Target recognition model detection method and device, equipment, medium and program product

By generating a three-dimensional simulation environment for multiple working conditions and weather scenes in the autonomous driving simulator, the projection gradient descent method is used to iteratively attack the image data, test and train the target recognition model, and the accuracy of recognition under image interference in the environment of autonomous driving vehicles is solved. The problem of sexuality improves the robustness and recognition ability of the model.

CN118968470AActive Publication Date: 2024-11-15CHERY AUTOMOBILE CO LTD

Patent Information

Application Number
CN202411066090.4
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2024-08-05
Publication Date
2024-11-15
Estimated Expiration
2044-08-05

AI Technical Summary

Technical Problem

The target recognition model of autonomous driving vehicles is easily affected by external interference during the environmental image acquisition process, resulting in the inability to guarantee the accuracy of the recognition. A method is needed to test and improve the robustness of the model.

Method used

By customizing a variety of working conditions and weather scenarios, using the autonomous driving simulator to generate a three-dimensional simulation environment, collect sample image data, and use the projection gradient descent method to iteratively attack the image data to generate image data against perturbations for testing and training Target recognition model.

Benefits of technology

The recognition accuracy and robustness of the target recognition model are improved, so that it can still correctly identify obstacles when facing interference, ensuring the safe driving of autonomous vehicles.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN118968470A_ABST
    Figure CN118968470A_ABST
Patent Text Reader

Abstract

The invention discloses a target recognition model detection method and device, equipment, a medium and a program product, and relates to the field of deep learning, and the method comprises the following steps: obtaining a weather scene configuration file and a working condition configuration file; generating a three-dimensional simulation driving environment based on the weather scene configuration file and the working condition configuration file, wherein the three-dimensional simulation driving environment comprises a three-dimensional model of an automatic driving vehicle and a three-dimensional model of an obstacle; collecting sample image data based on the three-dimensional simulation driving environment; performing iterative attack on the sample image data based on a projection gradient descent method to generate first image data; and identifying the first image data through a target identification model corresponding to the autonomous vehicle to obtain a first identification result, the first identification result being used for testing the target identification model. The robustness of the target recognition model can be detected.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The embodiments of the present application relate to the field of deep learning, and in particular to a detection method, device, equipment, medium and program product for a target recognition model. Background Art

[0002] Environmental perception technology is a key technology for the safe driving of autonomous vehicles, among which the vehicle's ability to accurately perceive depends on the target recognition model.

[0003] In related technologies, the target recognition model can collect environmental images, identify the environmental images, determine the target objects in the images (such as pedestrians, vehicles, traffic signs, etc. around the autonomous driving vehicle), and provide the autonomous driving vehicle with the necessary information to make correct driving decisions.

[0004] However, the implementation of relevant functions of autonomous driving vehicles depends on the network. If the environmental image is attacked by the outside world when it is collected, the content of the environmental image will be disturbed and there is a risk of modification. The accuracy of the target recognition model in recognizing the interfered environmental image cannot be guaranteed. Therefore, it is an urgent problem to accurately test the robustness of the target recognition model before the autonomous driving vehicle is put into use. Summary of the invention

[0005] The embodiments of the present application provide a method, apparatus, device, medium and program product for detecting a target recognition model, which can detect the robustness of the target recognition model. The technical solution is as follows:

[0006] On the one hand, a detection method of a target recognition model is provided, the method comprising:

[0007] Obtaining a weather scenario configuration file and a working condition configuration file, wherein the weather scenario configuration file is used to generate a simulated weather scenario when the autonomous driving vehicle is driving, and the working condition configuration file is used to generate a working condition scenario when the autonomous driving vehicle is driving, and the working condition configuration file includes an obstacle configuration part and a driving condition configuration part, wherein the obstacle configuration part includes the configuration of obstacles encountered by the autonomous driving vehicle when driving, and the driving condition configuration part includes the working condition of the autonomous driving vehicle when driving, and the working condition includes at least one of vehicle emergency braking, vehicle emergency avoidance, and vehicle traffic congestion;

[0008] Generate a three-dimensional simulated driving environment based on the weather scenario configuration file and the operating condition configuration file, wherein the three-dimensional simulated driving environment includes three-dimensional models of the autonomous driving vehicle and the obstacles;

[0009] Collecting sample image data based on the three-dimensional simulated driving environment, wherein the sample image data is an environment image collected from the perspective of the autonomous driving vehicle, and the environment image contains the obstacle;

[0010] Iteratively attack the sample image data based on a projected gradient descent method to generate first image data, where the first image data is image data obtained by superimposing noise on the sample image data;

[0011] The first image data is recognized by a target recognition model corresponding to the autonomous driving vehicle to obtain a first recognition result. The first recognition result includes the recognition status of the obstacle in the first image data by the target recognition model. The first recognition result is used to test the target recognition model.

[0012] On the other hand, a detection device for a target recognition model is provided, the device comprising:

[0013] an acquisition module, configured to acquire a weather scenario configuration file and a working condition configuration file, wherein the weather scenario configuration file is used to generate a simulated weather scenario when the autonomous driving vehicle is driving, and the working condition configuration file is used to generate a working condition scenario when the autonomous driving vehicle is driving, and the working condition configuration file includes an obstacle configuration part and a driving condition configuration part, wherein the obstacle configuration part includes the configuration of obstacles encountered by the autonomous driving vehicle when driving, and the driving condition configuration part includes the working condition of the autonomous driving vehicle when driving, and the working condition includes at least one of vehicle emergency braking, vehicle emergency avoidance, and vehicle traffic congestion;

[0014] A generating module, configured to generate a three-dimensional simulated driving environment based on the weather scenario configuration file and the operating condition configuration file, wherein the three-dimensional simulated driving environment includes three-dimensional models of the autonomous driving vehicle and the obstacles;

[0015] A collection module, used to collect sample image data based on the three-dimensional simulated driving environment, wherein the sample image data is an environment image collected from the perspective of the autonomous driving vehicle, and the environment image contains the obstacle;

[0016] The generating module is further used to perform an iterative attack on the sample image data based on a projected gradient descent method to generate first image data, where the first image data is image data obtained by superimposing noise on the sample image data;

[0017] An identification module is used to identify the first image data through a target recognition model corresponding to the autonomous driving vehicle to obtain a first recognition result, wherein the first recognition result includes the recognition status of the obstacle in the first image data by the target recognition model, and the first recognition result is used to test the target recognition model.

