driving vehicle.

[0048] The reference obstacle in the embodiment of the present invention is determined based on the reference point cloud grid; the reference point cloud grid is obtained by screening from the candidate point cloud grids. The reference point cloud grid can be screened from the candidate point cloud grids according to a preset probability threshold.

[0049] S140: Select a target obstacle from reference obstacles according to a preset second filtering range and a preset obstacle recognition strategy.

[0050] The preset second filtering range can be understood as the target range of the identified obstacles. The embodiment of the present invention does not impose any limitation on the size of the preset second identification range, which can be set by technicians based on experience. It should be noted that the preset first filtering range in the embodiment of the present invention is larger than the preset second filtering range.

[0051] It can be understood that if the preset first filtering range is larger than the preset second filtering range, the comprehensiveness of the candidate point cloud data is improved, omissions are avoided, and the accuracy of subsequent obstacle determination is improved.

[0052] The target obstacle refers to an obstacle selected from the reference obstacles. The reference obstacle refers to an obstacle in the current point cloud frame that may affect the operation of the autonomous driving vehicle. The reference obstacle may include a first reference obstacle and a second reference obstacle.

[0053] The obstacle identification strategy refers to a pre-set strategy that can be used to identify obstacles. The embodiment of the present invention does not impose any limitation on the setting of the obstacle identification strategy, which can be set by technicians based on experience.

[0054] In an optional embodiment, selecting a target obstacle from reference obstacles according to a preset second filtering range and a preset obstacle identification strategy includes: determining a reference retained obstacle in the reference obstacles according to the preset second filtering range; and selecting a target obstacle from the reference retained obstacles based on a preset obstacle identification strategy; wherein the obstacle identification strategy includes an obstacle retention strategy and an obstacle filtering strategy.

[0055] The reference retained obstacles refer to obstacles that do not need to be deleted and are screened from the reference obstacles according to the preset second filtering range. The obstacle retention strategy refers to a strategy for determining whether the reference obstacles can be retained. The obstacle filtering strategy refers to a strategy for determining whether the reference obstacles need to be deleted. The embodiment of the present invention does not impose any restrictions on the obstacle retention strategy and the obstacle filtering strategy, which can be set by technicians based on experience.

[0056] Exemplarily, the obstacle retention strategy may include at least one of the following sub-strategies: size retention strategy, length and width retention strategy, projection area retention strategy, and height and width retention strategy. Among them, the size retention strategy may be used to retain the reference reserved obstacles that meet the size of the corresponding obstacles. For example, if the reference reserved obstacle is a car, then if the size of the reference reserved obstacle meets the size of the car, it can be retained. Among them, the length and width retention strategy may be used to retain the reference reserved obstacles that meet the length, width, and length-to-width ratio of the corresponding obstacles. This strategy may be specifically used in the scenario where the reference reserved obstacle is a car and the car has only one face. Among them, the projection area retention strategy may be used to retain the reference reserved obstacles that meet the projection area of ​​the corresponding obstacles. This strategy may be specifically used in the scenario where the reference reserved obstacle is a car and the car has only one face, and the reference reserved obstacle is on the side or in front of the autonomous driving vehicle. Among them, the height and width retention strategy may be used to retain the reference reserved obstacles that meet the height and width of the corresponding obstacles. This strategy may be specifically used in the scenario where the reference reserved obstacle is a car and the reference reserved obstacle is in front of the autonomous driving vehicle.

[0057] Exemplarily, obstacle filtering strategies may include the following: filtering reference retained obstacles that are cars and do not meet car standards; filtering reference retained obstacles with smaller points or smaller sizes, that is, filtering out small objects or filtering out reference retained obstacles with fewer points; filtering low reference retained obstacles, that is, filtering out small obstacles that cars can drive over; filtering suspended reference retained obstacles, that is, filtering out reference retained obstacles floating in the air; filtering low-density reference retained obstacles, that is, filtering vegetation that is similar in shape to cars but has very little point density; front view filtering, that is, filtering out obstacles such as cars under large trees or in tunnels; large obstacle cluster filtering, that is, filtering out large obstacles such as small cars under large trees or in tunnels; filtering large obstacles by clustering, that is, filtering out large obstacles by clustering. For large obstacles, set a smaller grid size, re-cluster large obstacles, and subdivide smaller obstacles; bunny and ghost filtering, that is, bunny filtering is to filter out obstacles with a point number of more than 2 times the symmetrical position and a very high intensity value. For highly reflective objects, the lidar will produce mirror reflections and form ghost obstacles, so ghost filtering is required; ground obstacle filtering, that is, objects under the ground may be reflections formed by water on the road, so they need to be filtered; rain or water splash obstacle filtering, that is, obstacles formed by rain or water splashes from cars are filtered according to the number of points, area and density; rod filtering, that is, filtering objects that are as high as poles and have a small area.

[0058] It should be noted that, in the embodiment of the present invention, the obstacle retention strategy can be first used to filter the reference retained obstacles. If the reference retained obstacle does not satisfy any seed strategy in the obstacle retention strategy, the reference retained obstacle is deleted; if the reference retained obstacle satisfies any one of the obstacle retention strategies, the obstacle filtering strategy is used to filter the reference retained obstacle again; if the reference retained obstacle satisfies any one of the obstacle filtering strategies, the reference retained obstacle is deleted; if not, the target obstacle is selected from the reference retained obstacles.

