Vehicle perception information processing method, device and electronic equipment

Abstract

The present invention discloses a method, device and electronic device for processing vehicle perception information. The method relates to the field of autonomous driving, and includes: obtaining perception information collected by an autonomous driving vehicle; dividing a preset space corresponding to the perception information according to a preset direction to obtain a plurality of first grids, wherein the plurality of first grids are non-uniformly distributed in the preset space; performing obstacle recognition on the perception information based on the sub-perception information corresponding to the plurality of first grids to obtain a recognition result, wherein the recognition result is used to characterize at least one of the following: whether there are obstacles in the preset space, the semantic category of the existing obstacles, and the location of the existing obstacles. The present invention solves the technical problem in the related art of high-precision processing of perception information, resulting in large consumption of computing resources.

Description

Technical Field

[0001] The present invention relates to the field of autonomous driving, and in particular to a method, device and electronic device for processing vehicle perception information. Background Art

[0002] In the field of autonomous driving, processing the perception information obtained by the vehicle can help the autonomous driving system understand the surrounding environment more accurately and improve driving safety and efficiency. By processing the perception information, the vehicle can monitor the road conditions and the location of other vehicles in real time to avoid collisions and traffic accidents. It can also help the vehicle make smarter driving decisions, such as choosing the best path, speed and lane, and improving driving comfort and energy efficiency. Effective processing of vehicle perception information can improve the performance and reliability of the autonomous driving system, and provide important support for achieving safe and efficient autonomous driving.

[0003] At present, in order to improve the accuracy of processing vehicle perception information, it is necessary to perform high-resolution grid division on the perception information, which takes up a lot of resources and takes a long time to calculate. It is not conducive to deployment on vehicles, resulting in high-precision processing of perception information and large consumption of computing resources.

[0004] To address the above-mentioned problems, no effective solution has been proposed yet. Summary of the invention

[0005] The embodiments of the present invention provide a method, device and electronic device for processing vehicle perception information, so as to at least solve the technical problem in the related art that high-precision processing of perception information leads to large consumption of computing resources.

[0006] According to one aspect of an embodiment of the present invention, a method for processing vehicle perception information is provided, including: acquiring perception information collected by an autonomous driving vehicle; dividing a preset space corresponding to the perception information according to a preset direction to obtain a plurality of first grids, wherein the plurality of first grids are non-uniformly distributed in the preset space; and performing obstacle recognition on the perception information based on sub-perception information corresponding to the plurality of first grids to obtain a recognition result, wherein the recognition result is used to characterize at least one of the following: whether there is an obstacle in the preset space, the semantic category of the existing obstacle, and the position of the existing obstacle.

[0007] Optionally, the method also includes: determining the recognition requirements of the autonomous driving vehicle, wherein the recognition requirements are related to at least one of the following information: recognition accuracy, recognition efficiency; based on the recognition requirements, determining a preset direction from multiple directions, wherein the multiple directions include: a horizontal direction and a vertical direction.

[0008] Optionally, based on the recognition requirements, a preset direction is determined from multiple directions, including: when the recognition requirement is that the recognition accuracy reaches the target size accuracy, determining that the preset direction includes a vertical direction, and the target size accuracy is less than a first accuracy value; when the recognition requirement is that the recognition efficiency reaches the target recognition efficiency, determining that the preset direction includes a horizontal direction, and the target recognition efficiency is greater than the first efficiency value.

[0009] Optionally, based on the sub-perception information corresponding to the multiple first grids, obstacle recognition is performed on the perception information to obtain a recognition result, including: obtaining the current driving scene of the autonomous driving vehicle; inputting the sub-perception information corresponding to the multiple first grids into an obstacle recognition model, controlling the obstacle recognition model to perform obstacle recognition according to the current driving scene, and obtaining sub-recognition results of the multiple second grids, wherein the recognition result includes the sub-recognition results of the multiple second grids, and the scales of the multiple second grids match the current driving scene.

[0010] Optionally, the sub-perception information corresponding to the plurality of first grids is input into an obstacle recognition model, and the obstacle recognition model is controlled to perform obstacle recognition according to the current driving scene to obtain sub-recognition results of the plurality of second grids, including: determining a target output scale from a plurality of output scales based on the current driving scene, wherein different output scales correspond to different driving scenes; and the sub-perception information corresponding to the plurality of first grids is input into the obstacle recognition model, and the obstacle recognition model is controlled to perform obstacle recognition according to the target output scale to obtain sub-recognition results of the plurality of second grids.

[0011] Optionally, the sub-perception information corresponding to the plurality of first grids is input into an obstacle recognition model, and the obstacle recognition model is controlled to perform obstacle recognition according to the current driving scene to obtain sub-recognition results of the plurality of second grids, including: the sub-perception information corresponding to the plurality of first grids is input into the obstacle recognition model, and the obstacle recognition model is controlled to perform obstacle recognition according to the plurality of output scales to obtain sub-recognition result sets corresponding to the plurality of output scales, wherein different sub-recognition result sets contain sub-recognition results of the second grids at different output scales; based on the current driving scene, a sub-recognition result set corresponding to a target output scale is determined from the sub-recognition result set to obtain sub-recognition results of the plurality of second grids, wherein the target output scale corresponds to the current driving scene.

[0012] Optionally, the loss function of the obstacle recognition model is obtained by weighted processing of scale loss functions corresponding to multiple output scales, and the scale loss function is obtained by weighted processing of the grid loss function of the second grid corresponding to the corresponding output scale. The grid loss function of the second grid is determined based on the real recognition result of the second grid and the training recognition result of the second grid. The training recognition result is obtained by controlling the obstacle recognition model to perform obstacle recognition on the training data according to the corresponding output scale.

[0013] Optionally, the weight value corresponding to the scale loss function is determined based on the number of grids corresponding to the minimum output scale in the second grid corresponding to the corresponding output scale; and / or, the weight value corresponding to the grid loss function is determined based on the distance between the corresponding second grid and a preset position in a preset space.

[0014] According to another aspect of an embodiment of the present invention, a device for processing vehicle perception information is also provided, including: an acquisition module, used to acquire perception information collected by an autonomous driving vehicle; a division module, used to divide a preset space corresponding to the perception information according to a preset direction to obtain a plurality of first grids, wherein the plurality of first grids are non-uniformly distributed in the preset space; an identification module, used to perform obstacle identification on the perception information based on sub-perception information corresponding to the plurality of first grids to obtain an identification result, wherein the identification result is used to characterize at least one of the following: whether there is an obstacle in the preset space, the semantic category of the existing obstacle, and the position of the existing obstacle.

[0015] According to another aspect of an embodiment of the present invention, there is further provided an electronic device, comprising: a memory storing an executable program; and a processor for running the program, wherein the method in each embodiment of the present invention is executed when the program is running.

[0016] According to another aspect of an embodiment of the present invention, a computer-readable storage medium is provided. The computer-readable storage medium includes a stored executable program, wherein when the executable program is running, the device where the computer-readable storage medium is located is controlled to execute the methods in various embodiments of the present invention.

[0017] According to another aspect of an embodiment of the present invention, a computer program product is provided, including a computer program. When the computer program is executed by a processor, the method in each embodiment of the present invention is implemented.