[0018] In an optional embodiment, the generation module is further used to input the weather scenario configuration file into the autonomous driving simulator, and construct the background part in the three-dimensional simulated driving environment through the autonomous driving simulator; input the obstacle configuration part in the working condition configuration file into the autonomous driving simulator, and generate a three-dimensional model of the obstacle in the three-dimensional simulated driving environment through the autonomous driving simulator; the obstacle includes at least one of a vehicle and a pedestrian; input the driving condition configuration part in the working condition configuration file into the autonomous driving simulator, and generate a three-dimensional model of the autonomous driving vehicle in the three-dimensional simulated driving environment through the autonomous driving simulator.

[0019] In an optional embodiment, the acquisition module is also used to control the three-dimensional models of the autonomous driving vehicle and the obstacle to move in the three-dimensional simulated driving environment through the autonomous driving simulator; and to acquire the environmental image corresponding to the three-dimensional simulated driving environment based on the perspective of the autonomous driving vehicle to obtain the sample image data, wherein the sample image data includes the distribution and position information of the obstacle in the environmental image.

[0020] In an optional embodiment, the generation module is also used to convert the format of the sample image data to obtain first test data that meets preset format requirements; based on the projected gradient descent method, the first test data is iteratively attacked to generate the first image data, the iterative attack refers to the process of superimposing noise on the first test data, and the number of iterations of the iterative attack is a preset threshold; wherein the data obtained after the i-th attack on the first test data is used as the data used for the i+1-th attack, and i is a positive integer.

[0021] In an optional embodiment, before the generating module, the device further includes:

[0022] The category determination module is used to perform category analysis on the sample image data to determine a category label of the sample image data, wherein the category label is used to indicate category information of the obstacle in the sample image data.

[0023] In an optional embodiment, after the identification module, the device further includes:

[0024] A training module is used to determine that the target recognition model does not meet the preset accuracy requirement when recognizing image data when the difference between the first recognition result and the category label reaches a preset difference threshold; obtain a sample image data set, wherein the sample image data set includes multiple image data, and the format of the multiple image data is the same as that of the sample image data; iteratively attack the sample image data set based on the projected gradient descent method to generate a training image data set; wherein the training image data set includes multiple training image data, and the format of the multiple training image data is the same as that of the first image data; and train the target recognition model based on the training image data set.

[0025] On the other hand, a computer device is provided, which includes a processor and a memory, wherein the memory stores at least one instruction, at least one program, a code set or an instruction set, and the at least one instruction, the at least one program, the code set or the instruction set is loaded and executed by the processor to implement a detection method for a target recognition model as described in any of the above-mentioned embodiments of the present application.

[0026] On the other hand, a computer-readable storage medium is provided, in which at least one instruction, at least one program, a code set or an instruction set is stored, and the at least one instruction, the at least one program, the code set or the instruction set is loaded and executed by a processor to implement a detection method for a target recognition model as described in any of the above-mentioned embodiments of the present application.

[0027] On the other hand, a computer program product or a computer program is provided, the computer program product or the computer program includes computer instructions, the computer instructions are stored in a computer-readable storage medium. A processor of a computer device reads the computer instructions from the computer-readable storage medium, and the processor executes the computer instructions, so that the computer device executes the detection method of the target recognition model described in any of the above embodiments.

[0028] The beneficial effects brought by the technical solution provided by the embodiment of the present application include at least:

[0029] By customizing a variety of working conditions and weather scenarios, and using an autonomous driving simulator to simulate the corresponding simulation environment, we can fully cover all kinds of situations that autonomous driving vehicles may encounter when driving, which is close to reality and improves the simulation test effect. Generate environmental image information to simulate the images that may be collected from the perspective of the autonomous driving vehicle. The images contain the target content to be identified. Use images with adversarial perturbations as input data for the target recognition model of the autonomous driving vehicle. Determine the recognition accuracy and robustness of the target recognition model based on whether the target recognition model can accurately identify the disturbed target content. Use the projected gradient descent method to iteratively attack the image to generate adversarial perturbations, and you can obtain a large amount of perturbed image data. These image data can be used as training data for the target recognition model, specifically improving the training effect of the target recognition model, so that the target recognition model has the ability to resist attack perturbations. BRIEF DESCRIPTION OF THE DRAWINGS

[0030] In order to more clearly illustrate the technical solutions in the embodiments of the present application, the drawings required for use in the description of the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work.

[0031] Figure 1 is a schematic diagram of a detection system of a target recognition model provided by an exemplary embodiment of the present application;

[0032] Figure 2 is a flow chart of a detection method of a target recognition model provided by an exemplary embodiment of the present application;

[0033] Figure 3 It is a schematic diagram of comparison of environmental images collected when the obstacle is a bus in a sunny day and before and after being attacked, provided by an exemplary embodiment of the present application;

[0034] Figure 4 is a structural block diagram of a detection device for a target recognition model provided by an exemplary embodiment of the present application;

[0035] Figure 5 is a structural block diagram of a detection device for a target recognition model provided by another exemplary embodiment of the present application;

[0036] Figure 6 It is a structural block diagram of a computer device provided by an exemplary embodiment of the present application. DETAILED DESCRIPTION

[0037] In order to make the objectives, technical solutions and advantages of the present application clearer, the implementation methods of the present application will be further described in detail below with reference to the accompanying drawings.

[0038] Exemplary embodiments will be described in detail herein, examples of which are shown in the accompanying drawings. When the following description refers to the drawings, the same numbers in different drawings represent the same or similar elements unless otherwise indicated. The implementations described in the following exemplary embodiments do not represent all implementations consistent with the present application. Instead, they are merely examples of devices and methods consistent with some aspects of the present application as detailed in the appended claims.

[0039] The terms used in this application are for the purpose of describing specific embodiments only and are not intended to limit this application. The singular forms of "a", "said" and "the" used in this application and the appended claims are also intended to include plural forms unless the context clearly indicates other meanings. It should also be understood that the term "and / or" used herein refers to and includes any or all possible combinations of one or more associated listed items.

[0040] It should be noted that the information and data involved in this application are all information and data authorized by the user or fully authorized by all parties, and the collection, use and processing of relevant data must comply with relevant laws, regulations and standards of relevant countries and regions.

[0041] It should be understood that, although the terms first, second, etc. may be used in the present application to describe various information, these information should not be limited to these terms. These terms are only used to distinguish the same type of information from each other. For example, without departing from the scope of the present application, the first parameter may also be referred to as the second parameter, and similarly, the second parameter may also be referred to as the first parameter. Depending on the context, the word "if" as used herein may be interpreted as "at the time of" or "when" or "in response to determining".