[0059] Specifically, reference obstacles within a preset second filtering range are used as reference retained obstacles; the reference retained strategy is initially filtered according to the obstacle retention strategy; the reference retained obstacles that meet any seed strategy in the reference retained strategy are filtered again according to the obstacle filtering strategy; and the target obstacle is selected from the reference retained obstacles that do not meet any of the obstacle filtering strategies.

[0060] It can be understood that by determining the target obstacle according to the preset second filtering range, the obstacle retention strategy and the obstacle filtering strategy, the accuracy of the determined target obstacle is improved.

[0061] In an optional embodiment, after determining the reference retained obstacle in the reference obstacles according to a preset second filtering range, the method further includes: determining the reference filtered obstacle in the reference obstacles according to the preset second filtering range; obtaining the current position data of the autonomous driving vehicle, and determining the road boundary line of the lane where the autonomous driving vehicle is located according to the current position data; determining the boundary distance between the reference filtered obstacle and the road boundary line; and selecting the reference retained obstacle from the reference filtered obstacles according to the boundary distance and a preset boundary distance threshold.

[0062] Among them, the reference filtering obstacle refers to a reference obstacle that is not within the preset second filtering range. The current position data refers to the data of the position of the autonomous driving vehicle at the current moment. The road boundary line refers to the boundary line of the road where the autonomous driving vehicle is located. The boundary distance refers to the distance between the reference filtering obstacle and the road boundary line. The embodiment of the present invention does not impose any limitation on the size of the preset boundary distance threshold, which can be set by technicians based on experience, or repeatedly determined through a large number of experiments.

[0063] Exemplarily, a reference retained obstacle is selected from the reference filtered obstacles based on the boundary distance and the preset boundary distance threshold, including: for any reference filtered obstacle, if the boundary distance of the reference filtered obstacle is less than or equal to the preset boundary distance threshold, the reference filtered obstacle is used as a reference retained obstacle; if the boundary distance of the reference filtered obstacle is greater than the preset boundary distance threshold, the reference filtered obstacle is deleted.

[0064] In the embodiment of the present invention, since there are two road boundary lines, the boundary distances between the reference filtered obstacle and the two road boundary lines at different positions can be determined, and the boundary distance with a smaller value is compared with a preset boundary distance threshold to determine whether the reference filtered obstacle is a reference retained obstacle.

[0065] Specifically, a reference obstacle that is not within a preset second filtering range is used as a reference filtering obstacle; the current position data of the autonomous driving vehicle is obtained, and based on the current position data, a road boundary line of a lane where the autonomous driving vehicle is located is determined; a boundary distance between the reference filtering obstacle and the road boundary line is determined; and a reference retained obstacle is selected from the reference filtering obstacles based on the boundary distance and a preset boundary distance threshold.

[0066] It can be understood that by introducing the boundary distance and the preset boundary distance threshold, the reference reserved obstacles are updated, the omission of the determined reference reserved obstacles is avoided, and the comprehensiveness of the determined reference reserved obstacles is improved.

[0067] S150: Determine an actual obstacle in the reference point cloud frame, and update the target obstacle according to the actual obstacle in the reference point cloud frame to obtain the actual obstacle in the current point cloud frame.

[0068] Among them, actual obstacles refer to obstacles that actually affect the operation of autonomous driving vehicles, that is, actual obstacles refer to obstacles that can affect the operation of autonomous driving vehicles.

[0069] It should be noted that the actual obstacles in the reference point cloud frame in the embodiment of the present invention can be determined using the method provided by the embodiment of the present invention.

[0070] In an optional embodiment, the target obstacle is updated according to the actual obstacle in the reference point cloud frame to obtain the actual obstacle in the current point cloud frame, including: determining a tracking obstacle corresponding to the target obstacle among the actual obstacles in the reference point cloud frame; determining a running trajectory of the tracking obstacle in the reference point cloud frame, and updating the target obstacle according to the running trajectory to obtain the actual obstacle in the current point cloud frame.

[0071] Among them, the tracking obstacle refers to the actual obstacle that is the same as the target obstacle matched from the reference point cloud frame. The running trajectory refers to the running change trajectory of the tracking obstacle in each reference point cloud frame.

[0072] Specifically, for any target obstacle, the actual obstacle that is the same as the target obstacle in each reference point cloud frame is used as the tracking obstacle of the target obstacle; according to the driving changes of the tracking obstacle in each reference point cloud frame, the running trajectory of the tracking obstacle in each reference point cloud frame is determined; according to the running trajectory, it is judged whether the target obstacle is an abnormal obstacle; if so, the target obstacle is filtered; if not, the target obstacle is determined to be the actual obstacle in the current point cloud frame.

[0073] Among them, abnormal obstacles refer to abnormal target obstacles, that is, they can be understood as target obstacles that should not appear in the current operating scenario of the autonomous vehicle. For example, abnormal obstacles may include target obstacles that the vehicle repeatedly moves forward and backward, and stationary pedestrians or cyclists at a speed of 0 on the highway. The current operating scenario refers to the driving scenario of the autonomous vehicle at the current moment.