[0018] According to another aspect of an embodiment of the present invention, a computer program product is provided, including a non-volatile computer-readable storage medium, wherein the non-volatile computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, the method in each embodiment of the present invention is implemented.

[0019] According to another aspect of the embodiments of the present invention, a computer program is provided. When the computer program is executed by a processor, the methods in the embodiments of the present invention are implemented.

[0020] In an embodiment of the present invention, a method for processing vehicle perception information is provided, including: acquiring perception information collected by an autonomous driving vehicle; dividing a preset space corresponding to the perception information according to a preset direction to obtain a plurality of first grids, wherein the plurality of first grids are non-uniformly distributed in the preset space; and performing obstacle recognition on the perception information based on sub-perception information corresponding to the plurality of first grids to obtain a recognition result, wherein the recognition result is used to characterize at least one of the following: whether there is an obstacle in the preset space, the semantic category of the existing obstacle, and the position of the existing obstacle. It is easy to notice that in actual road conditions, obstacles in different areas are unevenly distributed, and more processing resources are needed to process dense obstacle areas. Dividing the preset space corresponding to the perception information into multiple non-uniformly distributed first grids according to the preset direction can help improve the processing efficiency of the perception information and avoid invalid processing of the entire space. Multiple non-uniformly distributed grids can better adapt to complex road environments, enable the vehicle system to perceive the surrounding environment more accurately, and invest more resources and energy in important areas, avoiding excessive processing of unimportant areas, thereby reducing the computational burden, which can improve the accurate identification of obstacles, while reducing the processing volume of the entire space and saving computing resources, thereby solving the technical problem of high-precision processing of perception information in related technologies, resulting in large consumption of computing resources. BRIEF DESCRIPTION OF THE DRAWINGS

[0021] The drawings described herein are used to provide a further understanding of the present invention and constitute a part of this application. The exemplary embodiments of the present invention and their descriptions are used to explain the present invention and do not constitute an improper limitation of the present invention. In the drawings:

[0022] Figure 1 is a flow chart of a method for processing vehicle perception information according to an embodiment of the present invention;

[0023] Figure 2 is a schematic diagram of an optional process of processing vehicle perception information according to an embodiment of the present invention;

[0024] Figure 3 is a schematic diagram of an optional horizontal division of a preset space according to an embodiment of the present invention;

[0025] Figure 4 is a schematic diagram of an optional vertical division of a preset space according to an embodiment of the present invention;

[0026] Figure 5 is a schematic diagram of an optional method for determining a loss function of an obstacle recognition model according to an embodiment of the present invention;

[0027] Figure 6is a schematic diagram corresponding to an optional model training process scale according to an embodiment of the present invention;

[0028] Figure 7 is a schematic diagram of an optional scale adjustment in the model application stage according to an embodiment of the present invention;

[0029] Figure 8 is a schematic diagram of a vehicle perception information processing device according to an embodiment of the present invention. DETAILED DESCRIPTION

[0030] In order to enable those skilled in the art to better understand the scheme of the present invention, the technical scheme in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work should fall within the scope of protection of the present invention.

[0031] It should be noted that the terms "first", "second", etc. in the specification and claims of the present invention and the above-mentioned drawings are used to distinguish similar objects, and are not necessarily used to describe a specific order or sequence. It should be understood that the data used in this way can be interchanged where appropriate, so that the embodiments of the present invention described herein can be implemented in an order other than those illustrated or described herein. In addition, the terms "including" and "having" and any variations thereof are intended to cover non-exclusive inclusions, for example, a process, method, system, product or device that includes a series of steps or units is not necessarily limited to those steps or units that are clearly listed, but may include other steps or units that are not clearly listed or inherent to these processes, methods, products or devices.

[0032] According to one aspect of an embodiment of the present invention, a method for processing vehicle perception information is provided. It should be noted that the steps shown in the flowchart of the accompanying drawings can be executed in a computer system such as a set of computer executable instructions, and although a logical order is shown in the flowchart, in some cases, the steps shown or described may be executed in an order different from that shown here.

[0033] Figure 1 is a flow chart of a method for processing vehicle perception information according to an embodiment of the present invention. Figure 1 As shown, the method comprises the following steps:

[0034] Step S102, obtaining perception information collected by the autonomous driving vehicle.

[0035] The above-mentioned autonomous driving vehicles may refer to vehicles that can automatically or semi-automatically drive, avoid obstacles, identify traffic signals, park, and other functions. Autonomous driving vehicles may be equipped with sensors, cameras, radars and other equipment, which can obtain surrounding environment information and be processed and decided by the on-board computer to achieve autonomous driving. The autonomous driving capability and vehicle type of autonomous driving vehicles can be determined according to actual needs and are not limited here.

[0036] The above-mentioned perception information may refer to the information about the surrounding environment obtained by the autonomous driving vehicle through various sensors, which may include but is not limited to various data such as road conditions, other vehicles, pedestrians, traffic lights, etc. The perception information can help the vehicle perceive the surrounding environment in real time, so as to make corresponding decisions and driving plans. The specific type of perception information can be determined according to actual needs and is not limited here; the data type of the perception information may include but is not limited to point cloud data, image data, etc. Point cloud data may be the coordinate information of a point in a three-dimensional space obtained by sensors such as lidar. By processing and analyzing the point cloud data, the vehicle can realize environmental modeling, obstacle detection, road recognition and other functions. Image data may be two-dimensional image information of the vehicle's surrounding environment obtained by sensors such as cameras. By processing and analyzing the image data, the vehicle can realize traffic sign recognition, vehicle detection, pedestrian detection and other functions. The specific data type of the perception information can also be determined according to actual needs and is not limited here.

[0037] In an optional embodiment, perception information can be remotely acquired in real time from various sensors, cameras, radars and other devices on the autonomous driving vehicle. Perception information can also be periodically acquired from a database on the autonomous driving vehicle for storing perception information. The perception information that can be acquired may include, but is not limited to, acquiring information on road conditions, other vehicles, pedestrians and the like through cameras; acquiring information such as the distance and speed of surrounding vehicles and obstacles through radar sensors; acquiring precise three-dimensional position and shape information of the surrounding environment through lidar sensors; acquiring the vehicle's position and navigation information through a global positioning system; acquiring distance information of surrounding objects through ultrasonic sensors, etc., to facilitate subsequent processing of the acquired perception information.

[0038] Step S104: dividing the preset space corresponding to the perception information according to the preset direction to obtain a plurality of first grids.

[0039] The plurality of first grids are non-uniformly distributed in the preset space.

[0040] The above-mentioned preset space may refer to the spatial range of the area corresponding to the perception information. The preset space may be determined according to the specific application scenarios and requirements, and may be set according to the vehicle's surrounding environment and road conditions. For example, in an autonomous driving vehicle, the preset space may refer to a road area within a certain range around the vehicle. The preset space may be determined according to the vehicle's sensor range and perception capabilities, and may include the front, rear, left and right sides of the vehicle. The preset space may be determined based on the specific perception information, and is not limited here.