[0042] First, a brief introduction is given to the terms involved in the embodiments of this application:

[0043] Projected Gradient Descent (PGD): is an optimization algorithm that is often used to solve nonlinear optimization problems with constraints. In machine learning and deep learning, PGD can be used to train models, especially in adversarial training scenarios, to generate adversarial samples, or for regularization in model optimization. The basic idea of ​​PGD is to iteratively search along the negative gradient direction of the objective function, but at each iteration, the solution is projected back to the feasible domain to satisfy the constraints of the problem. In adversarial training, the PGD algorithm can be used to generate adversarial samples by looking for tiny perturbations of the input samples near the decision boundary of the model, which are almost imperceptible to human vision but enough to make the model make wrong predictions. PGD iteratively adjusts the input samples, each adjustment along the gradient direction of the model output, and then ensures that the generated adversarial samples satisfy specific constraints through a projection step.

[0044] This application is based on the perception of the external environment by an autonomous vehicle. The autonomous vehicle collects environmental images from its own perspective, recognizes the environmental images through a target recognition model, determines whether there is a target content (which can be called an obstacle) in the environmental image, and promptly feeds back the recognition results. The autonomous vehicle determines whether to continue driving or stop based on the recognition results. Therefore, testing the robustness of the target recognition model is the key to the autonomous vehicle's ability to correctly perceive the environment and travel safely.

[0045] In this application, the gradient projection descent method is used to iteratively attack the image to obtain images with interference. These images with interference will be input into the target recognition model of the autonomous driving vehicle as input data to test the robustness of the target recognition model. That is, whether the target recognition model needs to be trained in a targeted manner is determined based on whether the recognition result of the target recognition model for the target content in the image (the target content can be referred to as an obstacle in this application, that is, the subject that the autonomous driving vehicle needs to avoid, etc.) is accurate. If the gradient projection descent method effectively attacks the image, which means that the recognition of the target recognition model is interfered with, the gradient projection descent method can be used to generate a large number of interfered images as training data for targeted training of the target recognition model, thereby improving the accuracy and robustness of the model recognition results.

[0046] KITTI / kitti format (a data format): KITTI format is a general term for a series of data formats related to the KITTI dataset, which is a large benchmark dataset widely used in computer vision and autonomous driving research. Exemplarily, the specific file formats of the KITTI dataset include the following. Image files: usually stored in 8-bit PNG (Portable Network Graphics) format, which may contain cropped and distortion-corrected images; LiDAR point cloud data: stored in binary format, containing three-dimensional point cloud data obtained by LiDAR scanning, each point contains x, y, z coordinates and reflection intensity information.

[0047] SMOKE model (Single-Stage Monocular 3D Object Detection via KeypointEstimation): It directly predicts the 3D attribute information of the target through key point estimation and 3D (Dimension) box regression, instead of relying on traditional 2D detection methods, thus avoiding the noise that may be introduced by 2D detection and improving the performance of 3D detection.

[0048] The target recognition model in this application is a model trained with the SMOKE model as the prototype, and is mainly used to perform target recognition on the collected images, identify the target content in the image (the target content can be an obstacle, such as a vehicle or pedestrian, etc.) The position and type of the target content in the image. The automatic recognition model is applied to the environmental perception of the autonomous driving vehicle. The recognition result of the automatic recognition model is used as the judgment basis of the autonomous driving vehicle to determine the next action of the autonomous driving vehicle.

[0049] Automated Driving Simulator (CARLA): provides a modular and flexible application programming interface for solving a range of tasks in autonomous driving research and development.

[0050] In the present application, an autonomous driving simulator is used to generate a three-dimensional simulation environment to simulate the environment in which an autonomous driving vehicle is driving. The three-dimensional simulation environment includes specified weather scenes, three-dimensional models of obstacles, and three-dimensional models of autonomous driving vehicles. In the three-dimensional simulation environment, all vehicles and obstacles move according to preset operating scenarios.

[0051] Next, the detection system of the target recognition model involved in the embodiment of the present application is described. For illustration, please refer to Figure 1 The implementation environment involves a terminal 110 and a server 120. The terminal 110 and the server 120 are connected via a communication network 130.

[0052] Terminal 110 includes an autonomous driving simulator, which can build a three-dimensional simulation environment to simulate the driving environment of an autonomous driving vehicle. In order to cover comprehensive weather scenarios and working conditions, the technician first uploads a weather scenario configuration file and a working condition configuration file to terminal 110. The weather scenario configuration file is used to generate a simulated weather scenario when the autonomous driving vehicle is driving, and the working condition configuration file is used to generate a working condition scenario when the autonomous driving vehicle is driving. Among them, the working condition configuration file covers the driving conditions of the autonomous driving vehicle itself and the types of obstacles and road conditions encountered by the autonomous driving vehicle.

[0053] The autonomous driving simulator in terminal 110 can construct a corresponding three-dimensional simulation environment based on the weather scenario configuration file and the operating condition configuration file. In the three-dimensional simulation environment, the autonomous driving vehicle drives as the main body and collects images.

[0054] For example, the three-dimensional simulation environment simulates the situation of an autonomous driving vehicle driving on the road on a sunny day. During the driving process of the autonomous driving vehicle, there is an obstacle in the direction of its travel, which is a bus parked on the side of the road.

[0055] After the terminal 110 constructs the three-dimensional simulation environment, it collects environmental images, which are images collected from the perspective of the autonomous driving vehicle and are used to reflect the surrounding environment during driving. Some environmental images contain obstacles as target content, for example, obstacles are vehicles and pedestrians that require the autonomous driving vehicle to avoid or stop in time. In some embodiments, the target content also includes road conditions. For example, when the number of obstacles in the environmental image reaches a preset number threshold, it represents current road congestion.

[0056] The terminal 110 performs an iterative attack on the collected environment image based on the projected gradient descent method (PGD). The purpose of the iterative attack operation is to interfere with the target recognition model of the autonomous driving vehicle when recognizing the environment image. In other words, the attack on the environment image is actually to change part of the information in the environment image, so that the target recognition model of the autonomous driving vehicle cannot accurately recognize the target content, make an erroneous judgment, and affect the driving safety of the autonomous driving vehicle.

[0057] The terminal 110 sends the first image data obtained after the iterative attack to the server 120. The server 120 includes a target recognition model. The target recognition model recognizes the first image data to obtain a first recognition result. The first recognition result includes the situation in which the target recognition model recognizes the target content, including the type of the target content and the location information of the target content in the image. The first recognition result is used to test the target recognition model, reflecting the accuracy and robustness of the target recognition model when recognizing the attacked image.

[0058] In some embodiments, the target recognition model cannot accurately identify the target content in the image attacked by the gradient projection descent method. Therefore, a batch of attacked images can be obtained based on the gradient projection descent method as training data to train the target recognition model.

[0059] The above-mentioned terminal can be a mobile phone, a tablet computer, a desktop computer, a portable laptop computer, a smart TV, a car terminal, a smart home device and other terminal devices in various forms, which is not limited in the embodiments of the present application.