[0074] It can be understood that by tracking and filtering the target obstacle according to the running trajectory of the tracked obstacle to determine the actual obstacle in the current point cloud frame, the accuracy of determining the actual obstacle in the current point cloud frame is improved.

[0075] In an embodiment of the present invention, the actual obstacles in the reference point cloud frame can be used to track and identify the actual obstacles in the current point cloud frame, thereby improving the accuracy of the actual obstacles in the current point cloud frame. For example, target obstacles that the vehicle repeatedly moves forward and backward can be deleted based on the actual obstacles in the reference point cloud frame, and target obstacles detected on the highway that are stationary pedestrians or cyclists with a speed of 0 for multiple consecutive frames can be deleted.

[0076] An embodiment of the present invention provides an obstacle filtering solution for an autonomous driving vehicle, which obtains current point cloud data in a current point cloud frame collected by a laser radar and a reference point cloud frame associated with the current point cloud frame, and performs a primary filtering on the current point cloud data according to a preset first filtering range to obtain candidate point cloud data; inputs the candidate point cloud data into a trained obstacle detection model, outputs candidate obstacles and candidate obstacle probabilities of obstacle types to which the candidate obstacles belong, and determines a first reference obstacle and a ground from the candidate obstacles according to the candidate obstacle probabilities and a preset probability threshold; wherein the candidate obstacles include at least one candidate point cloud data; determines reference point cloud data according to the candidate point cloud data, the ground and the first reference obstacle, and determines a second reference obstacle in the current point cloud frame according to the reference point cloud data; selects a target obstacle from the reference obstacles according to a preset second filtering range and a preset obstacle recognition strategy; wherein the reference obstacles include a first reference obstacle and a second reference obstacle; determines an actual obstacle in the reference point cloud frame, and updates the target obstacle according to the actual obstacle in the reference point cloud frame to obtain an actual obstacle in the current point cloud frame. The above scheme, by introducing a preset first filtering range, an obstacle detection model, a preset probability threshold, a preset second filtering range, an obstacle recognition strategy and actual obstacles in the reference point cloud frame, realizes layer-by-layer filtering of the current point cloud data, thereby improving the accuracy of determining the actual obstacles in the current point cloud frame, that is, improving the accuracy of obstacle recognition.

[0077] Based on the above technical solutions, in an embodiment of the present invention, candidate obstacles of road boundaries or road area boundaries output by the obstacle detection model can be filtered out. Alternatively, if the obstacle detection model cannot predict the road boundaries or road area boundaries, they can be extracted based on point cloud geometric features to prepare for subsequent filtering, thereby providing boundary filtering guarantees in the absence of high-precision maps.

[0078] In the embodiments of the present invention, there are multiple ways to generate road boundaries: the road boundaries can be generated by obstacle detection models and traditional geometric methods. When high-precision maps are available, they can be used as an aid. When high-precision maps do not provide road boundaries or areas, road boundary ranges are provided to filter obstacles outside the road. When traditional geometric methods do not generate them, clustered super-long obstacles can also be used as road boundaries to ensure effective generation of road boundaries.

[0079] Embodiment 2

[0080] Figure 2This is a flow chart of an obstacle filtering method for an autonomous driving vehicle provided in the second embodiment of the present invention. Based on the above embodiments, this embodiment further refines the operation of "determining the second reference obstacle in the current point cloud frame based on the reference point cloud data" into "dividing the reference point cloud data into at least one reference point cloud grid, and determining the grid point cloud parameters of each reference point cloud data in the reference point cloud grid; determining the second reference obstacle in the current point cloud frame based on the grid point cloud parameters; wherein the reference obstacle is determined based on the reference point cloud grid" to improve the determination mechanism of the second reference obstacle. It should be noted that for the parts not described in detail in the embodiments of the present invention, reference may be made to the descriptions of other embodiments.

[0081] See also Figure 2 The obstacle filtering scheme for the autonomous driving vehicle shown includes:

[0082] S210, obtaining current point cloud data in a current point cloud frame collected by the laser radar, and a reference point cloud frame associated with the current point cloud frame, and performing an initial filtering on the current point cloud data according to a preset first filtering range to obtain candidate point cloud data.

[0083] S220: input the candidate point cloud data into a trained obstacle detection model, output the candidate obstacles and the candidate obstacle probabilities of the obstacle types to which the candidate obstacles belong, and determine a first reference obstacle and a ground obstacle from the candidate obstacles according to the candidate obstacle probabilities and a preset probability threshold.

[0084] The candidate obstacles include at least one candidate point cloud data.

[0085] In an optional embodiment, determining a first reference obstacle and a ground obstacle from candidate obstacles according to the candidate obstacle probability and a preset probability threshold includes: comparing the candidate obstacle probability of the candidate obstacle with the preset probability threshold; if the comparison result is that the candidate obstacle probability is greater than or equal to the preset probability threshold, determining the candidate obstacle as the first reference obstacle or the ground according to the obstacle type of the candidate obstacle.