[0041] The above-mentioned preset directions may include horizontal and vertical directions. The horizontal direction may be further divided into a direction parallel to the vehicle's travel direction, or perpendicular to the vehicle's travel direction, or a direction at a preset angle, etc. In practical applications, the corresponding preset directions may be determined according to business needs, such as recognition accuracy, etc., so that the preset space may be divided based on the horizontal or vertical direction, or the preset space may be divided based on both the horizontal and vertical directions. The preset direction may be determined specifically according to actual needs and is not limited here.

[0042] In an optional embodiment, a preset space corresponding to the vehicle perception information can be determined, and the size and shape of the preset space can be determined according to specific application scenarios and requirements; the preset space corresponding to the perception information can be divided according to a preset direction, and the preset space can be divided into multiple small areas by a grid division method, each of which can be a first grid, and the granularity of the division can be adjusted according to actual conditions, and can be determined according to the density and distribution of the perception information; the position of each first grid in the preset space can be determined according to the preset direction, and this can be achieved by calculating and converting coordinates to ensure that each first grid does not overlap in the preset space and covers the range included in the perception information. Multiple first grids are non-uniformly distributed in the preset space, and the first grids in different areas can be allocated and adjusted differently according to the density and importance of the perception information to ensure that the perception information is consistent with the preset direction. Effective processing and analysis of perceived information. Traffic conditions on the road are usually unevenly distributed. Some areas may have more vehicles or obstacles, while other areas have relatively fewer. The preset space corresponding to the perceived information is divided into multiple non-uniformly distributed grids to better adapt to the complex road environment, so that the vehicle system can perceive the surrounding environment more accurately. By dividing the preset space corresponding to the perceived information into multiple non-uniformly distributed first grids, more resources and energy can be invested in important areas, avoiding excessive processing of unimportant areas, thereby reducing the computational burden. Through the above steps, the preset space corresponding to the perceived information can be divided according to the preset direction to obtain multiple first grids, and ensure that the multiple first grids are non-uniformly distributed in the preset space. The vehicle perception information can be effectively processed and analyzed, providing more accurate and reliable data support for intelligent transportation systems and autonomous driving technologies.

[0043] Step S106: Based on the sub-perception information corresponding to the plurality of first grids, obstacle recognition is performed on the perception information to obtain a recognition result.

[0044] The recognition result is used to characterize at least one of the following: whether there is an obstacle in the preset space, the semantic category of the existing obstacle, and the location of the existing obstacle.

[0045] The above-mentioned sub-perception information may refer to information corresponding to each first grid in the perception information, and may refer to information located within the coordinates filtered out from the perception information based on the coordinates of the first grid.

[0046] The above-mentioned semantic category may refer to the obstacle category or type obtained by classifying the obstacles. Describing obstacles based on semantic categories can better understand the types and characteristics of objects in the surrounding environment. By semantically classifying obstacles, different types of objects can be handled more accurately, and corresponding processing and decisions can be made.

[0047] The position of the existing obstacle mentioned above may refer to the physical position of the identified obstacle in the real environment, or may refer to the relative position of the identified obstacle with respect to the vehicle, but is not limited thereto.

[0048] In an optional embodiment, the perception information in each first grid can be further processed to extract sub-perception information related to the obstacle, such as the position, speed, size, etc. of the target; the sub-perception information can be analyzed and processed, and a machine learning algorithm or a deep learning model can be used to identify obstacles. Relevant technical personnel can also use manual labeling to identify obstacles, which is not limited here. It can be determined whether the target is an obstacle and the obstacles can be classified, such as pedestrians, vehicles, buildings, etc.; according to the result of obstacle identification, it can be characterized as whether there is an obstacle in the preset space, the semantic category of the existing obstacle, and the location of the existing obstacle. In the above steps, by performing obstacle identification on the sub-perception information corresponding to multiple first grids, the situation of the surrounding environment can be more accurately judged, providing important information support for the automatic driving system. By identifying and characterizing obstacles, the vehicle's driving path can be better planned to ensure safety and efficiency.

[0049] In an embodiment of the present invention, a method for processing vehicle perception information is provided, including: acquiring perception information collected by an autonomous driving vehicle; dividing a preset space corresponding to the perception information according to a preset direction to obtain a plurality of first grids, wherein the plurality of first grids are non-uniformly distributed in the preset space; and performing obstacle recognition on the perception information based on sub-perception information corresponding to the plurality of first grids to obtain a recognition result, wherein the recognition result is used to characterize at least one of the following: whether there is an obstacle in the preset space, the semantic category of the existing obstacle, and the position of the existing obstacle. It is easy to notice that in actual road conditions, obstacles in different areas are unevenly distributed, and more processing resources are needed to process dense obstacle areas. Dividing the preset space corresponding to the perception information into multiple non-uniformly distributed first grids according to the preset direction can help improve the processing efficiency of the perception information and avoid invalid processing of the entire space. Multiple non-uniformly distributed grids can better adapt to complex road environments, enable the vehicle system to perceive the surrounding environment more accurately, and invest more resources and energy in important areas, avoiding excessive processing of unimportant areas, thereby reducing the computational burden, which can improve the accurate identification of obstacles, while reducing the processing volume of the entire space and saving computing resources, thereby solving the technical problem of high-precision processing of perception information in related technologies, resulting in large consumption of computing resources.

[0050] Optionally, the method also includes: determining the recognition requirements of the autonomous driving vehicle, wherein the recognition requirements are related to at least one of the following information: recognition accuracy, recognition efficiency; based on the recognition requirements, determining a preset direction from multiple directions, wherein the multiple directions include: a horizontal direction and a vertical direction.

[0051] The above-mentioned recognition accuracy can refer to the granularity or accuracy of the autonomous driving vehicle in identifying other objects, and can also be used to identify the size of obstacles that need to be identified. For example, a larger recognition accuracy can be set for the recognition processing of obstacles such as buildings, and a smaller recognition accuracy can be set for the recognition processing of obstacles such as small animals on the road. The specific setting of the recognition accuracy can be determined according to actual needs and is not limited here.

[0052] The above-mentioned recognition efficiency may refer to the speed and response time of an autonomous driving vehicle in identifying other objects. The higher the recognition efficiency, the faster and more accurately the vehicle can perceive and identify the surrounding environment, thereby making decisions and adjusting the driving direction faster.

[0053] In an optional embodiment, in the process of processing vehicle perception information, the recognition requirements of the autonomous driving vehicle may include determining at least one of the recognition accuracy and the recognition efficiency, wherein the recognition accuracy indicates the accuracy of the recognition, and the recognition efficiency indicates the speed and real-time nature of the recognition. When determining the recognition requirements of the autonomous driving vehicle, the actual application scenarios and requirements of the system may be analyzed to understand the specific requirements of the recognition task, and then the recognition accuracy and efficiency requirements may be determined based on these requirements, such as requiring accurate classification and recognition of the vehicle, or requiring rapid recognition of the vehicle in a real-time scenario; the preset direction may be determined from multiple directions, wherein the multiple directions may include horizontal directions and vertical directions, the horizontal direction may refer to the recognition of the horizontal position and motion trajectory of the vehicle, and the vertical direction may refer to the recognition of the vertical position and height of the vehicle. When determining the preset direction, the capabilities of the system's sensors and algorithms, as well as various situations that may occur in actual scenarios, may be considered, and the preset direction may be verified and optimized through experiments and tests to ensure that the system can meet the requirements of the recognition requirements in actual applications.