[0060] It is worth noting that the above-mentioned servers can be independent physical servers, or they can be server clusters or distributed systems composed of multiple physical servers. They can also be cloud servers that provide basic cloud computing services such as cloud services, cloud databases, cloud computing, cloud functions, cloud storage, network services, cloud communications, middleware services, domain name services, security services, content delivery networks (CDN), as well as big data and artificial intelligence platforms.

[0061] In some embodiments, the above server can also be implemented as a node in a blockchain system.

[0062] In combination with the above-mentioned noun introduction and application scenarios, the detection method of the target recognition model provided by the present application is described. The method can be executed by a server or a terminal, or can be executed by a server and a terminal together. In the embodiment of the present application, the method is executed by a terminal as an example. Figure 2 As shown, Figure 2 FIG. 1 is a flow chart of a detection method of a target recognition model provided by an exemplary embodiment of the present application. The method includes the following steps.

[0063] Step 210: Obtain a weather scenario configuration file and an operating condition configuration file.

[0064] The weather scenario configuration file is used to generate simulated weather scenarios when the autonomous vehicle is driving, and the operating condition configuration file is used to generate operating condition scenarios when the autonomous vehicle is driving.

[0065] Optionally, the weather scene configuration file includes meteorological pictures and descriptions corresponding to various weather conditions. For example, if the weather scene is snowy, the meteorological picture includes various pictures for representing the snowy climate, such as a picture of snowfall over a city road; the description is as follows: the snowfall is 2.5 mm thick in 24 hours (i.e., the accumulated thickness on the ground is 2.5 mm), the color is white, and the shape is a hexagon.

[0066] The operating condition configuration file includes an obstacle configuration part and a driving condition configuration part. The obstacle configuration part includes the configuration of obstacles encountered by the autonomous driving vehicle when driving, and the driving condition configuration part includes the working conditions of the autonomous driving vehicle when driving. The working conditions include at least one of vehicle emergency braking, vehicle emergency avoidance, and vehicle traffic congestion.

[0067] Optionally, the obstacle configuration part includes pictures and obstacle information of various types of obstacles. For example, if the obstacle is a bus, the obstacle configuration part includes pictures of buses of various shapes; the obstacle information is as follows: the name of the obstacle, and the three-dimensional size (length, width, height).

[0068] The weather scenario configuration file and the operating condition configuration file are files prepared in advance by a technician and uploaded to the terminal. In some embodiments, the terminal includes an application capable of generating weather scenario configuration and operating condition configuration files, and the technician can directly configure the above files in the terminal.

[0069] Optionally, the weather scene profile corresponds to 10 weather scenes, namely: (1) sunny;

[0070] (2) Light rain; (3) Moderate rain; (4) Heavy rain; (5) Fog; (6) Snow; (7) Shadow: The in-car field of vision is blocked and creates a shadow. For example, when an autonomous vehicle passes under the shade of a tree, the surrounding environment from the perspective of the in-car is covered by shadows and the light is weak. (8) Midnight: refers to the scene when the vehicle is driving during the midnight period. For example, the period from 00:00:00 to 4:00:00 on the same day can be regarded as midnight, and the surrounding environment from the perspective of the autonomous vehicle is dim or invisible. (9) Backlight: The in-car field of vision is affected by strong light, making it impossible to accurately observe the surrounding environment. For example, when an autonomous vehicle is driving, there is sunrise / strong light in front and the vehicle is driving against the light. (10) Motion blur: The in-car field of vision is blurred due to movement when the autonomous vehicle is driving at a high speed, making a sharp turn, or bumping.

[0071] Optionally, the obstacle configuration part corresponds to 7 types of obstacles, namely: buses, trucks, motorcycles, special vehicles, cars, passenger cars, and pedestrians. These obstacles are obstacles that the autonomous driving vehicle may encounter during driving.

[0072] Optionally, the driving condition configuration part corresponds to five working conditions, namely: (1) vehicle traffic congestion: the road where the autonomous driving vehicle is located is congested, for example, there are 10 vehicles visible in the field of vision inside the vehicle, and the distance between each vehicle does not exceed 30 cm; (2) vehicle emergency avoidance: other obstacles suddenly appear during the driving of the autonomous driving vehicle, and the autonomous driving vehicle needs to make emergency avoidance. For example, when the autonomous driving vehicle passes an intersection, a car or a pedestrian suddenly appears from the turning point of the intersection. At this time, the autonomous driving vehicle needs to make emergency avoidance; (3) vehicle emergency braking: the autonomous driving vehicle suddenly brakes / brakes during driving. For example, when the autonomous driving vehicle is driving at a speed of 40km / h, the vehicle in front suddenly breaks down and stops, and the autonomous driving vehicle brakes urgently; (4) the vehicle in front leaves: the vehicle in front of the autonomous driving vehicle leaves the current lane. For example, after the vehicle in front changes lanes, there is no vehicle in front of the autonomous driving vehicle; (5) adaptive cruise: when the autonomous driving vehicle follows a vehicle in front, it needs to maintain the following distance in real time according to the speed of the vehicle in front.

[0073] Among them, the files of the obstacle configuration part include files for describing the shape of the obstacle and generating a three-dimensional model of the obstacle. For example, the obstacle configuration part includes obstacle pictures, and the pictures include the shapes of the obstacles in different views (including left view, right view, main view, etc.).

[0074] Step 220 , generating a three-dimensional simulated driving environment based on the weather scenario configuration file and the operating condition configuration file.

[0075] Among them, the three-dimensional simulated driving environment includes three-dimensional models of autonomous driving vehicles and obstacles, and the terminal includes an autonomous driving simulator. The autonomous driving simulator can generate a corresponding three-dimensional model based on the input configuration file to simulate the driving environment of the autonomous driving vehicle.

[0076] Optionally, the weather scenario profile is input into the autonomous driving simulator, and the autonomous driving simulator is used to construct a background portion of the three-dimensional simulated driving environment.

[0077] The background part here refers to the physical environment when the autonomous driving vehicle is driving under the specified weather scenario. For example, the weather scenario configuration file contains a part for configuring the light rain weather scene. In the scene constructed by the autonomous driving simulator, the autonomous driving vehicle is driving on a country road. Then, the background part is used to indicate the road conditions of the country road in light rain weather. The autonomous driving vehicle will simulate movement in the physical environment indicated by the background part.

[0078] Optionally, the obstacle configuration portion of the operating condition configuration file is input into an autonomous driving simulator, and a three-dimensional model of the obstacle is generated in a three-dimensional simulated driving environment by the autonomous driving simulator, wherein the obstacle includes at least one of a vehicle and a pedestrian.