[0086] Specifically, for any candidate obstacle, the candidate probability threshold of the candidate obstacle is compared with the preset probability threshold; if the comparison result is that the candidate obstacle probability is greater than or equal to the preset probability threshold, and if the obstacle type of the candidate obstacle is a non-ground type, the candidate obstacle is determined to be the first reference obstacle; if the obstacle type of the candidate obstacle is a ground type, the candidate obstacle is determined to be the ground; if the comparison result is that the candidate obstacle probability is less than the preset probability threshold, the candidate obstacle is eliminated.

[0087] It can be understood that by comparing the probability of the candidate obstacle with the preset probability threshold to determine the first reference obstacle and the ground, the accuracy of the determined first reference obstacle or the ground is improved.

[0088] S230: Determine reference point cloud data according to the candidate point cloud data, the ground, and the first reference obstacle.

[0089] S240: Divide the reference point cloud data into at least one reference point cloud grid, and determine grid point cloud parameters of each reference point cloud data in the reference point cloud grid.

[0090] The reference point cloud grid refers to a grid including at least one reference point cloud data obtained by dividing the reference point cloud data. The embodiment of the present invention does not limit the method of dividing the reference point cloud data to obtain the reference point cloud grid, which can be set by technicians based on experience.

[0091] The grid point cloud parameters refer to the basic attribute data of the reference point cloud data in the reference point cloud grid. The embodiment of the present invention does not impose any limitation on the grid point cloud parameters, which can be set by a technician based on experience. Exemplarily, the grid point cloud parameters may include at least one of the center of gravity of the reference point cloud grid, the number of reference point cloud data in the reference point cloud grid, the minimum height and maximum height of the reference point cloud data in the reference point cloud grid, etc.

[0092] S250: Determine a second reference obstacle in the current point cloud frame according to the grid point cloud parameters.

[0093] The second reference obstacle is determined based on the reference point cloud grid.

[0094] Exemplarily, determining a second reference obstacle in the current point cloud frame according to the grid point cloud parameters includes: filtering the reference point cloud grid according to the grid point cloud parameters; generating an intermediate obstacle according to the filtered reference point cloud grid; determining obstacle parameters of the intermediate obstacle according to the grid point cloud parameters of each reference point cloud grid in the intermediate obstacle; filtering the intermediate obstacle according to the obstacle parameters to obtain a second reference obstacle.

[0095] Among them, the intermediate obstacle refers to the obstacle determined according to the filtered reference point cloud grid. The obstacle parameter refers to the basic attribute data of the reference point cloud grid in the intermediate obstacle. Exemplarily, the obstacle parameter may include at least one of the point number variance, the center of gravity variance and the grid distribution. Exemplarily, the obstacle parameter may be determined based on the grid point cloud parameter, that is, the grid point cloud parameters of each reference point cloud grid of the intermediate obstacle may be summarized and sorted to obtain the obstacle parameter.

[0096] It can be understood that by introducing the obstacle parameter, the intermediate obstacles are filtered, and then the second reference obstacle is determined, thereby improving the accuracy of the determined reference obstacle.

[0097] In the embodiment of the present invention, since the reasoning of the obstacle detection model may miss detections, clustering is performed through traditional geometric methods to identify missed obstacles. However, in this process, some objects such as vegetation will be identified as obstacles, but these obstacles are unnecessary and need to be filtered out.

[0098] Specifically, the geometric clustering detection method generally uses the bev grid projection method for clustering. Therefore, during the clustering process, the point cloud will be projected onto a 2D grid plane, and the connected areas will be clustered to form obstacles. After the point cloud is projected, it can be filtered based on the centroid of the reference point cloud grid, the number of reference point cloud data in the reference point cloud grid, the minimum height and maximum height of the reference point cloud data in the reference point cloud grid, and other features; after clustering into intermediate obstacles, the reference point cloud grids contained in the intermediate obstacles can be counted, and the variance of the number of points, the variance of the centroid, the grid distribution, etc. can be counted, and the intermediate obstacles that do not meet the requirements can be filtered out to obtain the second reference obstacle.

[0099] It should be noted that for roadside fences, fences on the side of the highway, and vegetation isolation belts in the middle of the highway, geometric clustering will detect them as an extra-long obstacle, and such obstacles can be converted into road boundary categories.

[0100] S260: Select a target obstacle from reference obstacles according to a preset second filtering range and a preset obstacle recognition strategy.

[0101] The reference obstacles include a first reference obstacle and a second reference obstacle.

[0102] S270: Determine the actual obstacle in the reference point cloud frame, and update the target obstacle according to the actual obstacle in the reference point cloud frame to obtain the actual obstacle in the current point cloud frame.

[0103] An obstacle filtering scheme for an autonomous driving vehicle provided by an embodiment of the present invention is to refine the operation of determining the second reference obstacle in the current point cloud frame based on the reference point cloud data into dividing the reference point cloud data into at least one reference point cloud grid, and determining the grid point cloud parameters of each reference point cloud data in the reference point cloud grid; determining the second reference obstacle in the current point cloud frame based on the grid point cloud parameters; wherein the reference obstacle is determined based on the reference point cloud grid, thus improving the determination mechanism of the second reference obstacle. The above scheme determines the second reference obstacle through the geometric clustering detection method, thus avoiding the occurrence of missed detection and improving the comprehensiveness of the reference obstacles subsequently determined based on the second reference obstacle.

