Risk identification method and system for autonomous vehicle facing complex urban scene

Abstract

The application discloses a kind of complex urban scene-oriented automatic driving vehicle risk identification method and system, method includes the following steps: constructing three-dimensional occupancy grid in the local coordinate system of vehicle;Potential obstacle dynamic reachable set is constructed;With the initial set of occlusion grid as the set of invisible grid, the set of occlusion grid of last time is expanded outward according to the dynamic reachable set of each potential obstacle, and the set of occlusion grid after expansion is carried out with the set of invisible grid at current time Space intersection operation, obtain the set of occlusion grid at current time;Three-dimensional anisotropic risk kernel function of visible obstacle and potential obstacle is constructed;According to visible occupancy matrix, occlusion occupancy matrix and three-dimensional anisotropic risk kernel function, the obstacle risk field based on kernel diffusion superposition is constructed.The application has stronger three-dimensional space risk identification ability, improves the safety of vehicle driving under complex urban scene.

Description

Technical Field

[0001] This invention relates to the field of risk identification algorithms for autonomous vehicles, and in particular to a method and system for risk identification of autonomous vehicles in complex urban scenarios. Background Technology

[0002] As autonomous driving technology is gradually implemented, vehicles need to operate safely in urban environments with complex road structures and diverse facilities. Unlike highways or relatively simple road environments, urban scenarios have the following characteristics: First, large areas of obscurity are easily formed at intersections, roadside parking, and building obstructions. These areas are not directly observable at any given moment, but may contain potential traffic participants such as vehicles or pedestrians. These participants are not arbitrarily distributed within these areas but are typically constrained by their speed and traffic rules. Therefore, obstructed areas cannot be ignored, nor can all obscured areas be considered obstructed areas. Second, urban scenarios also widely contain height-related risk factors, such as low bridges and overhead structures. Whether such obstacles pose a risk to vehicles depends not only on their planar location but also on the relative height between the obstacle and the vehicle.

[0003] In existing autonomous driving risk identification technologies, some solutions focus on constructing risks based on detection results of visible areas, while handling occluded areas often employs empirical inflation or conservative assignment strategies, making it difficult to form a stable and reliable risk representation for occluded areas. On the other hand, for risks in the height direction, many solutions directly reduce 3D perception results to a 2D representation, constructing a risk map only in a plane, making it difficult to reliably retain obstacle risks related to vehicle height. Therefore, it is necessary to propose a risk identification method and system for autonomous vehicles in complex urban scenarios.

[0004] To address the risk identification needs of autonomous vehicles in complex urban scenarios, existing technologies typically have the following shortcomings: (1) When modeling the sensor occlusion area, some schemes directly equate the sensor's invisible area with the occlusion area, which can easily cause the occlusion area to expand or change abruptly over time without constraints, resulting in problems such as excessive expansion of potential risks within the occlusion area, high false alarm rate of risks, and unstable risk results.

[0005] (2) In urban scenarios with high-altitude obstacles such as interchanges, ramps, low bridges, and suspended facilities, existing methods often only express risks in a two-dimensional plane, making it difficult to reliably retain information about high-altitude obstacles related to vehicle height in the risk identification results, thus affecting the subsequent planning and control tasks in determining whether there is a risk of collision in the height direction.

[0006] To address the above-mentioned problems, the technical problems to be solved by the present invention include at least the following: The problem of occlusion area risk calculation: How to stably extract occlusion areas from sensor visibility information under the three-dimensional occupancy grid representation, and combine the vehicle motion continuity, the kinematic reachability of potential obstacles and road topology constraints to update the occlusion areas in a time-consistent manner, and the potential occupancy probability distribution that evolves consistently with the vehicle motion within the occlusion area, so as to improve the stability and accuracy of occupancy area risk identification. The problem of preserving height-direction risk: How to construct a three-dimensional risk field in three-dimensional space that can simultaneously characterize ground collision risk and overhead obstacle-related risk, and introduce an aggregation mechanism related to vehicle height when converting to two-dimensional risk expression, so that the risk identification results can still reliably reflect the height-direction risk characteristics related to vehicle geometry at the two-dimensional level, and avoid the loss of three-dimensional information in the two-dimensional conversion process. Summary of the Invention

[0007] The main objective of this invention is to solve the above-mentioned problems and provide a method and system for risk identification of autonomous vehicles in complex urban scenarios, which has stronger three-dimensional spatial risk identification capabilities and can also improve the safety of vehicle driving in complex urban scenarios.