[0079] For example, the operating condition configuration file indicates that the obstacle is a bus or a motorcycle, and the obstacle configuration part contains a picture of the obstacle and information describing the size of the obstacle.

[0080] The autonomous driving simulator will generate a three-dimensional model of the obstacle based on the obstacle configuration part and put it into the background part of the three-dimensional simulated driving environment. Among them, the ratio between the three-dimensional model of the obstacle and the three-dimensional model of the physical environment meets the preset mapping requirements. For example, the three-dimensional model of the physical environment simulates the real space volume of 10,000 cubic meters, and the obstacle size information indicates that the volume of the obstacle is 90 cubic meters. In fact, the volume of the three-dimensional model displayed on the terminal display screen is proportionally reduced, and the space volume: three-dimensional model volume = 10,000:1.

[0081] Optionally, the driving condition configuration part of the condition configuration file is input into the autonomous driving simulator, and a three-dimensional model of the autonomous driving vehicle is generated in a three-dimensional simulated driving environment through the autonomous driving simulator.

[0082] After the three-dimensional models of obstacles, autonomous driving vehicles, and physical environments are generated, a three-dimensional simulated driving environment is obtained, in which the three-dimensional models of obstacles and the three-dimensional model of autonomous driving vehicles perform corresponding activities based on the working conditions indicated by the driving condition configuration part.

[0083] For example, if the weather condition is light rain, the three-dimensional simulated driving environment will simulate and display the rainfall. If the physical environment is a city road, the three-dimensional simulated driving environment will simulate and display the interlaced relationship between different road trunk roads and the traffic lights arranged on each road. The driving condition configuration part indicates that the autonomous driving vehicle is in a congested road condition, and the obstacle configuration part configures 10 cars and 2 pedestrians as obstacles. The three-dimensional models of the above obstacles are distributed around the three-dimensional model of the autonomous driving vehicle to form a congested road condition. In some embodiments, obstacles (vehicles, pedestrians) and autonomous driving vehicles travel at a slow speed along the specified direction of the road.

[0084] Exemplarily, the specific implementation steps are as follows. S1: Model file import: put the pictures of obstacle models such as vehicles and pedestrians into the import directory of the autonomous driving simulator (CARLA), modify the configuration file related information to the actual information of the model file, execute the make import command in the directory where CARLA is located, wait for the compilation to complete, and the import will be successful; wherein, the actual information of the model file includes the shape, size, and name of the obstacle; S2: Weather scene construction: based on the autonomous driving simulator, a total of 10 weather scenes including sunny days, light rain, moderate rain, heavy rain, fog, snow, shadow, midnight, backlight and motion blur are constructed; S3: Working condition scene construction: based on the autonomous driving simulator, at least 12 test working conditions including buses, trucks, motorcycles, special vehicles, cars, passenger cars, pedestrians, traffic congestion, emergency avoidance, emergency braking, front vehicle departure, and adaptive cruise are constructed.

[0085] Step 230 : collecting sample image data based on the three-dimensional simulated driving environment.

[0086] Among them, the sample image data is an environmental image collected from the perspective of an autonomous driving vehicle, and the environmental image contains obstacles.

[0087] The obstacles in the environment image are the parts that the target recognition model is expected to be able to identify. If the target recognition model can accurately identify the type of obstacles and their location information in the environment image, it means that the target recognition model has the ability to provide environmental perception for the autonomous driving vehicle. In some embodiments, in addition to the obstacles as target content, the environment image also reflects the driving status of the autonomous driving vehicle itself, or the environment image contains prompt information to guide the autonomous driving vehicle's next driving status, such as a sign, which is used to instruct the vehicle to stop, slow down, return to the original route, etc.

[0088] The autonomous driving simulator controls the three-dimensional models of autonomous driving vehicles and obstacles to move in a three-dimensional simulated driving environment.

[0089] Based on the perspective of the autonomous driving vehicle, an environmental image corresponding to the three-dimensional simulated driving environment is collected to obtain sample image data, wherein the sample image data includes the distribution and position information of obstacles in the environmental image.

[0090] Exemplarily, image information of the autonomous driving vehicle under different weather and working conditions is collected based on the autonomous driving simulator. A total of 105 images are generated in each working condition in a single scene. The image information includes two parts: PNG (Portable Network Graphics, a common image file format) file and NPZ (NumPy Zip, a compressed file format) file. The PNG file contains visual data of the image information, that is, pixel information that the user can see. The NPZ file contains additional information related to the image, including the numerical value of each pixel in the image.

[0091] In this embodiment, the PNG file and the NPZ file represent the vehicle driving state and the vehicle positioning information respectively. The vehicle driving state refers to the driving state of the autonomous driving vehicle in the three-dimensional simulated driving environment and the driving state of obstacles collected from its own perspective. The vehicle positioning information includes the position information of the obstacles in the image, and the vehicle positioning information can reflect the distance and relative position between the autonomous driving vehicle and the obstacles.

[0092] Optionally, the obtained PNG file and NPZ file are converted into testable data in kiti format, and the files are integrated to generate category labels for the corresponding images. Category labels are used to indicate the name and type of obstacles in the image. For example, if the obstacle contained in one of the collected environment images is a bus, the environment image is converted to obtain test data in kiti format, and the category label obtained based on the test data is "bus".

[0093] Among them, the format conversion of image data is to obtain a data format suitable for detection by the target recognition model. The target recognition model is a model trained based on the SMOKE model. The image data in the kiti data format can be input into the target recognition model as input data. The target recognition model recognizes the kitti format data, determines the target content in the data, and outputs the recognition result.

[0094] Step 240: Perform an iterative attack on the sample image data based on the projected gradient descent method to generate first image data.

[0095] The first image data is image data obtained by superimposing noise on the sample image data.

[0096] The attack on sample image data by the gradient projection descent method refers to the process of superimposing noise on the sample image data. The superimposed noise is to interfere with the target recognition model when recognizing the image. The impact of these noise interferences on the image is not obvious to the human eye. However, for the target recognition model, slight changes will also interfere with the target recognition model, resulting in inaccurate recognition results of the target recognition model and wrong judgments, which in turn affects the safety of the autonomous driving vehicle during driving.

[0097] The sample image data is format converted to obtain first test data that meets the preset format requirements.

[0098] Exemplarily, the sample image data is converted into KITTI format, then the first test data is test data in KITTI format, and the data in KITTI format is a data type recognizable by the target recognition model.

[0099] An iterative attack is performed on the first test data based on the projected gradient descent method to generate first image data. The iterative attack refers to the process of superimposing noise on the first test data. The number of iterations of the iterative attack is a preset threshold. The data obtained after the i-th attack on the first test data is used as the data used for the i+1-th attack, and i is a positive integer.