[0104] On the basis of the above technical solutions, an embodiment of the present invention provides a specific obstacle filtering process. Exemplarily, the on-board sensor of the autonomous driving vehicle collects the current point cloud data in the current point cloud frame; filters out invalid points in the current point cloud data, and then filters the current point cloud data according to a preset first filtering range to obtain candidate point cloud data; inputs the candidate point cloud data into a trained obstacle detection model, and the obstacle detection model infers the input candidate point cloud data, outputs the candidate obstacle and the candidate obstacle probability of the obstacle type to which the candidate obstacle belongs, and determines the first reference obstacle, the ground, the road boundary and the road area from the candidate obstacles based on the candidate obstacle probability and the preset probability threshold; selects the reference point cloud data from the candidate point cloud data, and based on the geometric clustering detection method, filters and identifies the overlong grid point cloud layered according to the obstacle parameters and the grid point cloud parameters. The obstacle is a road boundary method, the reference point cloud data is filtered to obtain a second reference obstacle; a reference obstacle including a first reference obstacle and a second reference obstacle is generated; based on the traditional road boundary detection method, the reference obstacle is boundary filtered; according to the preset second filtering range and the preset obstacle recognition strategy, the background obstacles (i.e., reference obstacles outside the preset second filtering range and with a boundary distance greater than the preset boundary distance threshold), boundary obstacles and general foreground obstacles (i.e., reference obstacles in the road area and having no effect on the operation of the autonomous driving vehicle) in the reference obstacle are filtered out to obtain the target obstacle; according to the running trajectory of the tracking obstacle in the reference point cloud frame, the target obstacle is tracked and filtered to determine the actual obstacle in the current point cloud frame. The embodiment of the present invention can also provide a plug-in mechanism, which is a dynamic loading mechanism that is lighter than the component, and its purpose is to allow developers to focus more on the functional logic and more conveniently expand the function. The plug-in development is highly independent and can be flexibly used by the component to load the plug-in as needed. The advantage is that different filtering strategies can be customized individually to face different perception scenarios or to increase filtering strategies, making the program more flexible and easy to expand, that is, ensuring that subsequent filtering strategies can be effectively and conveniently added to the program, making the code efficiently integrated and effectively supplemented when facing different scenarios.

[0105] Figure 3 This is a schematic diagram of the structure of an obstacle filtering device for an autonomous driving vehicle provided in Embodiment 3 of the present invention. This embodiment can be applied to filtering obstacles on the road during the operation of the autonomous driving vehicle to determine the actual obstacles on the road. The method can be executed by the obstacle filtering device of the autonomous driving vehicle, which can be implemented in software and / or hardware and can be configured in an electronic device that carries the obstacle filtering function of the autonomous driving vehicle.

[0106] like Figure 3 As shown, the device includes: a point cloud data acquisition module 310, a first reference obstacle determination module 320, a second reference obstacle determination module 330, a target obstacle determination module 340 and an actual obstacle determination module 350. Among them,

[0107] The point cloud data acquisition module 310 is used to acquire the current point cloud data in the current point cloud frame collected by the laser radar, and the reference point cloud frame associated with the current point cloud frame, and perform a primary filtering on the current point cloud data according to a preset first filtering range to obtain candidate point cloud data;

[0108] The first reference obstacle determination module 320 is used to input the candidate point cloud data into the trained obstacle detection model, output the candidate obstacle and the candidate obstacle probability of the obstacle type to which the candidate obstacle belongs, and determine the first reference obstacle and the ground from the candidate obstacles according to the candidate obstacle probability and a preset probability threshold; wherein the candidate obstacle includes at least one candidate point cloud data;

[0109] A second reference obstacle determination module 330, configured to determine reference point cloud data according to the candidate point cloud data, the ground and the first reference obstacle, and determine a second reference obstacle in the current point cloud frame according to the reference point cloud data;

[0110] A target obstacle determination module 340, configured to select a target obstacle from reference obstacles according to a preset second filtering range and a preset obstacle identification strategy; wherein the reference obstacles include the first reference obstacle and the second reference obstacle;

[0111] The actual obstacle determination module 350 is used to determine the actual obstacle in the reference point cloud frame, and update the target obstacle according to the actual obstacle in the reference point cloud frame to obtain the actual obstacle in the current point cloud frame.

[0112] An embodiment of the present invention provides an obstacle filtering solution for an autonomous driving vehicle, which obtains current point cloud data in a current point cloud frame collected by a laser radar and a reference point cloud frame associated with the current point cloud frame, and performs a primary filtering on the current point cloud data according to a preset first filtering range to obtain candidate point cloud data; inputs the candidate point cloud data into a trained obstacle detection model, outputs candidate obstacles and candidate obstacle probabilities of obstacle types to which the candidate obstacles belong, and determines a first reference obstacle and a ground from the candidate obstacles according to the candidate obstacle probabilities and a preset probability threshold; wherein the candidate obstacles include at least one candidate point cloud data; determines reference point cloud data according to the candidate point cloud data, the ground and the first reference obstacle, and determines a second reference obstacle in the current point cloud frame according to the reference point cloud data; selects a target obstacle from the reference obstacles according to a preset second filtering range and a preset obstacle recognition strategy; wherein the reference obstacles include a first reference obstacle and a second reference obstacle; determines an actual obstacle in the reference point cloud frame, and updates the target obstacle according to the actual obstacle in the reference point cloud frame to obtain an actual obstacle in the current point cloud frame. The above scheme, by introducing a preset first filtering range, an obstacle detection model, a preset probability threshold, a preset second filtering range, an obstacle recognition strategy and actual obstacles in the reference point cloud frame, realizes layer-by-layer filtering of the current point cloud data, thereby improving the accuracy of determining the actual obstacles in the current point cloud frame, that is, improving the accuracy of obstacle recognition.