[0008] The technical solution adopted in this invention is: A method for risk identification of autonomous vehicles in complex urban scenarios is provided, including the following steps: Based on the vehicle information and vehicle pose information collected by the sensors, a three-dimensional occupancy grid is constructed in the local coordinate system of the vehicle. Visibility is determined based on the relationship between the sensor rays and obstacle occlusion, and the grid is divided into two sets, including the visible grid set and the invisible grid set. Construct the dynamic reachability set of potential obstacles, including the dynamic reachability sets of vehicle obstacles and pedestrian obstacles per unit time; The invisible grid set is used as the initial occlusion grid set. The occlusion grid set of the previous time step is expanded outward according to the dynamic reachability set of various potential obstacles. The expanded occlusion grid set is then spatially intersected with the invisible grid set of the current time step to obtain the occlusion grid set of the current time step. The visible occupancy matrix is ​​obtained based on the probability of each visible region occupying a grid in the 3D occupancy grid, and the occupancy matrix is ​​obtained based on the probability of each occupancy region occupying a grid. Construct a three-dimensional anisotropic risk kernel function for visible and potential obstacles; An obstacle risk field based on kernel diffusion superposition is constructed using the visible occupancy matrix, the occupancy matrix, and the three-dimensional anisotropic risk kernel function.

[0009] Following the above technical solution, the method further includes the step of: aggregating three-dimensional risks within a preset height range where risks may occur to obtain a two-dimensional planning layer risk field.

[0010] According to the above technical solution, the vehicle information includes vehicle surrounding environment data collected by the vehicle-mounted LiDAR and video image data of the vehicle surrounding environment captured by the depth camera.

[0011] Following the above technical solution, the visible grid set specifically refers to the grids within the sensor's detection radius that are not obstructed; the invisible grid set specifically refers to the invisible grids within the sensor's detection radius that are in blind spots caused by buildings, large vehicles, or road geometry.

[0012] Following the above technical solution, the specific calculation process for the probability of each grid cell occupying the occupancy area is as follows: based on the overlap ratio between the occupancy grid expansion set at the previous moment and the invisible grid cell set at the current moment, the probability of each occupancy grid cell being occupied by a potential obstacle is calculated and normalized to obtain the probability that a potential obstacle exists in a grid cell within the occupancy area.

[0013] Following the above technical solution, the obstacle risk field based on nuclear diffusion superposition specifically includes a weighted sum of a motion collision risk field and an occlusion potential risk field. The motion collision risk field is specifically constructed based on the visible occupancy matrix and the three-dimensional anisotropic risk kernel function of visible obstacles; the occupancy potential risk field is specifically constructed based on the occupancy matrix and the three-dimensional anisotropic risk kernel function.

[0014] Following the above technical solution, the corresponding weighting coefficients of the motion collision risk field and the occlusion potential risk field are determined based on the proportion of the visible area and the occluded area in the three-dimensional occupied grid.

[0015] Following the above technical solution, the preset potential risk height range is specifically calculated based on the ground elevation at the vehicle's ground coordinates, the vehicle's body height, and the safety height margin above the vehicle, determining the potential risk height range at the vehicle's current location.