[0100] That is, when attacking the first test data, the noise superposition process is cumulative. When the previous round of attack is completed, an intermediate output result will be obtained, and the next round of attack will be carried out on the intermediate output result, that is, noise will continue to be superimposed on the data of the intermediate output result.

[0101] Exemplarily, a digital attack is performed on the first test data after conversion into the kitti format. The attack method is the gradient projection descent method (PGD). Noise is superimposed on the original image, and the noise is updated in the direction of maximizing the loss value of the target recognition model (the SMOKE model in this application).

[0102] Here’s how it works:

[0103] Among them, the noise superposition of the image is achieved by changing the numerical value of the pixels in the image. x is the input of the target recognition model, which is specifically reflected in the pixel value of the image; θ is the parameter value of the target model, y is the label value of the target recognition model, and L(θ, x, y) is the loss function of the target recognition model. is the gradient of the input x, sgn is the direction of the gradient, S is the added perturbation, t is the number of attack iterations, and α is the perturbation factor. t+1 It refers to the value of the pixel at the t+1th iteration.

[0104] The iterative process mainly involves the following key parameters: number of iterations t=20, threshold eps=76, and perturbation factor α=2. That is, the number of attacks on the first test data is 20 times. The threshold refers to the upper limit of the change of a single pixel of the image during each attack. The pixel value range is 0-255. The change value of a single pixel before and after each attack on the image does not exceed 76. In digital attacks, especially when generating adversarial perturbations, the perturbation factor refers to a parameter used to quantify the strength of the adversarial perturbation. The perturbation factor is usually expressed as a small value, which defines the limit of the size or amplitude of the adversarial perturbation. The perturbation factor controls the difference between the adversarial sample (a sample with noise superimposed, such as the test data obtained after attacking the first test data in this application) and the original sample (such as the first test data in this application). For example, the original input is x, and the adversarial sample can be expressed as x+α*noise, where α is the perturbation factor and noise is the perturbation vector calculated according to the attack algorithm (such as the PGD algorithm in this application).

[0105] For illustration, please refer to Figure 3 , Figure 3 It is a schematic diagram comparing environmental images collected when the obstacle is a bus on a sunny day before and after the attack, wherein the difference between the sample image 310 before the attack and the sample image 320 after the attack is that the surface coating of the obstacle bus 330 has changed.

[0106] Step 250: recognize the first image data using a target recognition model corresponding to the autonomous driving vehicle to obtain a first recognition result.

[0107] The first recognition result includes recognition of obstacles in the first image data by the target recognition model, and the first recognition result is used to test the target recognition model.

[0108] Optionally, before iteratively attacking the sample image data based on the projected gradient descent method to generate the first image data, a category analysis is performed on the sample image data to determine a category label of the sample image data, where the category label is used to indicate category information of obstacles in the sample image data.

[0109] When the difference between the first recognition result and the category label reaches a preset difference threshold, it is determined that the target recognition model does not meet the preset accuracy requirement when recognizing the image data.

[0110] For example, the obstacle in the sample image data is a signboard. The pattern on the signboard indicates that there is construction ahead and the road is impassable. The autonomous driving vehicle should return along the original route. Then, the category label indicates that the obstacle in the sample image data is: construction signboard. The first image data is obtained after iteratively attacking the sample image data based on the projected gradient descent method. The first recognition result of the target recognition model when recognizing the first image data indicates that the automatic recognition model recognizes the construction sign as a deceleration sign. The first recognition result is different from the meaning expressed by the category label. The difference is 80%, which reaches the preset difference threshold of 60%. At this time, it is determined that the target recognition model does not meet the preset accuracy requirement when recognizing the image data.

[0111] When the projected gradient descent method causes an effective attack on the target recognition model, a batch of attacked data can be generated based on this method to train the target recognition model, so that the target recognition model has the ability to resist digital attacks and improve the robustness and recognition accuracy of the target recognition model.

[0112] Optionally, a sample image data set is obtained, wherein the sample image data set includes multiple image data, and the format of the multiple image data is the same as that of the sample image data. The sample image data set is iteratively attacked based on the projected gradient descent method to generate a training image data set; wherein the training image data set includes multiple training image data, and the format of the multiple training image data is the same as that of the first image data; and the target recognition model is trained based on the training image data set.

[0113] In summary, the method provided by the present application, by customizing a variety of working conditions and weather scenes, uses an autonomous driving simulator to simulate the corresponding simulation environment, comprehensively covers various situations that may be encountered when an autonomous driving vehicle is driving, is in line with reality, and improves the simulation test effect. Generate environmental image information, simulate the images that may be collected from the perspective of the autonomous driving vehicle, the images contain target content to be identified, use images with adversarial disturbances as input data for the target recognition model of the autonomous driving vehicle, and determine the recognition accuracy and robustness of the target recognition model based on whether the target recognition model can accurately identify the disturbed target content. Use the projected gradient descent method to iteratively attack the image to generate adversarial disturbances, and a large amount of disturbed image data can be obtained. These image data can be used as training data for the target recognition model, and the training effect of the target recognition model can be improved in a targeted manner, so that the target recognition model has the ability to resist attack disturbances.

[0114] Figure 4 is a structural block diagram of a detection device for a target recognition model provided by an exemplary embodiment of the present application, such as Figure 4 As shown, the device includes the following parts.

[0115] an acquisition module 410, configured to acquire a weather scenario configuration file and a working condition configuration file, wherein the weather scenario configuration file is used to generate a simulated weather scenario when the autonomous driving vehicle is driving, and the working condition configuration file is used to generate a working condition scenario when the autonomous driving vehicle is driving, wherein the working condition configuration file includes an obstacle configuration part and a driving condition configuration part, wherein the obstacle configuration part includes the configuration of obstacles encountered by the autonomous driving vehicle when driving, and the driving condition configuration part includes the working condition of the autonomous driving vehicle when driving, and the working condition includes at least one of vehicle emergency braking, vehicle emergency avoidance, and vehicle traffic congestion;

[0116] A generating module 420, configured to generate a three-dimensional simulated driving environment based on the weather scenario configuration file and the operating condition configuration file, wherein the three-dimensional simulated driving environment includes three-dimensional models of the autonomous driving vehicle and the obstacles;

[0117] An acquisition module 430 is used to acquire sample image data based on the three-dimensional simulated driving environment, wherein the sample image data is an environment image acquired from the perspective of the autonomous driving vehicle, and the environment image contains the obstacle;

[0118] The generating module 420 is further configured to perform an iterative attack on the sample image data based on a projected gradient descent method to generate first image data, where the first image data is image data obtained by superimposing noise on the sample image data;

[0119] The recognition module 440 is used to recognize the first image data through the target recognition model corresponding to the autonomous driving vehicle to obtain a first recognition result. The first recognition result includes the recognition of the obstacle in the first image data by the target recognition model. The first recognition result is used to test the target recognition model.