[0113] Optionally, the second reference obstacle determination module 330 includes:

[0114] A grid point cloud parameter determination unit, used for dividing the reference point cloud data into at least one reference point cloud grid, and determining a grid point cloud parameter of each reference point cloud data in the reference point cloud grid;

[0115] The second reference obstacle determining unit is used to determine a second reference obstacle in the current point cloud frame according to the grid point cloud parameters; wherein the second reference obstacle is determined based on the reference point cloud grid.

[0116] Optionally, the first reference obstacle determination module 320 includes:

[0117] A comparing unit, configured to compare the candidate obstacle probability of the candidate obstacle with the preset probability threshold;

[0118] The second reference obstacle determining unit is configured to determine, if the comparison result shows that the probability of the candidate obstacle is greater than or equal to the preset probability threshold, that the candidate obstacle is a first reference obstacle or the ground according to the obstacle type of the candidate obstacle.

[0119] Optionally, the target obstacle determination module 340 includes:

[0120] a reference reserved obstacle determining unit, configured to determine a reference reserved obstacle among the reference obstacles according to the preset second filtering range;

[0121] The target obstacle selection unit is used to select the target obstacle from the reference reserved obstacles based on a preset obstacle recognition strategy; wherein the obstacle recognition strategy includes an obstacle retention strategy and an obstacle filtering strategy.

[0122] Optionally, after referring to the reserved obstacle determination unit, the device further includes:

[0123] a reference filtering obstacle determining unit, configured to determine a reference filtering obstacle among the reference obstacles according to the preset second filtering range;

[0124] A road boundary line determination unit, used to obtain current position data of the autonomous driving vehicle, and determine the road boundary line of the lane where the autonomous driving vehicle is located based on the current position data;

[0125] a boundary distance determination unit, configured to determine a boundary distance between the reference filtered obstacle and the road boundary line;

[0126] The reference reserved obstacle determining unit is used to select a reference reserved obstacle from the reference filtered obstacles according to the boundary distance and a preset boundary distance threshold.

[0127] Optional, actual obstacle determination module, including:

[0128] A tracking obstacle determination unit, configured to determine a tracking obstacle corresponding to the target obstacle from among the actual obstacles in the reference point cloud frame;

[0129] The actual obstacle determination unit is used to determine the running trajectory of the tracked obstacle in the reference point cloud frame, and update the target obstacle according to the running trajectory to obtain the actual obstacle in the current point cloud frame.

[0130] Optionally, the preset first filtering range is larger than the preset second filtering range.

[0131] The obstacle filtering device for an autonomous driving vehicle provided in an embodiment of the present invention can execute the obstacle filtering method for an autonomous driving vehicle provided in any embodiment of the present invention, and has functional modules and beneficial effects corresponding to executing the obstacle filtering method for each autonomous driving vehicle.

[0132] In the technical solution of the present invention, the collection, storage, use, processing, transmission, provision and disclosure of the current point cloud frame, current point cloud data and reference point cloud frame, etc., all comply with the provisions of relevant laws and regulations and do not violate public order and good morals.

[0133] Embodiment 4

[0134] Figure 4 4 is a schematic diagram of the structure of an electronic device for implementing an obstacle filtering method for an autonomous driving vehicle provided by Embodiment 4 of the present invention. The electronic device 410 is intended to represent various forms of digital computers, such as laptop computers, desktop computers, workbenches, personal digital assistants, servers, blade servers, mainframe computers, and other suitable computers. The electronic device may also represent various forms of mobile devices, such as personal digital processing, cellular phones, smart phones, wearable devices (such as helmets, glasses, watches, etc.) and other similar computing devices. The components shown herein, their connections and relationships, and their functions are merely examples and are not intended to limit the implementation of the present invention described and / or required herein.

[0135] like Figure 4 As shown, the electronic device 410 includes at least one processor 411, and a memory connected to the at least one processor 411 in communication, such as a read-only memory (ROM) 412, a random access memory (RAM) 413, etc., wherein the memory stores a computer program that can be executed by at least one processor, and the processor 411 can perform various appropriate actions and processes according to the computer program stored in the read-only memory (ROM) 412 or the computer program loaded from the storage unit 418 to the random access memory (RAM) 413. In the RAM 413, various programs and data required for the operation of the electronic device 410 can also be stored. The processor 411, the ROM 412, and the RAM 413 are connected to each other via a bus 414. An input / output (I / O) interface 415 is also connected to the bus 414.