[0016] This invention also provides a risk identification system for autonomous vehicles in complex urban scenarios, specifically including: The grid set division module is used to construct a three-dimensional occupying grid in the local coordinate system of the vehicle based on the vehicle information and vehicle pose information collected by the sensors. It determines the visibility based on the relationship between the sensor rays and the obstacle occlusion and divides the grid into two sets, including the visible grid set and the invisible grid set. The Dynamic Reachability Set Construction Module is used to construct the dynamic reachability set of potential obstacles, including the dynamic reachability set of vehicle obstacles and pedestrian obstacles per unit time. The occlusion grid set update module is used to take the invisible grid set as the initial occlusion grid set, expand the occlusion grid set of the previous time step outward according to the dynamic reachability set of various potential obstacles, and perform a spatial intersection operation between the expanded occlusion grid set and the invisible grid set of the current time step to obtain the occlusion grid set of the current time step. The obstacle risk field construction module is used to obtain the visible occupancy matrix based on the probability of each occupied grid in the visible region of the 3D occupancy grid, and to obtain the occupancy matrix based on the probability of each occupied grid in the occupancy region; to construct the 3D anisotropic risk kernel function for visible obstacles and potential obstacles; and to construct an obstacle risk field based on kernel diffusion superposition based on the visible occupancy matrix, the occupancy matrix and the 3D anisotropic risk kernel function.

[0017] The present invention also provides a computer storage medium storing a computer program executable by a processor, the computer program being used to implement the risk identification method for autonomous vehicles in complex urban scenarios described in the above technical solution.

[0018] The beneficial effects of this invention are as follows: This invention constructs a dynamic reachability set model by combining road topology constraints and the motion states of potential obstacles. To ensure the temporal consistency of the occupancy area as the vehicle moves, an occupancy grid set update rule is proposed, ultimately yielding the probability distribution of potential obstacle occupancy within the occupancy area. This avoids simply equating sensor-invisible areas with high-risk areas or excessive risk expansion caused by empirical inflation, making the evolution of occupancy-related risks more continuous over time and the risk boundaries more reasonable, thereby reducing false alarms in occupancy areas and improving the reliability of risk identification results.

[0019] Furthermore, traditional two-dimensional risk field models, due to the loss of height information, cannot accurately represent the risks posed by high-altitude obstacles such as bridges, suspended structures, and traffic signs. This invention constructs a three-dimensional multi-source driving risk field in three-dimensional space, capable of simultaneously expressing the collision risk from ground obstacles and the height-direction risk caused by high-altitude road facilities.

[0020] Furthermore, when using three-dimensional risks to express two-dimensional planning layers, risks are aggregated only within the height range related to vehicle height. This ensures that the two-dimensional risk results can still reflect the key risk characteristics of whether high-altitude obstacles may collide with vehicles, avoiding the problem of weakening or losing high-altitude risks in scenarios such as low bridges, suspended facilities, and underpasses in traditional two-dimensional risk maps.

[0021] Furthermore, the two-dimensional or three-dimensional risk field, i.e., the risk identification result, output by this invention can be provided to the subsequent planning and control module in the form of a grid or function query, enabling rapid query of the risk level at any spatial location and providing more effective risk quantification information for the safe driving of autonomous vehicles.

[0022] Of course, any product implementing this invention does not necessarily need to achieve all of the advantages described above at the same time. Attached Figure Description

[0023] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0024] Figure 1 This is a flowchart of the risk identification method for autonomous vehicles in complex urban scenarios according to an embodiment of the present invention. Figure 1 ; Figure 2 This is a flowchart of the risk identification method for autonomous vehicles in complex urban scenarios according to an embodiment of the present invention. Figure 2 ; Figure 3 This is a schematic diagram of a scenario illustrating the occlusion grid set update rules in an embodiment of the present invention; Figure 4 This is a schematic diagram showing the distribution of the three-dimensional anisotropic risk kernel function of visible obstacles in an embodiment of the present invention; Figure 5 This is a schematic diagram showing the distribution of the three-dimensional anisotropic risk kernel function of potential obstacles in an embodiment of the present invention. Detailed Implementation

[0025] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the invention.

[0026] It should be noted that the illustrations provided in the embodiments of the present invention are only schematic representations of the basic concept of the present invention. Therefore, the illustrations only show the components related to the present invention and are not drawn according to the number, shape and size of the components in actual implementation. In actual implementation, the form, quantity and proportion of each component can be arbitrarily changed, and the layout of the components may also be more complex.