[0120] In an optional embodiment, the generation module 420 is further used to input the weather scenario configuration file into the autonomous driving simulator, and construct the background part in the three-dimensional simulated driving environment through the autonomous driving simulator; input the obstacle configuration part in the working condition configuration file into the autonomous driving simulator, and generate a three-dimensional model of the obstacle in the three-dimensional simulated driving environment through the autonomous driving simulator; the obstacle includes at least one of a vehicle and a pedestrian; input the driving condition configuration part in the working condition configuration file into the autonomous driving simulator, and generate a three-dimensional model of the autonomous driving vehicle in the three-dimensional simulated driving environment through the autonomous driving simulator.

[0121] In an optional embodiment, the acquisition module 430 is also used to control the three-dimensional model of the autonomous driving vehicle and the obstacle to move in the three-dimensional simulated driving environment through the autonomous driving simulator; and to acquire the environmental image corresponding to the three-dimensional simulated driving environment based on the perspective of the autonomous driving vehicle to obtain the sample image data, wherein the sample image data includes the distribution and position information of the obstacle in the environmental image.

[0122] In an optional embodiment, the generation module 420 is further used to perform format conversion on the sample image data to obtain first test data that meets preset format requirements; perform an iterative attack on the first test data based on a projected gradient descent method to generate the first image data, wherein the iterative attack refers to a process of superimposing noise on the first test data, and the number of iterations of the iterative attack is a preset threshold; wherein the data obtained after the i-th attack on the first test data is used as the data used for the i+1-th attack, and i is a positive integer.

[0123] In an optional embodiment, before the generating module 420, as Figure 5 As shown, the device also includes:

[0124] The category determination module 450 is used to perform category analysis on the sample image data to determine a category label of the sample image data, where the category label is used to indicate category information of the obstacle in the sample image data.

[0125] In an optional embodiment, after the identification module, the device further includes:

[0126] A training module 460 is used to determine that the target recognition model does not meet a preset accuracy requirement when recognizing image data when the difference between the first recognition result and the category label reaches a preset difference threshold; obtain a sample image data set, wherein the sample image data set includes multiple image data, and the multiple image data are in the same format as the sample image data; iteratively attack the sample image data set based on the projected gradient descent method to generate a training image data set; wherein the training image data set includes multiple training image data, and the multiple training image data are in the same format as the first image data; and train the target recognition model based on the training image data set.

[0127] In summary, the detection device of the target recognition model provided by the present application can fully cover various situations that may be encountered when the autonomous driving vehicle is driving by customizing a variety of working conditions and weather scenes and using an autonomous driving simulator to simulate the corresponding simulation environment, which is in line with reality and improves the simulation test effect. Generate environmental image information to simulate the images that may be collected from the perspective of the autonomous driving vehicle, the images contain the target content to be identified, use the images with adversarial disturbances as the input data of the target recognition model of the autonomous driving vehicle, and determine the recognition accuracy and robustness of the target recognition model based on whether the target recognition model can accurately identify the disturbed target content. Use the projected gradient descent method to iteratively attack the image to generate adversarial disturbances, and a large amount of image data with disturbances can be obtained. These image data can be used as training data for the target recognition model to improve the training effect of the target recognition model in a targeted manner, so that the target recognition model has the ability to resist attack disturbances.

[0128] It should be noted that the detection device of the target recognition model provided in the above embodiment is only illustrated by the division of the above functional modules. In actual applications, the above functions can be assigned to different functional modules as needed, that is, the internal structure of the device is divided into different functional modules to complete all or part of the functions described above. In addition, the detection device of the target recognition model provided in the above embodiment and the detection method embodiment of the target recognition model belong to the same concept. The specific implementation process is detailed in the method embodiment and will not be repeated here.

[0129] Figure 6 The block diagram of a computer device 600 provided by an exemplary embodiment of the present application is shown. The computer device 600 may be: a smart phone, a tablet computer, an MP3 player (Moving Picture Experts Group Audio Layer III, Moving Picture Experts Group Audio Layer 3), an MP4 player (Moving Picture Experts Group Audio Layer IV, Moving Picture Experts Group Audio Layer 4), a laptop computer or a desktop computer. The computer device 600 may also be called a user device, a portable terminal, a laptop terminal, a desktop terminal or other names.

[0130] Typically, the computer device 600 includes a processor 601 and a memory 602 .

[0131] The processor 601 may include one or more processing cores, such as a 4-core processor, an 8-core processor, etc. The processor 601 may be implemented in at least one hardware form of DSP (Digital Signal Processing), FPGA (Field-Programmable Gate Array), and PLA (Programmable Logic Array). The processor 601 may also include a main processor and a coprocessor. The main processor is a processor for processing data in the awake state, also known as a CPU (Central Processing Unit); the coprocessor is a low-power processor for processing data in the standby state. In some embodiments, the processor 601 may be integrated with a GPU (Graphics Processing Unit), which is responsible for rendering and drawing the content to be displayed on the display screen. In some embodiments, the processor 601 may also include an AI (Artificial Intelligence) processor, which is used to process computing operations related to machine learning.

[0132] The memory 602 may include one or more computer-readable storage media, which may be non-transitory. The memory 602 may also include a high-speed random access memory, and a non-volatile memory, such as one or more disk storage devices, flash memory storage devices. In some embodiments, the non-transitory computer-readable storage medium in the memory 602 is used to store at least one instruction, which is used to be executed by the processor 601 to implement the detection method of the target recognition model provided in the method embodiment of the present application.

[0133] In some embodiments, the computer device 600 further includes some other components 603, and the type and quantity of the other components 603 can be selected based on the functional requirements of the computer device 600. Those skilled in the art will appreciate that Figure 6 The structure shown in the figure does not constitute a limitation on the computer device 600, and the computer device 600 may include more or less components than those shown in the figure, or combine some components, or adopt a different arrangement of components.

[0134] Optionally, the computer readable storage medium may include: a read-only memory (ROM), a random access memory (RAM), a solid state drive (SSD), or an optical disk. Among them, the random access memory may include a resistance random access memory (ReRAM) and a dynamic random access memory (DRAM). The serial numbers of the above embodiments of the present application are only for description and do not represent the advantages and disadvantages of the embodiments.

[0135] An embodiment of the present application also provides a computer device, which includes a processor and a memory, wherein the memory stores at least one instruction, at least one program, a code set or an instruction set, and the at least one instruction, the at least one program, the code set or the instruction set is loaded and executed by the processor to implement a detection method for a target recognition model as described in any of the above embodiments of the present application.