[0136] Multiple components in the electronic device 410 are connected to the I / O interface 415, including: an input unit 416, such as a keyboard, a mouse, etc.; an output unit 417, such as various types of displays, speakers, etc.; a storage unit 418, such as a disk, an optical disk, etc.; and a communication unit 419, such as a network card, a modem, a wireless communication transceiver, etc. The communication unit 419 allows the electronic device 410 to exchange information / data with other devices through a computer network such as the Internet and / or various telecommunication networks.

[0137] Processor 411 may be a variety of general and / or dedicated processing components with processing and computing capabilities. Some examples of processor 411 include, but are not limited to, a central processing unit (CPU), a graphics processing unit (GPU), various dedicated artificial intelligence (AI) computing chips, various processors running machine learning model algorithms, digital signal processors (DSPs), and any appropriate processors, controllers, microcontrollers, etc. Processor 411 performs the various methods and processes described above, such as an obstacle filtering method for an autonomous driving vehicle.

[0138] In some embodiments, the obstacle filtering method for an autonomous driving vehicle may be implemented as a computer program, which is tangibly contained in a computer-readable storage medium, such as a storage unit 418. In some embodiments, part or all of the computer program may be loaded and / or installed on the electronic device 410 via the ROM 412 and / or the communication unit 419. When the computer program is loaded into the RAM 413 and executed by the processor 411, one or more steps of the obstacle filtering method for an autonomous driving vehicle described above may be performed. Alternatively, in other embodiments, the processor 411 may be configured to execute the obstacle filtering method for an autonomous driving vehicle in any other appropriate manner (e.g., by means of firmware).

[0139] Various implementations of the systems and techniques described above herein can be implemented in digital electronic circuit systems, integrated circuit systems, field programmable gate arrays (FPGAs), application specific integrated circuits (ASICs), application specific standard products (ASSPs), systems on chips (SOCs), load programmable logic devices (CPLDs), computer hardware, firmware, software, and / or combinations thereof. These various implementations can include: being implemented in one or more computer programs that can be executed and / or interpreted on a programmable system including at least one programmable processor, which can be a special purpose or general purpose programmable processor that can receive data and instructions from a storage system, at least one input device, and at least one output device, and transmit data and instructions to the storage system, the at least one input device, and the at least one output device.

[0140] Computer programs for implementing the methods of the present invention may be written in any combination of one or more programming languages. These computer programs may be provided to a processor of a general-purpose computer, a special-purpose computer, or other programmable data processing device, so that when the computer program is executed by the processor, the functions / operations specified in the flow chart and / or block diagram are implemented. The computer program may be executed entirely on the machine, partially on the machine, partially on the machine and partially on a remote machine as a stand-alone software package, or entirely on a remote machine or server.

[0141] In the context of the present invention, a computer-readable storage medium may be a tangible medium that may contain or store a computer program for use by or in combination with an instruction execution system, device or equipment. A computer-readable storage medium may include, but is not limited to, electronic, magnetic, optical, electromagnetic, infrared, or semiconductor systems, devices or equipment, or any suitable combination of the foregoing. Alternatively, a computer-readable storage medium may be a machine-readable signal medium. A more specific example of a machine-readable storage medium may include an electrical connection based on one or more lines, a portable computer disk, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), an optical fiber, a portable compact disk read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination of the foregoing.

[0142] To provide interaction with a user, the systems and techniques described herein may be implemented on an electronic device having: a display device (e.g., a CRT (cathode ray tube) or LCD (liquid crystal display) monitor) for displaying information to the user; and a keyboard and a pointing device (e.g., a mouse or trackball) through which the user can provide input to the electronic device. Other types of devices may also be used to provide interaction with the user; for example, the feedback provided to the user may be any form of sensory feedback (e.g., visual feedback, auditory feedback, or tactile feedback); and input from the user may be received in any form (including acoustic input, voice input, or tactile input).

[0143] The systems and techniques described herein may be implemented in a computing system that includes backend components (e.g., as a data server), or a computing system that includes middleware components (e.g., an application server), or a computing system that includes frontend components (e.g., a user computer with a graphical user interface or a web browser through which a user can interact with implementations of the systems and techniques described herein), or a computing system that includes any combination of such backend components, middleware components, or frontend components. The components of the system may be interconnected by any form or medium of digital data communication (e.g., a communication network). Examples of communication networks include: a local area network (LAN), a wide area network (WAN), a blockchain network, and the Internet.

[0144] A computing system may include a client and a server. The client and the server are generally remote from each other and usually interact through a communication network. The client and server relationship is generated by computer programs running on the corresponding computers and having a client-server relationship with each other. The server may be a cloud server, also known as a cloud computing server or cloud host, which is a host product in the cloud computing service system to solve the defects of difficult management and weak business scalability in traditional physical hosts and VPS services.

[0145] It should be understood that the various forms of processes shown above can be used to reorder, add or delete steps. For example, the steps described in the present invention can be executed in parallel, sequentially or in different orders, as long as the desired results of the technical solution of the present invention can be achieved, and this document does not limit this.

[0146] The above specific implementations do not constitute a limitation on the protection scope of the present invention. It should be understood by those skilled in the art that various modifications, combinations, sub-combinations and substitutions can be made according to design requirements and other factors. Any modification, equivalent substitution and improvement made within the spirit and principle of the present invention should be included in the protection scope of the present invention.