[0054] This application comprehensively considers the recognition accuracy, recognition efficiency and requirements of multiple directions when determining the recognition requirements of the autonomous driving vehicle and determining the preset direction, which can ensure that the system can accurately and quickly process the perception information of the autonomous driving vehicle, thereby realizing a safe and stable autonomous driving function.

[0055] Optionally, based on the recognition requirements, a preset direction is determined from multiple directions, including: when the recognition requirement is that the recognition accuracy reaches the target size accuracy, determining that the preset direction includes a vertical direction, and the target size accuracy is less than a first accuracy value; when the recognition requirement is that the recognition efficiency reaches the target recognition efficiency, determining that the preset direction includes a horizontal direction, and the target recognition efficiency is greater than the first efficiency value.

[0056] The first precision value mentioned above may refer to a preset precision value, and the first precision value may be determined according to actual needs and is not limited here. For example, in order to be able to identify obstacles of smaller sizes, the first precision value may be higher, and in order to be able to identify obstacles of larger sizes, the first precision value may be lower.

[0057] The first efficiency value mentioned above may refer to a preset efficiency value, and the first efficiency value may be determined according to actual needs, and is not limited here. For example, the setting of the first efficiency value may take into account the data processing capability of the data processing system on the vehicle, and may include information such as hardware parameters and ambient temperature, and may also take into account the driving scene of the vehicle, and the setting of the first efficiency value may be determined according to actual needs, and is not limited here.

[0058] In an optional embodiment, the recognition requirements of the vehicle perception system can be determined first, which may include recognition accuracy and recognition efficiency; the target size accuracy and target recognition efficiency can be determined based on the recognition requirements; the preset direction can be determined based on the requirements of the target size accuracy and target recognition efficiency; if the target size accuracy is the main consideration, the preset direction should include a vertical direction; if the target recognition efficiency is the main consideration, the preset direction can include a horizontal direction; and then the preset space corresponding to the perception information can be divided according to the preset direction to obtain multiple first grids, and the preset space can be divided into multiple small areas by a grid division method, each small area can be a first grid, and in the divided first grid, the recognition algorithm of the vehicle perception information is implemented, and the target is identified and classified according to the preset direction and recognition requirements. Through the implementation of the above steps, the vehicle perception information can be effectively processed, the accurate recognition and classification of the target can be achieved, and the performance and reliability of the vehicle perception system can be improved.

[0059] Optionally, based on the sub-perception information corresponding to the multiple first grids, obstacle recognition is performed on the perception information to obtain a recognition result, including: obtaining the current driving scene of the autonomous driving vehicle; inputting the sub-perception information corresponding to the multiple first grids into an obstacle recognition model, controlling the obstacle recognition model to perform obstacle recognition according to the current driving scene, and obtaining sub-recognition results of the multiple second grids, wherein the recognition result includes the sub-recognition results of the multiple second grids, and the scales of the multiple second grids match the current driving scene.

[0060] The above-mentioned current driving scenario may refer to a driving scenario corresponding to the environment and situation in which the autonomous driving vehicle is currently located, which may include factors such as road type, traffic conditions around the vehicle, pedestrians, obstacles, etc. For example, if an autonomous driving car is driving on a busy city road, there are many other vehicles, pedestrians and traffic lights around it. This current driving scenario requires the vehicle system to be able to perceive the surrounding vehicles and pedestrians in a timely manner, abide by traffic rules and make safe driving decisions. For example, the driving scenario of an autonomous driving vehicle driving on a highway may mainly be to maintain a safe distance from other vehicles, abide by lane rules, etc., perceive the surrounding vehicles in a timely manner, and make corresponding driving decisions to ensure driving safety. The specific content of the current driving scenario can be determined according to actual needs and is not limited here.

[0061] The above-mentioned obstacle recognition model may refer to a model that senses, detects and recognizes the surrounding environment through the vehicle-mounted perception system, thereby determining whether there are obstacles in a preset space, the semantic category of the obstacles, and the location information of the obstacles. The obstacle recognition model may use a deep learning algorithm, such as a convolutional neural network or a recurrent neural network, to process images or radar data to achieve accurate recognition and positioning of obstacles. The specific type of the obstacle recognition model may be determined according to actual needs and is not limited here.

[0062] The above-mentioned second grid may be the same as the first grid or different from the first grid, and may be determined according to actual needs; when the second grid is the same as the first grid, the perception information collected by the autonomous driving vehicle may be obtained, and then the preset space corresponding to the perception information may be divided according to the preset direction to obtain a plurality of first grids, and the first grids may be non-uniformly distributed in the preset space, that is, the space at different positions is divided into grids of different numbers and sizes; based on the sub-perception information corresponding to these first grids, obstacle recognition may be performed on the perception information to obtain recognition results, including whether there are obstacles in the preset space, the semantic category of the existing obstacles, and the position information; and the current movement of the autonomous driving vehicle may be obtained. The method comprises the following steps: a first driving scene, and then inputting sub-perception information corresponding to the plurality of first grids into the obstacle recognition model; according to the current driving scene, controlling the obstacle recognition model to perform obstacle recognition, and obtaining sub-recognition results of the plurality of second grids, including the sub-recognition results of the plurality of second grids, and the scale of the second grid matches the current driving scene, that is, the size and position of the grid can better adapt to the current driving environment, so as to improve the perception ability of the autonomous driving vehicle to the surrounding environment and improve the driving safety and reliability by rasterizing the perception information and identifying obstacles; and by reasonably dividing the space and grids, and adjusting the obstacle recognition model according to the current driving scene, obstacles can be more accurately identified and corresponding driving decisions can be made.

[0063] In an optional embodiment, the current driving scene of the autonomous driving vehicle can be obtained, and the environmental information around the vehicle, including road conditions, surrounding vehicles and obstacles, etc., can be obtained through on-board sensors such as lidar and cameras, so as to construct data of the current driving scene. The entire driving scene can be divided into multiple first grids, each grid containing environmental information within a certain range, and the area around the vehicle that needs to be sensed is determined according to the current position and direction of the vehicle; multiple sub-perception information can be extracted in each first grid, including obstacles, traffic signs, road lines and other information around the vehicle, which can be obtained through sensors or extracted through the vehicle's own data processing system; the sub-perception information corresponding to each first grid can be input into an obstacle recognition model for processing, and the obstacle recognition model can be a neural network model based on deep learning or other machine learning algorithms. The obstacle recognition model The information in the grid can be feature extracted and identified to identify the location, type and other information of the obstacle; the input and parameters of the obstacle recognition model can be controlled according to the characteristics of the current driving scene to ensure that the model can accurately identify obstacles. For example, when driving on a highway, the obstacle recognition model may require higher resolution and sensitivity to detect high-speed vehicles and obstacles. The obstacle recognition model can output sub-recognition results of multiple second grids, that is, the obstacle information detected in each second grid, which can include the location, speed, size, etc. of the obstacle, which can help the vehicle make decisions and plans. Finally, the sub-recognition results of the second grid can be matched with the current driving scene to ensure that the recognition results are consistent with the actual road conditions. If there is a mismatch, re-recognition or adjustment of model parameters can be performed, so that the surrounding environment can be accurately perceived, providing reliable information support for the autonomous driving system.