[0027] In this invention, it should also be noted that the terms "center," "upper," "lower," "left," "right," "vertical," "horizontal," "inner," and "outer," etc., indicate the orientation or positional relationship based on the orientation or positional relationship shown in the accompanying drawings. They are used only for the convenience of describing this application and simplifying the description, and do not indicate or imply that the device or element referred to must have a specific orientation, or be constructed and operated in a specific orientation. Therefore, they should not be construed as limitations on this application. Furthermore, the terms "first" and "second" are used only for descriptive and distinguishing purposes and should not be construed as indicating or implying relative importance.

[0028] Furthermore, it should be noted that the features of the various embodiments of the present invention can be combined or integrated in whole or in part, and as those skilled in the art will understand, they can interact and operate in different ways. Each embodiment can be implemented independently of each other or in association with one another.

[0029] like Figure 1 As shown, the risk identification method for autonomous vehicles in complex urban scenarios according to an embodiment of the present invention includes the following steps: S1. Based on the vehicle information and vehicle pose information collected by the sensors, a three-dimensional occupancy grid is constructed in the local coordinate system of the vehicle. Visibility is determined based on the relationship between the sensor rays and the obstacle occlusion. The grid is divided into two sets, including the visible grid set and the invisible grid set. S2. Construct the dynamic reachability set of potential obstacles, including the dynamic reachability set of vehicle obstacles and pedestrian obstacles per unit time. S3. Using the invisible grid set as the initial occlusion grid set, expand the occlusion grid set of the previous time step outward according to the dynamic reachability set of various potential obstacles, and perform a spatial intersection operation between the expanded occlusion grid set and the invisible grid set of the current time step to obtain the occlusion grid set of the current time step. S4. Obtain the visible occupation matrix based on the probability of each visible region occupying a grid in the 3D occupation grid, and obtain the occlusion occupation matrix based on the probability of each occupied region occupying a grid. S5. Construct the three-dimensional anisotropic risk kernel function for visible and potential obstacles; S6. Construct an obstacle risk field based on kernel diffusion superposition using the visible occupancy matrix, the occupancy matrix, and the three-dimensional anisotropic risk kernel function.

[0030] Furthermore, such as Figure 2 As shown, the method further includes step S7: aggregating the three-dimensional risks within a preset height range where risks may occur to obtain a two-dimensional planning layer risk field.

[0031] The construction of the 3D occupancy grid in step S1 mainly includes the following steps: Based on sensor information such as vehicle-mounted LiDAR and depth camera, and the vehicle's pose information, a 3D occupancy grid is constructed in the vehicle's local coordinate system. Simultaneously, visibility is determined based on the relationship between the sensor ray and obstacle occlusion, dividing the grid into two sets: the set of visible grids within the sensor detection radius that are not occluded. And the set of invisible grid cells within the sensor's detection radius but in blind spots caused by buildings, large vehicles, or road geometry. .

[0032] This invention mainly divides occlusion regions based on dynamic reachability sets. The construction of the dynamic reachability set in step S2 mainly includes the following steps: To ensure that the potential occupancy inference of the occupancy area satisfies the road traffic prior, potential obstacles must include at least two categories: vehicles and pedestrians. For vehicle obstacles, their motion is constrained by lane topology, advancing longitudinally along the lane direction and laterally offset by the lane width. Constraints. Assume that vehicle-type obstacles occur at time... The position is Then vehicle-type obstacles are in The dynamically reachable set within can be written as:

[0033] in, The maximum speed of the vehicle. Lane width; and Let be the unit vectors for the tangential and normal directions of the lane.

[0034] For pedestrian obstacles, their activity area is restricted to the vicinity of the sidewalk and zebra crossing, and an omnidirectional dynamic model is adopted. Let the pedestrian obstacle at time... The position is Then, human obstacles in The dynamically reachable set within can be written as:

[0035] in, The reachable radius.