Claims

1. An obstacle filtering method for an autonomous driving vehicle, characterized in that: include: Acquire current point cloud data in a current point cloud frame collected by the laser radar, and a reference point cloud frame associated with the current point cloud frame, and perform a primary filtering on the current point cloud data according to a preset first filtering range to obtain candidate point cloud data; Inputting the candidate point cloud data into a trained obstacle detection model, outputting the candidate obstacle and the candidate obstacle probability of the obstacle type to which the candidate obstacle belongs, and determining a first reference obstacle and the ground from the candidate obstacles according to the candidate obstacle probability and a preset probability threshold; wherein the candidate obstacle includes at least one candidate point cloud data; Determine reference point cloud data according to the candidate point cloud data, the ground and the first reference obstacle, and determine a second reference obstacle in the current point cloud frame according to the reference point cloud data; Selecting a target obstacle from reference obstacles according to a preset second filtering range and a preset obstacle identification strategy; wherein the reference obstacles include the first reference obstacle and the second reference obstacle; An actual obstacle in the reference point cloud frame is determined, and the target obstacle is updated according to the actual obstacle in the reference point cloud frame to obtain the actual obstacle in the current point cloud frame.

2. The method according to claim 1, characterized in that The step of determining a second reference obstacle in the current point cloud frame according to the reference point cloud data includes: Dividing the reference point cloud data into at least one reference point cloud grid, and determining grid point cloud parameters of each reference point cloud data in the reference point cloud grid; According to the grid point cloud parameters, a second reference obstacle in the current point cloud frame is determined; wherein the second reference obstacle is determined based on the reference point cloud grid.

3. The method according to claim 1, characterized in that The determining a first reference obstacle and a ground from the candidate obstacles according to the candidate obstacle probability and a preset probability threshold comprises: comparing the candidate obstacle probability of the candidate obstacle with the preset probability threshold; If the comparison result is that the probability of the candidate obstacle is greater than or equal to the preset probability threshold, the candidate obstacle is determined to be a first reference obstacle or the ground according to the obstacle type of the candidate obstacle.

4. The method according to claim 1, characterized in that: The step of selecting a target obstacle from the reference obstacles according to a preset second filtering range and a preset obstacle identification strategy includes: Determining a reference reserved obstacle among the reference obstacles according to the preset second filtering range; Based on a preset obstacle recognition strategy, the target obstacle is selected from the reference retained obstacles; wherein the obstacle recognition strategy includes an obstacle retention strategy and an obstacle filtering strategy.

5. The method according to claim 4, characterized in that After determining the reference reserved obstacles among the reference obstacles according to the preset second filtering range, the method further includes: Determining a reference filtering obstacle among the reference obstacles according to the preset second filtering range; Acquire current position data of the autonomous driving vehicle, and determine a road boundary line of a lane where the autonomous driving vehicle is located based on the current position data; Determining a boundary distance between the reference filtered obstacle and the road boundary line; A reference retained obstacle is selected from the reference filtered obstacles according to the boundary distance and a preset boundary distance threshold.

6. The method according to any one of claims 1 to 5, characterized in that The updating of the target obstacle according to the actual obstacle in the reference point cloud frame to obtain the actual obstacle in the current point cloud frame includes: Determining a tracking obstacle corresponding to the target obstacle among the actual obstacles in the reference point cloud frame; The running trajectory of the tracked obstacle in the reference point cloud frame is determined, and the target obstacle is updated according to the running trajectory to obtain the actual obstacle in the current point cloud frame.

7. The method according to any one of claims 1 to 5, characterized in that The preset first filtering range is larger than the preset second filtering range.

8. An obstacle filtering device for an autonomous driving vehicle, characterized in that: include: A point cloud data acquisition module, used to acquire current point cloud data in a current point cloud frame collected by the laser radar, and a reference point cloud frame associated with the current point cloud frame, and perform a primary filtering on the current point cloud data according to a preset first filtering range to obtain candidate point cloud data; a first reference obstacle determination module, configured to input the candidate point cloud data into a trained obstacle detection model, output a candidate obstacle and a candidate obstacle probability of an obstacle type to which the candidate obstacle belongs, and determine a first reference obstacle and a ground surface from the candidate obstacles according to the candidate obstacle probability and a preset probability threshold; wherein the candidate obstacle includes at least one candidate point cloud data; a second reference obstacle determination module, configured to determine reference point cloud data according to the candidate point cloud data, the ground and the first reference obstacle, and determine a second reference obstacle in the current point cloud frame according to the reference point cloud data; a target obstacle determination module, configured to select a target obstacle from reference obstacles according to a preset second filtering range and a preset obstacle identification strategy; wherein the reference obstacles include the first reference obstacle and the second reference obstacle; The actual obstacle determination module is used to determine the actual obstacle in the reference point cloud frame, and update the target obstacle according to the actual obstacle in the reference point cloud frame to obtain the actual obstacle in the current point cloud frame.

9. An electronic device, characterized in that: include: one or more processors; A memory for storing one or more programs; When the one or more programs are executed by the one or more processors, the one or more processors implement an obstacle filtering method for an autonomous driving vehicle as described in any one of claims 1-7.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the program is executed by a processor, an obstacle filtering method for an autonomous driving vehicle as described in any one of claims 1-7 is implemented.

Images