[0064] Optionally, the sub-perception information corresponding to the plurality of first grids is input into an obstacle recognition model, and the obstacle recognition model is controlled to perform obstacle recognition according to the current driving scene to obtain sub-recognition results of the plurality of second grids, including: determining a target output scale from a plurality of output scales based on the current driving scene, wherein different output scales correspond to different driving scenes; and the sub-perception information corresponding to the plurality of first grids is input into the obstacle recognition model, and the obstacle recognition model is controlled to perform obstacle recognition according to the target output scale to obtain sub-recognition results of the plurality of second grids.

[0065] In an optional embodiment, in the vehicle perception information processing, different output scales can be determined according to different current driving scenes, and the sub-perception information of multiple first grids can be input into the obstacle recognition model, and the model can be controlled to perform obstacle recognition to obtain sub-recognition results of multiple second grids. Specifically, the target output scale can be determined according to the current driving scene of the vehicle. Different driving scenes may require different output scales. For example, a longer prediction distance may be required on a highway, while a shorter prediction distance may be required on an urban road. The sub-perception information of multiple first grids can be input into the obstacle recognition model, and the obstacle recognition model can be controlled to perform obstacle recognition according to the target output scale. According to the determined target output scale, the parameters and structure of the model are adjusted to meet the needs of the current driving scene. After the obstacle recognition model performs obstacle recognition, the sub-recognition results of multiple second grids can be obtained, that is, the obstacle position and attribute information within a larger range. Through the above steps, the vehicle perception information can be effectively processed according to the needs of the current driving scene, and the accuracy and efficiency of obstacle recognition can be improved.

[0066] Optionally, the sub-perception information corresponding to the plurality of first grids is input into an obstacle recognition model, and the obstacle recognition model is controlled to perform obstacle recognition according to the current driving scene to obtain sub-recognition results of the plurality of second grids, including: the sub-perception information corresponding to the plurality of first grids is input into the obstacle recognition model, and the obstacle recognition model is controlled to perform obstacle recognition according to the plurality of output scales to obtain sub-recognition result sets corresponding to the plurality of output scales, wherein different sub-recognition result sets contain sub-recognition results of the second grids at different output scales; based on the current driving scene, a sub-recognition result set corresponding to a target output scale is determined from the sub-recognition result set to obtain sub-recognition results of the plurality of second grids, wherein the target output scale corresponds to the current driving scene.

[0067] In an optional embodiment, sub-perception information corresponding to multiple first grids may be input into the obstacle recognition model, and the perception information in each first grid may be input into the obstacle recognition model, where the perception information may include information such as the position, speed, and size of obstacles around the vehicle; the obstacle recognition model may be controlled to perform obstacle recognition according to multiple output scales, and in the obstacle recognition model, multiple output scales may be set so as to recognize obstacles at different scales. By adjusting the output scales, obstacles of different sizes may be better adapted, and sub-recognition result sets corresponding to multiple output scales may be obtained. According to different output scales, the obstacle recognition model may output multiple A sub-recognition result set, each set may correspond to a sub-recognition result of the second grid under a different output scale, and the recognition result may include information such as the position, speed, and size of each obstacle; the sub-recognition result set corresponding to the target output scale may be determined based on the current driving scene, and the target output scale may be determined according to the characteristics of the current driving scene, that is, the output scale suitable for the current scene, and then the sub-recognition result set corresponding to the target output scale is selected from multiple sub-recognition result sets to obtain multiple sub-recognition results of the second grid, and finally multiple sub-recognition results of the second grid may be obtained, which can help the driving system to more accurately identify surrounding obstacles and provide support and guarantee for safe driving of the vehicle.

[0068] Optionally, the loss function of the obstacle recognition model is obtained by weighted processing of scale loss functions corresponding to multiple output scales, and the scale loss function is obtained by weighted processing of the grid loss function of the second grid corresponding to the corresponding output scale. The grid loss function of the second grid is determined based on the real recognition result of the second grid and the training recognition result of the second grid. The training recognition result is obtained by controlling the obstacle recognition model to perform obstacle recognition on the training data according to the corresponding output scale.

[0069] In an optional embodiment, during the construction and optimization of the obstacle recognition model in the vehicle perception information processing, the loss function of the obstacle recognition model may be obtained by weighting the scale loss functions corresponding to multiple output scales. During the model training process, multiple output results at different scales may be considered, and the loss function may be weighted according to the importance of the output results at different scales to ensure that the model can accurately recognize obstacles at different scales. The scale loss function may be obtained by weighting the grid loss function of the second grid corresponding to the corresponding output scale. At each output scale, the grid loss function of the second grid may be weighted to ensure that the model can accurately capture the information of the obstacle at each scale. The grid loss function of the second grid may be based on the first grid. The real recognition result of the second grid and the training recognition result of the second grid are determined. The grid loss function can compare the difference between the model output result and the real result, and calculate the loss value based on this difference to guide the training process of the model. The training recognition result can be obtained by controlling the obstacle recognition model to perform obstacle recognition on the training data according to the corresponding output scale. During the training process, the model can control the obstacle recognition of the training data according to the corresponding output scale to ensure that the model can learn the obstacle features at different scales during the training process and improve the generalization ability of the model. The above settings can effectively guide the training process of the obstacle recognition model and improve the accuracy and robustness of the model at different scales through weighted processing of loss functions of multiple scales and determination of the grid loss function, as well as control of the training data.

[0070] Optionally, the weight value corresponding to the scale loss function is determined based on the number of grids corresponding to the minimum output scale in the second grid corresponding to the corresponding output scale; and / or, the weight value corresponding to the grid loss function is determined based on the distance between the corresponding second grid and a preset position in a preset space.

[0071] In an optional embodiment, the weight value corresponding to the scale loss function can be determined based on the number of grids corresponding to the minimum output scale in the second grid corresponding to the corresponding output scale. In vehicle perception information processing, the number of grids corresponding to different output scales can be considered to determine the weight value of the loss function. Specifically, if the number of grids corresponding to the minimum output scale in the second grid corresponding to a certain output scale is large, then the weight value of the loss function of the scale can be increased accordingly to ensure that more attention is paid to the information loss of these scales during the training process; the weight value corresponding to the grid loss function can be determined based on the distance between the corresponding second grid and the preset position in the preset space. The weight value of the loss function can be determined by considering the distance between the preset position and the second grid. Specifically, if the distance between a certain second grid and the preset position is small, the weight value of the loss function of the grid can be increased accordingly to ensure that more attention is paid to the loss of these position information during the training process. Through the above settings, the importance of different scales and position information in vehicle perception information processing can be better balanced, thereby improving the performance and accuracy of the model, and can also better adapt to vehicle perception needs in different scenarios and environments, making the model more flexible and applicable.