[0036] The process of constructing the occlusion grid set in step S3 mainly involves: To maintain temporal consistency of the occlusion area as the vehicle moves at different times, this invention proposes an occlusion grid set update rule, such as... Figure 3 As shown. Specifically, in The invisible raster set at time As the initial set of occlusion grids .exist At that time, the set of occluded grid cells from the previous time step. ( Figure 3 (a) The right-side shaded area is expanded outward according to the dynamic reachability set model of potential obstacles for pedestrians and vehicles. After expansion, as shown in the figure... Figure 3 As shown in (b), the dotted filled region is an expanded grid set of vehicle obstacles, constrained by lane topology; the circular filled region is an expanded grid set of pedestrian obstacles, limited to the vicinity of sidewalks and zebra crossings. The expanded grid set is then compared with the currently invisible grid set. Perform spatial intersection operation to obtain, as follows Figure 3 (c) shows the set of occlusion grids at the current time. This avoids treating the entire set of invisible grid cells as an occluded grid cell set.

[0037] The rule for updating the occlusion grid set ensures that only those occlusion grids updated at time [time] are updated. Still not visible, and can be seen from the previous moment Only the grid cells reachable by various potential obstacles within the occluded area within a single time step are retained as the current set of occluded grid cells. Once a grid cell in a certain area becomes visible at a subsequent time, it automatically reverts to its previous state. The occlusion set is removed, thus ensuring temporal consistency as the vehicle moves. After obtaining... Then, expand the set based on its occlusion grid at the previous time step. With the set of invisible grid cells at the current time The overlap ratio is used to calculate the probability that a potential obstacle occupies each occluded grid cell, and this probability is normalized to obtain the probability that a potential obstacle exists in a grid cell within the occluded area. .

[0038] Furthermore, the probability that a grid cell within the visible area of ​​a 3D occupied grid is occupied. And the probability that there are potential obstacles in the grid within the occluded area. Subsequently, in order to simultaneously describe the motion collision risk of ground obstacles and the height-direction risk caused by overhead obstacles such as elevated roads, low bridges, and suspended facilities in complex urban scenarios, this invention constructs a three-dimensional multi-source driving risk field, mainly including the following steps: (1) Binarization modeling of the grid: To facilitate risk calculation within a unified framework, the visible area is divided into grids. And the occupancy of the grid area Perform consistent occupancy representation. This is based on the probability of each occupied grid cell within the visible area. Obtain the visible occupancy matrix:

[0039] Based on the probability that each occupied area occupies a grid cell Obtain the occupancy matrix:

[0040] (2) Three-dimensional anisotropic risk kernel function: It is evident that obstacles and potential obstacles within occluded areas possess different risk characteristics. If the same risk kernel function is used to model different risk sources, it will be difficult to accurately describe the actual impact of various risk sources on the overall risk of the scene. Furthermore, the risk propagation characteristics of the same risk source differ in the height and planar directions. Therefore, this invention constructs differentiated anisotropic risk kernel functions for different types of obstacles, enabling them to accurately characterize the risk propagation patterns of various obstacles in three-dimensional space. The three-dimensional anisotropic risk kernel function for the visible obstacle is as follows:

[0041] in, This is the amplitude coefficient of the visible obstacle risk kernel function, used to adjust the risk intensity of visible obstacles in space; and This represents the spatial scale coefficient of the visible obstacle risk kernel function, used to control the spatial extent of the risk posed by visible obstacles. The three-dimensional anisotropic risk kernel function distribution of visible obstacles is shown below. Figure 4 As shown, the kernel function makes the risk value reach its maximum in the central region of the obstacle, and then gradually decreases outward, forming a continuous risk distribution centered on the obstacle itself.

[0042] Compared to visible obstacles, the risks posed by potential obstacles in occluded areas have several characteristics: the source of the risk is unobservable, and the spatial distribution of risk propagation is ambiguous and widespread. Therefore, this invention modifies the correlation coefficient of the visible obstacle risk kernel function based on empirical values ​​to adapt it to potential obstacles in occluded areas:

[0043] in, The amplitude coefficient of the risk kernel function for occluded potential obstacles is lower than that for static obstacles, thus reflecting the lower confidence level of potential risks compared to visible risks. The spatial scale coefficients are used to represent the risk kernel function for occlusion potential obstacles. The three-dimensional anisotropic risk kernel function distribution of potential obstacles is shown below. Figure 5 As shown, compared to the visible obstacle risk kernel function, the potential obstacle risk kernel function has a wider risk coverage and a gentler risk boundary, which can form a low-intensity but wide-coverage risk distribution in space.

[0044] (3) Construction of a three-dimensional multi-source driving risk field based on nuclear diffusion superposition: An obstacle risk field modeling algorithm based on kernel diffusion superposition is applied to the visible occupancy matrix, the occupancy matrix, and each risk kernel function to obtain the motion collision risk field. and blocking potential risk fields :

[0045] By using a weighted overlay method, risks from different sources are unified into a three-dimensional space, generating a three-dimensional multi-source driving risk field. :

[0046] in, , These are weighting coefficients for the risk fields of visible obstacles and potential occlusion obstacles, respectively. The values ​​of these weighting coefficients are determined based on the proportions of the visible and occlusion regions within the 3D occupied grid.

[0047] in, The confidence coefficient for the occluded area. .

[0048] Furthermore, in obtaining the three-dimensional multi-source driving risk field Subsequently, to facilitate the application of three-dimensional risks in vehicle planning within a two-dimensional motion plane, and to avoid the mis-aggregation of irrelevant risks in the height direction into the two-dimensional planning plane during the two-dimensional dimensionality reduction process, thus preventing unreasonable impacts, this invention proposes a two-dimensional planning layer risk construction method related to vehicle height. Step S7 mainly includes the following steps: Let the vehicle's coordinates on the ground be... The ground elevation at that location is The vehicle's body height is The safety margin above the vehicle is Define the vehicle at this location. The range of possible risks for:

[0049] Aggregating the three-dimensional risks within this height range yields a two-dimensional planning layer risk field. :

[0050] The risk field of this two-dimensional planning layer ensures that the three-dimensional risks of obstacles such as low bridges and suspension facilities that are within the vehicle height range and may collide with the vehicle are aggregated into the two-dimensional planning layer. The risks of obstacles that are significantly higher than the vehicle height range and are unlikely to collide with the vehicle are not aggregated into the two-dimensional planning layer.

[0051] Through the above steps, a three-dimensional multi-source driving risk field is obtained. The risk field of the two-dimensional planning layer related to its vehicle body height range . and All of these can be used as risk identification results output by the present invention, and can be used to assess, screen and constrain the risk level of different spatial locations.

[0052] As can be seen, this invention, based on a dynamic reachability set-based occlusion region partitioning method, constructs a dynamic reachability set model by combining road topology constraints and the motion state of potential obstacles. To ensure the temporal consistency of occlusion regions as the vehicle moves, an occlusion grid set update rule is proposed, ultimately yielding the probability distribution of potential obstacle occupancy within the occlusion region. This avoids simply equating sensor-invisible areas with high-risk areas or excessive risk expansion caused by empirical inflation, making the evolution of occlusion-related risks more continuous over time and the risk boundaries more reasonable, thereby reducing false alarms in occlusion region risks and improving the reliability of risk identification results.

[0053] Traditional two-dimensional risk field models, due to the loss of height information, cannot accurately characterize the risks of high-altitude obstacles such as bridges, suspended structures, and traffic signs. This invention constructs a three-dimensional multi-source driving risk field in three-dimensional space, capable of simultaneously expressing the collision risk from ground obstacles and the height-direction risk caused by road overhead facilities. When using the three-dimensional risk for expression in the two-dimensional planning layer, risks are aggregated only within the height range relevant to vehicle height. This ensures that the two-dimensional risk results still reflect the key risk characteristics of whether high-altitude obstacles may collide with vehicles, avoiding the weakening or loss of high-altitude risks in scenarios such as low bridges, suspended structures, and underpasses in traditional two-dimensional risk maps. The risk identification results output by this invention include a three-dimensional risk field. With two-dimensional risk field It can be provided to subsequent planning and control modules in the form of grid or function queries, enabling rapid querying of the risk level at any spatial location, and providing more effective risk quantification information for the safe driving of autonomous vehicles.