[0072] The technical solution proposed in this application is explained below in combination with an optional application scenario. In the field of unmanned driving, the grid occupancy network can output whether each grid (voxel) within the perception range is occupied by the environment and its semantic category, which can include vehicles, pedestrians, obstacles, ground, walls, etc., for autonomous driving vehicles to perceive the environment. It can output more detailed and higher-resolution dynamic and static objects. The space and contours occupied can be output, and at the same time, a more refined expression can be formed for common obstacles or irregular objects that are not seen or rare in the annotated dataset. Especially for the field of unmanned driving in mines, the identification of common obstacles such as flexible participants and animals is more critical; the traditional grid occupancy network divides the entire perception environment into grids evenly according to the resolution, but the near side requires finer resolution than the far side. Grid division: The perception effect of distant grids is small, and the field of mine unmanned driving requires more accurate and precise detection of general obstacles in the near-ground area. The use of uniform grid division will lead to more resources and long calculation time, which is not conducive to deployment on the vehicle. At the same time, the requirements for grid resolution in different scenarios of unmanned driving may be inconsistent, and grids with uniform resolution are not conducive to the generalization and implementation of the business; that is, in related technologies, mine unmanned driving requires high resolution and high precision for general obstacle detection; the grid resolution of traditional grid occupancy networks is the same, resulting in more resources and long calculation time, which is not conducive to deployment on the vehicle; the traditional grid occupancy network business application is consistent with training, but the requirements for grid resolution in different scenarios are inconsistent, and uniform resolution is not conducive to business generalization and implementation.

[0073] In order to enable a high-resolution grid occupancy network to be put on board, the present application proposes uneven grid division and variable grid output, that is, an uneven zoom grid occupancy network. In order to enable a high-resolution grid occupancy network to be put on board, a method of uneven grid division and zoom grid output can be adopted; wherein the uneven grid division can only be an uneven division in a preset direction, and the selection of different directions is related to business needs. The horizontal direction can determine the scale of entering the model, affecting the operating efficiency of the model, and is a compromise between accuracy and efficiency; in order to solve the technical problems that uniform grids occupy a lot of resources, have a long calculation time, and have low revenue from distant refined grid services, a horizontally uneven grid division network can be adopted; in order to solve the problem of accurate detection of small obstacles in the near-ground area, a vertically uneven grid division network can be adopted; in order to solve the need for resolution changes in different scenes, a zoom grid output can be adopted.

[0074] Specifically, an unlimited number of lidars and cameras can be installed on unmanned mining vehicles to obtain perception data in real time, which may include point cloud and video data.

[0075] Figure 2 is a schematic diagram of an optional vehicle perception information processing process according to an embodiment of the present invention, such as Figure 2 As shown, the processing process of vehicle perception information can include acquiring perception data, performing uneven grid division, voxelizing the model, setting the backbone network of the model, setting the occupancy head of the model and the grid output process.

[0076] The above-mentioned acquisition of perception data, model voxelization, setting of the model's backbone network, and grid output process of this application can use the traditional occupancy head (Occupancy) network architecture. This application mainly optimizes and improves the two processes of uneven grid division and setting of the model's occupancy head.

[0077] Specifically, uneven grid division can be performed. The traditional occupancy network (Occupancy) is evenly divided in the three directions of x, y, and z. Voxelization operation (Voxelize) is performed according to the uniform grid to generate the input of the corresponding backbone network (Backbone). For example, the perception range is X: (-50~50m), Y: (-50~50m), Z: (-6~6m), the scale of the grid is (X: 0.1m, Y: 0.1m, Z: 0.2m), and the generated feature output dimension is (1000*1000*60). The uneven grid division of this application can be as follows, with varying resolutions in both the horizontal and vertical directions, that is, the scale of the grid. For example, a grid with a resolution of 0.5m can be generated near the vehicle, and a resolution of 0.4m can be generated at the far end. The final combined input dimension is smaller, for example (500*500*30), which can ensure the detection capability of the grid occupancy network in the vicinity, and reduce resource usage and computing latency.

[0078] Figure 3 is a schematic diagram of an optional horizontal division of a preset space according to an embodiment of the present invention, such as Figure 3 As shown, an optional schematic effect of dividing the preset space based on the horizontal direction is shown. It can be seen that the grid network is unevenly distributed, and the grid network close to the central preset position is denser than the grid network close to the central preset position.

[0079] Figure 4 is a schematic diagram of an optional vertical division of a preset space according to an embodiment of the present invention, such as Figure 4 As shown, an optional schematic effect of dividing the preset space based on the vertical direction is shown. It can be seen that the grid network is unevenly distributed, and the grid network close to the central preset position is denser than the grid network close to the central preset position.

[0080] Improvements to the model (OccHead). During the training phase, the predicted values ​​and true values ​​output by the backbone network (Backbone) and conversational inference (Conv) are matched to each other at N different scales. Taking three groups as an example, three different scales can be matched. In the first step, for grids at different positions, different loss weights can be used to form the overall loss corresponding to the scale, which can include loss 1, loss 2, and loss 3. The weight form can be the Euclidean distance of the farthest point from the center point of the four corners of the grid, or the reciprocal of other distance calculation forms. The weight value of the position marked here can be loss 1> loss 2> loss 3, thereby achieving a grid occupancy output that pays more attention to the area near the vehicle and has higher accuracy. In the second step, the losses are calculated for the combinations of N groups of different corresponding predicted values ​​and true values, which can include loss 1, loss 2, and loss 3. Then, according to the number of grids containing the minimum grid, the reciprocal is taken as the weight to obtain the final weighted loss as the loss of the model (OccHead) for model training, thereby achieving multiple groups of grid outputs of different scales, that is, zoom grids.

[0081] Figure 5 is a schematic diagram of a method for determining a loss function of an optional obstacle recognition model according to an embodiment of the present invention, such as Figure 5 As shown, the occupancy head of the model outputs the predicted value of the uneven grid and the true value of the uneven grid respectively. The predicted value of the uneven grid corresponds to three scales, including scale 1, scale 2 and scale 3; the true value of the uneven grid corresponds to three scales, including scale 1, scale 2 and scale 3; thus, loss 1 can be constructed based on two scales 1, loss 2 can be constructed based on two scales 2, loss 3 can be constructed based on two scales 3, and finally, weighted loss can be constructed based on loss 1, loss 2 and loss 3.

[0082] Figure 6 is a schematic diagram corresponding to an optional model training process scale according to an embodiment of the present invention, such as Figure 6 As shown, different scales can correspond to different frames during model training, and three different scales can correspond to different frames 1, 2 and 3. Frames of different specifications can be used to flexibly divide the preset space.

[0083] Figure 7 is a schematic diagram of an optional scale adjustment of the model application stage according to an embodiment of the present invention, such as Figure 7 As shown, the occupancy head of the model can be adjusted according to the business to obtain scale 1 output, scale 2 output and scale 3 output.

[0084] In the forward application stage of the model, the model can be adjusted according to different business needs, such as forward driving, reversing, obstacle avoidance, etc., or different scale switching environments such as the forward road, curve, crushing station, etc. to output and zoom the grid occupancy network of different scales.

[0085] According to another aspect of an embodiment of the present invention, a device for processing vehicle perception information is also provided. The device can execute the method for processing vehicle perception information of the above embodiment. The specific implementation method and preferred application scenario are the same as those of the above embodiment and will not be repeated here.

[0086] Figure 8 is a schematic diagram of a vehicle perception information processing device according to an embodiment of the present application, such as Figure 8 As shown, the device includes the following: an acquisition module 802 , a division module 804 and an identification module 806 .