[0054] To implement the above method embodiments, the present invention also provides a risk identification system for autonomous vehicles in complex urban scenarios, specifically including: The grid set division module is used to construct a three-dimensional occupying grid in the local coordinate system of the vehicle based on the vehicle information and vehicle pose information collected by the sensors. It determines the visibility based on the relationship between the sensor rays and the obstacle occlusion and divides the grid into two sets, including the visible grid set and the invisible grid set. The Dynamic Reachability Set Construction Module is used to construct the dynamic reachability set of potential obstacles, including the dynamic reachability set of vehicle obstacles and pedestrian obstacles per unit time. The occlusion grid set update module is used to take the invisible grid set as the initial occlusion grid set, expand the occlusion grid set of the previous time step outward according to the dynamic reachability set of various potential obstacles, and perform a spatial intersection operation between the expanded occlusion grid set and the invisible grid set of the current time step to obtain the occlusion grid set of the current time step. The obstacle risk field construction module is used to obtain the visible occupancy matrix based on the probability of each occupied grid in the visible region of the 3D occupancy grid, and to obtain the occupancy matrix based on the probability of each occupied grid in the occupancy region; to construct the 3D anisotropic risk kernel function for visible obstacles and potential obstacles; and to construct an obstacle risk field based on kernel diffusion superposition based on the visible occupancy matrix, the occupancy matrix and the 3D anisotropic risk kernel function.

[0055] Each module is mainly used to implement the above methods and steps, and will not be described in detail here.

[0056] This application also provides a computer-readable storage medium, such as flash memory, hard disk, multimedia card, card-type memory (e.g., SD or DX memory), random access memory (RAM), static random access memory (SRAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), programmable read-only memory (PROM), magnetic memory, disk, optical disk, server, App application store, etc., which stores a computer program. When the program is executed by a processor, it implements a corresponding function. This embodiment describes a risk identification method for autonomous vehicles in complex urban scenarios, implemented by a computer-readable storage medium executed by a processor.

[0057] It should be noted that, depending on the implementation needs, the various steps / components described in this application can be broken down into more steps / components, or two or more steps / components or parts of the operation of steps / components can be combined into new steps / components to achieve the purpose of this invention.

[0058] The order of the steps in the above embodiments does not imply the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of this application.

[0059] It should be understood that those skilled in the art can make improvements or modifications based on the above description, and all such improvements and modifications should fall within the protection scope of the appended claims.

Claims

1. A method for risk identification of autonomous vehicles in complex urban scenarios, characterized in that, Includes the following steps: Based on the vehicle information and vehicle pose information collected by the sensors, a three-dimensional occupancy grid is constructed in the local coordinate system of the vehicle. Visibility is determined based on the relationship between the sensor rays and obstacle occlusion, and the grid is divided into two sets, including the visible grid set and the invisible grid set. Construct the dynamic reachability set of potential obstacles, including the dynamic reachability sets of vehicle obstacles and pedestrian obstacles per unit time; The invisible grid set is used as the initial occlusion grid set. The occlusion grid set of the previous time step is expanded outward according to the dynamic reachability set of various potential obstacles. The expanded occlusion grid set is then spatially intersected with the invisible grid set of the current time step to obtain the occlusion grid set of the current time step. The visible occupancy matrix is ​​obtained based on the probability of each visible region occupying a grid in the 3D occupancy grid, and the occupancy matrix is ​​obtained based on the probability of each occupancy region occupying a grid. Construct a three-dimensional anisotropic risk kernel function for visible and potential obstacles; An obstacle risk field based on kernel diffusion superposition is constructed using the visible occupancy matrix, the occupancy matrix, and the three-dimensional anisotropic risk kernel function.

2. The method for risk identification of autonomous vehicles in complex urban scenarios according to claim 1, characterized in that, The method also includes the step of aggregating three-dimensional risks within a preset height range where risks may occur, to obtain a two-dimensional planning layer risk field.

3. The method for risk identification of autonomous vehicles in complex urban scenarios according to claim 1, characterized in that, Vehicle information includes environmental data collected by the vehicle's onboard LiDAR and video images of the vehicle's surroundings captured by a depth camera.

4. The method for risk identification of autonomous vehicles in complex urban scenarios according to claim 1, characterized in that, The visible grid set specifically refers to the grids within the sensor's detection radius that are not obstructed; the invisible grid set specifically refers to the invisible grids within the sensor's detection radius that are in blind spots due to buildings, large vehicles, or road geometry.

5. The method for risk identification of autonomous vehicles in complex urban scenarios according to claim 1, characterized in that, The specific calculation process for the probability of each grid cell in the occupied area is as follows: based on the overlap ratio between the expanded set of occupied grid cells at the previous time and the set of invisible grid cells at the current time, the probability of each occupied grid cell being occupied by a potential obstacle is calculated and normalized to obtain the probability that a potential obstacle exists in a grid cell in the occupied area.

6. The method for risk identification of autonomous vehicles in complex urban scenarios according to claim 1, characterized in that, The obstacle risk field based on nuclear diffusion superposition specifically includes a weighted sum of a motion collision risk field and an occlusion potential risk field. The motion collision risk field is specifically constructed based on the visible occupancy matrix and the three-dimensional anisotropic risk kernel function of visible obstacles; the occupancy potential risk field is specifically constructed based on the occupancy matrix and the three-dimensional anisotropic risk kernel function.

7. The method for risk identification of autonomous vehicles in complex urban scenarios according to claim 6, characterized in that, The corresponding weighting coefficients for the motion collision risk field and the occlusion potential risk field are determined based on the proportion of the visible and occluded regions in the 3D occupied grid.

8. The method for risk identification of autonomous vehicles in complex urban scenarios according to claim 2, characterized in that, The preset potential risk height range is specifically calculated based on the ground elevation at the vehicle's ground coordinates, the vehicle's body height, and the safety height margin above the vehicle, determining the potential risk height range at the vehicle's current location.

9. A risk identification system for autonomous vehicles in complex urban scenarios, characterized in that, Specifically, it includes: The grid set division module is used to construct a three-dimensional occupying grid in the local coordinate system of the vehicle based on the vehicle information and vehicle pose information collected by the sensors. It determines the visibility based on the relationship between the sensor rays and the obstacle occlusion and divides the grid into two sets, including the visible grid set and the invisible grid set. The Dynamic Reachability Set Construction Module is used to construct the dynamic reachability set of potential obstacles, including the dynamic reachability set of vehicle obstacles and pedestrian obstacles per unit time. The occlusion grid set update module is used to take the invisible grid set as the initial occlusion grid set, expand the occlusion grid set of the previous time step outward according to the dynamic reachability set of various potential obstacles, and perform a spatial intersection operation between the expanded occlusion grid set and the invisible grid set of the current time step to obtain the occlusion grid set of the current time step. The obstacle risk field construction module is used to obtain the visible occupancy matrix based on the probability of each occupied grid in the visible region of the 3D occupancy grid, and to obtain the occupancy matrix based on the probability of each occupied grid in the occupancy region; to construct the 3D anisotropic risk kernel function for visible obstacles and potential obstacles; and to construct an obstacle risk field based on kernel diffusion superposition based on the visible occupancy matrix, the occupancy matrix and the 3D anisotropic risk kernel function.

10. A computer storage medium, characterized in that, It contains a computer program that can be executed by a processor, which is used to implement the risk identification method for autonomous vehicles in complex urban scenarios according to any one of claims 1-7.

Images