[0087] Among them, the acquisition module is used to obtain the perception information collected by the autonomous driving vehicle; the division module is used to divide the preset space corresponding to the perception information according to the preset direction to obtain multiple first grids, wherein the multiple first grids are non-uniformly distributed in the preset space; the identification module is used to identify obstacles on the perception information based on the sub-perception information corresponding to the multiple first grids to obtain an identification result, wherein the identification result is used to characterize at least one of the following: whether there are obstacles in the preset space, the semantic category of the existing obstacles, and the location of the existing obstacles.

[0088] Optionally, the method also includes: determining the recognition requirements of the autonomous driving vehicle, wherein the recognition requirements are related to at least one of the following information: recognition accuracy, recognition efficiency; based on the recognition requirements, determining a preset direction from multiple directions, wherein the multiple directions include: a horizontal direction and a vertical direction.

[0089] Optionally, based on the recognition requirements, a preset direction is determined from multiple directions, including: when the recognition requirement is that the recognition accuracy reaches the target size accuracy, determining that the preset direction includes a vertical direction, and the target size accuracy is less than a first accuracy value; when the recognition requirement is that the recognition efficiency reaches the target recognition efficiency, determining that the preset direction includes a horizontal direction, and the target recognition efficiency is greater than the first efficiency value.

[0090] Optionally, based on the sub-perception information corresponding to the multiple first grids, obstacle recognition is performed on the perception information to obtain a recognition result, including: obtaining the current driving scene of the autonomous driving vehicle; inputting the sub-perception information corresponding to the multiple first grids into an obstacle recognition model, controlling the obstacle recognition model to perform obstacle recognition according to the current driving scene, and obtaining sub-recognition results of the multiple second grids, wherein the recognition result includes the sub-recognition results of the multiple second grids, and the scales of the multiple second grids match the current driving scene.

[0091] Optionally, the sub-perception information corresponding to the plurality of first grids is input into an obstacle recognition model, and the obstacle recognition model is controlled to perform obstacle recognition according to the current driving scene to obtain sub-recognition results of the plurality of second grids, including: determining a target output scale from a plurality of output scales based on the current driving scene, wherein different output scales correspond to different driving scenes; and the sub-perception information corresponding to the plurality of first grids is input into the obstacle recognition model, and the obstacle recognition model is controlled to perform obstacle recognition according to the target output scale to obtain sub-recognition results of the plurality of second grids.

[0092] Optionally, the sub-perception information corresponding to the plurality of first grids is input into an obstacle recognition model, and the obstacle recognition model is controlled to perform obstacle recognition according to the current driving scene to obtain sub-recognition results of the plurality of second grids, including: the sub-perception information corresponding to the plurality of first grids is input into the obstacle recognition model, and the obstacle recognition model is controlled to perform obstacle recognition according to the plurality of output scales to obtain sub-recognition result sets corresponding to the plurality of output scales, wherein different sub-recognition result sets contain sub-recognition results of the second grids at different output scales; based on the current driving scene, a sub-recognition result set corresponding to a target output scale is determined from the sub-recognition result set to obtain sub-recognition results of the plurality of second grids, wherein the target output scale corresponds to the current driving scene.

[0093] Optionally, the loss function of the obstacle recognition model is obtained by weighted processing of scale loss functions corresponding to multiple output scales, and the scale loss function is obtained by weighted processing of the grid loss function of the second grid corresponding to the corresponding output scale. The grid loss function of the second grid is determined based on the real recognition result of the second grid and the training recognition result of the second grid. The training recognition result is obtained by controlling the obstacle recognition model to perform obstacle recognition on the training data according to the corresponding output scale.

[0094] Optionally, the weight value corresponding to the scale loss function is determined based on the number of grids corresponding to the minimum output scale in the second grid corresponding to the corresponding output scale; and / or, the weight value corresponding to the grid loss function is determined based on the distance between the corresponding second grid and a preset position in a preset space.

[0095] An embodiment of the present application further provides an electronic device, comprising: a memory storing an executable program; and a processor for running the program, wherein the program executes the methods in various embodiments of the present invention when running.

[0096] The above-mentioned memory may refer to a device inside a computer for storing data and programs, and may include memory, hard disk, etc., wherein the memory may be used to temporarily store running programs and data, the hard disk may be used to store programs and data for a long time, and the memory may be used to enable the computer to read and write data, and execute programs; the above-mentioned processor may be responsible for executing instructions in computer programs and performing data processing, and may be responsible for controlling and executing various operations, including arithmetic operations, logical operations, data transmission, etc.

[0097] An embodiment of the present application further provides a computer-readable storage medium, which includes a stored executable program, wherein when the executable program is running, the device where the computer-readable storage medium is located is controlled to execute the methods in various embodiments of the present invention.

[0098] The above-mentioned computer storage medium may refer to a medium in a computer memory used to store certain discontinuous physical quantities. Computer storage media mainly include semiconductors, magnetic cores, magnetic drums, magnetic tapes, laser disks, etc.; the stored program included in the computer-readable storage medium may be a set of instructions that can be recognized and executed by a computer, running on an electronic computer, and is an information tool that meets certain needs of people.

[0099] An embodiment of the present application further provides a computer program product, including a computer program, which implements the methods in various embodiments of the present invention when executed by a processor.

[0100] The above-mentioned computer program product may refer to a software program that has been written, tested and released and can be run on a computer or other device. The computer program product may include an application, an operating system, tool software, etc., which is used to implement specific functions or solve specific problems.

[0101] An embodiment of the present application further provides a computer program product, including a non-volatile computer-readable storage medium, wherein the non-volatile computer-readable storage medium is used to store a computer program, and when the computer program is executed by a processor, the method in each embodiment of the present invention is implemented.

[0102] The above-mentioned non-volatile computer-readable storage medium may refer to a medium for storing data. The non-volatile computer-readable storage medium can keep the data from being lost when the power is off, and can be used to store long-term data, such as operating systems, applications, and user files. The non-volatile storage medium may include hard disk drives, solid-state drives, optical disks, and flash memory storage devices, etc.

[0103] The embodiments of the present application further provide a computer program, which implements the methods in the above-mentioned embodiments of the present invention when executed by a processor.

[0104] The above-mentioned computer program may refer to a collection of instructions used to tell a computer to perform a specific task or operation. A computer program may be written by a programmer using a specific programming language and may include algorithms, data structures, logic, and control flows. A computer program may be used for a variety of purposes, including application software, operating systems, and the like.

[0105] In the above embodiments of the present invention, the description of each embodiment has its own emphasis. For parts that are not described in detail in a certain embodiment, reference can be made to the relevant descriptions of other embodiments.

[0106] In the several embodiments provided in this application, it should be understood that the disclosed technical content can be implemented in other ways. Among them, the device embodiments described above are only schematic. For example, the division of the units can be a logical function division. There may be other division methods in actual implementation. For example, multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the mutual coupling or direct coupling or communication connection shown or discussed can be through some interfaces, indirect coupling or communication connection of units or modules, which can be electrical or other forms.

[0107] The units described as separate components may or may not be physically separated, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed on multiple units. Some or all of the units may be selected according to actual needs to achieve the purpose of the present embodiment.

[0108] In addition, each functional unit in each embodiment of the present invention may be integrated into one processing unit, or each unit may exist physically separately, or two or more units may be integrated into one unit. The above-mentioned integrated unit may be implemented in the form of hardware or in the form of software functional units.

[0109] If the integrated unit is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, in essence, or the part that contributes to the prior art, or all or part of the technical solution can be embodied in the form of a software product. The computer software product is stored in a storage medium, including several instructions for a computer device (which can be a personal computer, server or network device, etc.) to perform all or part of the steps of the method described in each embodiment of the present invention. The aforementioned storage medium includes: U disk, read-only memory (ROM, Read-Only Memory), random access memory (RAM, Random Access Memory), mobile hard disk, disk or optical disk, etc. Various media that can store program codes.

[0110] The above is only a preferred embodiment of the present invention. It should be pointed out that for ordinary technicians in this technical field, several improvements and modifications can be made without departing from the principle of the present invention. These improvements and modifications should also be regarded as the scope of protection of the present invention.

Claims

1. A method for processing vehicle perception information, characterized in that: include: Obtaining perception information collected by the autonomous vehicle; Dividing a preset space corresponding to the perception information according to a preset direction to obtain a plurality of first grids, wherein the plurality of first grids are non-uniformly distributed in the preset space; Based on the sub-perception information corresponding to the plurality of first grids, obstacle recognition is performed on the perception information to obtain a recognition result, wherein the recognition result is used to characterize at least one of the following: whether there is an obstacle in the preset space, a semantic category of the existing obstacle, and a position of the existing obstacle; Wherein, the performing obstacle recognition on the perception information based on the sub-perception information corresponding to the plurality of first grids to obtain a recognition result includes: inputting the sub-perception information corresponding to the plurality of first grids into an obstacle recognition model, controlling the obstacle recognition model to perform obstacle recognition, and obtaining sub-recognition results of the plurality of second grids, wherein the recognition result includes the sub-recognition results of the plurality of second grids; The method further includes: determining an identification requirement of the autonomous driving vehicle, wherein the identification requirement is related to at least one of the following information: identification accuracy and identification efficiency; based on the identification requirement, determining the preset direction from multiple directions, wherein the multiple directions include: a horizontal direction and a vertical direction, wherein the vertical direction is used to characterize the identification of the vertical position and height of the autonomous driving vehicle; Among them, based on the recognition requirement, determining the preset direction from multiple directions includes: when the recognition requirement is that the recognition accuracy reaches the target size accuracy, determining that the preset direction includes a vertical direction, and the target size accuracy is less than a first accuracy value; when the recognition requirement is that the recognition efficiency reaches the target recognition efficiency, determining that the preset direction includes a horizontal direction, and the target recognition efficiency is greater than a first efficiency value.

2. The method according to claim 1, characterized in that The step of inputting the sub-perception information corresponding to the plurality of first grids into an obstacle recognition model, controlling the obstacle recognition model to perform obstacle recognition, and obtaining the sub-recognition results of the plurality of second grids includes: Acquire a current driving scene of the autonomous driving vehicle; The sub-perception information corresponding to the multiple first grids is input into the obstacle recognition model, and the obstacle recognition model is controlled to perform obstacle recognition according to the current driving scene to obtain sub-recognition results of the multiple second grids, wherein the scales of the multiple second grids match the current driving scene.

3. The method according to claim 2, characterized in that The step of inputting the sub-perception information corresponding to the plurality of first grids into an obstacle recognition model, controlling the obstacle recognition model to perform obstacle recognition according to the current driving scene, and obtaining the sub-recognition results of the plurality of second grids includes: Based on the current driving scene, determining a target output scale from a plurality of output scales, wherein different output scales correspond to different driving scenes; The sub-perception information corresponding to the plurality of first grids is input into the obstacle recognition model, and the obstacle recognition model is controlled to perform obstacle recognition according to the target output scale to obtain sub-recognition results of the plurality of second grids.

4. The method according to claim 2, characterized in that: The step of inputting the sub-perception information corresponding to the plurality of first grids into an obstacle recognition model, controlling the obstacle recognition model to perform obstacle recognition according to the current driving scene, and obtaining the sub-recognition results of the plurality of second grids includes: Inputting the sub-perception information corresponding to the plurality of first grids into the obstacle recognition model, controlling the obstacle recognition model to perform obstacle recognition according to the plurality of output scales, and obtaining sub-recognition result sets corresponding to the plurality of output scales, wherein different sub-recognition result sets include sub-recognition results of the second grid at different output scales; Based on the current driving scene, a sub-recognition result set corresponding to a target output scale is determined from the sub-recognition result set to obtain sub-recognition results of the plurality of second grids, wherein the target output scale corresponds to the current driving scene.

5. The method according to claim 3 or 4, characterized in that: The loss function of the obstacle recognition model is obtained by weighting the scale loss functions corresponding to the multiple output scales, and the scale loss function is obtained by weighting the grid loss function of the second grid corresponding to the corresponding output scale. The grid loss function of the second grid is determined based on the real recognition result of the second grid and the training recognition result of the second grid. The training recognition result is obtained by controlling the obstacle recognition model to perform obstacle recognition on training data according to the corresponding output scale.

6. The method according to claim 5, characterized in that The weight value corresponding to the scale loss function is determined based on the number of grids corresponding to the minimum output scale in the second grid corresponding to the corresponding output scale; and / or, the weight value corresponding to the grid loss function is determined based on the distance between the corresponding second grid and the preset position in the preset space.

7. A vehicle perception information processing device, characterized in that: include: An acquisition module, used to acquire the perception information collected by the autonomous driving vehicle; A division module, configured to divide the preset space corresponding to the perception information according to a preset direction to obtain a plurality of first grids, wherein the plurality of first grids are non-uniformly distributed in the preset space; an identification module, configured to perform obstacle identification on the perception information based on the sub-perception information corresponding to the plurality of first grids, and obtain an identification result, wherein the identification result is used to characterize at least one of the following: whether there is an obstacle in the preset space, a semantic category of the existing obstacle, and a position of the existing obstacle; The recognition module is further used to input the sub-perception information corresponding to the plurality of first grids into the obstacle recognition model, control the obstacle recognition model to perform obstacle recognition, and obtain the sub-recognition results of the plurality of second grids, wherein the recognition result includes the sub-recognition results of the plurality of second grids; The division module is further used to determine the recognition requirements of the autonomous driving vehicle, wherein the recognition requirements are related to at least one of the following information: recognition accuracy and recognition efficiency; based on the recognition requirements, the preset direction is determined from multiple directions, wherein the multiple directions include: a horizontal direction and a vertical direction, and the vertical direction is used to characterize the recognition of the vertical position and height of the autonomous driving vehicle; The division module is also used to determine that the preset direction includes a vertical direction and the target size accuracy is less than a first accuracy value when the recognition requirement is that the recognition accuracy reaches the target size accuracy; and to determine that the preset direction includes a horizontal direction and the target recognition efficiency is greater than a first efficiency value when the recognition requirement is that the recognition efficiency reaches the target recognition efficiency.

8. An electronic device, characterized in that: include: A memory storing an executable program; A processor, configured to run the program, wherein the program executes the method according to any one of claims 1 to 6 when running.

Images