[0136] An embodiment of the present application also provides a computer-readable storage medium, in which at least one instruction, at least one program, a code set or an instruction set is stored. The at least one instruction, the at least one program, the code set or the instruction set is loaded and executed by a processor to implement a detection method for a target recognition model as described in any of the above embodiments of the present application.

[0137] The embodiment of the present application also provides a computer program product or a computer program, which includes a computer instruction stored in a computer-readable storage medium. The processor of the computer device reads the computer instruction from the computer-readable storage medium, and the processor executes the computer instruction, so that the computer device executes the detection method of the target recognition model described in any of the above embodiments.

[0138] A person skilled in the art will understand that all or part of the steps to implement the above embodiments may be accomplished by hardware or by instructing related hardware through a program, and the program may be stored in a computer-readable storage medium, and the above-mentioned storage medium may be a read-only memory, a disk or an optical disk, etc.

[0139] The above description is only an optional embodiment of the present application and is not intended to limit the present application. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present application shall be included in the protection scope of the present application.

Claims

1. A detection method for a target recognition model, characterized in that: The method comprises: Obtaining a weather scenario configuration file and a working condition configuration file, wherein the weather scenario configuration file is used to generate a simulated weather scenario when the autonomous driving vehicle is driving, and the working condition configuration file is used to generate a working condition scenario when the autonomous driving vehicle is driving, and the working condition configuration file includes an obstacle configuration part and a driving condition configuration part, wherein the obstacle configuration part includes the configuration of obstacles encountered by the autonomous driving vehicle when driving, and the driving condition configuration part includes the working condition of the autonomous driving vehicle when driving, and the working condition includes at least one of vehicle emergency braking, vehicle emergency avoidance, and vehicle traffic congestion; Generate a three-dimensional simulated driving environment based on the weather scenario configuration file and the operating condition configuration file, wherein the three-dimensional simulated driving environment includes three-dimensional models of the autonomous driving vehicle and the obstacles; Collecting sample image data based on the three-dimensional simulated driving environment, wherein the sample image data is an environment image collected from the perspective of the autonomous driving vehicle, and the environment image contains the obstacle; Iteratively attack the sample image data based on a projected gradient descent method to generate first image data, where the first image data is image data obtained by superimposing noise on the sample image data; The first image data is recognized by a target recognition model corresponding to the autonomous driving vehicle to obtain a first recognition result. The first recognition result includes the recognition status of the obstacle in the first image data by the target recognition model. The first recognition result is used to test the target recognition model.

2. The method according to claim 1, characterized in that The generating of the three-dimensional simulated driving environment based on the weather scenario configuration file and the operating condition configuration file comprises: Inputting the weather scenario configuration file into an autonomous driving simulator, and constructing a background portion of the three-dimensional simulated driving environment through the autonomous driving simulator; Inputting the obstacle configuration part in the working condition configuration file into the autonomous driving simulator, and generating a three-dimensional model of the obstacle in the three-dimensional simulated driving environment by the autonomous driving simulator; the obstacle includes at least one of a vehicle and a pedestrian; The driving condition configuration part in the condition configuration file is input into the autonomous driving simulator, and the autonomous driving simulator is used to generate a three-dimensional model of the autonomous driving vehicle in the three-dimensional simulated driving environment.

3. The method according to claim 1, characterized in that The collecting of sample image data based on the three-dimensional simulated driving environment includes: Controlling the autonomous driving vehicle and the three-dimensional model of the obstacle to move in the three-dimensional simulated driving environment through the autonomous driving simulator; The environmental image corresponding to the three-dimensional simulated driving environment is collected based on the perspective of the autonomous driving vehicle to obtain the sample image data, wherein the sample image data includes the distribution and position information of the obstacles in the environmental image.

4. The method according to claim 1, characterized in that: The iterative attack on the sample image data based on the projected gradient descent method to generate the first image data includes: Performing format conversion on the sample image data to obtain first test data that meets preset format requirements; An iterative attack is performed on the first test data based on a projected gradient descent method to generate the first image data, wherein the iterative attack refers to a process of superimposing noise on the first test data, and the number of iterations of the iterative attack is a preset threshold; wherein the data obtained after the i-th attack on the first test data is used as the data used for the i+1-th attack, and i is a positive integer.

5. The method according to any one of claims 1 to 4, characterized in that: Before the iterative attack on the sample image data based on the projected gradient descent method to generate the first image data, the method further includes: A category analysis is performed on the sample image data to determine a category label of the sample image data, where the category label is used to indicate category information of the obstacle in the sample image data.

6. The method according to claim 5, characterized in that After the first image data is identified by the target recognition model corresponding to the autonomous driving vehicle to obtain a first recognition result, the method further includes: When the difference between the first recognition result and the category label reaches a preset difference threshold, determining that the target recognition model does not meet a preset accuracy requirement when recognizing the image data; Acquire a sample image data set, wherein the sample image data set includes a plurality of image data, and the plurality of image data has the same format as the sample image data; Iteratively attacking the sample image data set based on the projected gradient descent method to generate a training image data set; wherein the training image data set includes a plurality of training image data, and the plurality of training image data has the same format as the first image data; The object recognition model is trained based on the training image dataset.

7. A detection device for a target recognition model, characterized in that: The device comprises: an acquisition module, configured to acquire a weather scenario configuration file and a working condition configuration file, wherein the weather scenario configuration file is used to generate a simulated weather scenario when the autonomous driving vehicle is driving, and the working condition configuration file is used to generate a working condition scenario when the autonomous driving vehicle is driving, and the working condition configuration file includes an obstacle configuration part and a driving condition configuration part, wherein the obstacle configuration part includes the configuration of obstacles encountered by the autonomous driving vehicle when driving, and the driving condition configuration part includes the working condition of the autonomous driving vehicle when driving, and the working condition includes at least one of vehicle emergency braking, vehicle emergency avoidance, and vehicle traffic congestion; A generating module, configured to generate a three-dimensional simulated driving environment based on the weather scenario configuration file and the operating condition configuration file, wherein the three-dimensional simulated driving environment includes three-dimensional models of the autonomous driving vehicle and the obstacles; A collection module, used to collect sample image data based on the three-dimensional simulated driving environment, wherein the sample image data is an environment image collected from the perspective of the autonomous driving vehicle, and the environment image contains the obstacle; The generating module is further used to perform an iterative attack on the sample image data based on a projected gradient descent method to generate first image data, where the first image data is image data obtained by superimposing noise on the sample image data; An identification module is used to identify the first image data through a target recognition model corresponding to the autonomous driving vehicle to obtain a first recognition result, wherein the first recognition result includes the recognition status of the obstacle in the first image data by the target recognition model, and the first recognition result is used to test the target recognition model.

8. A computer device, characterized in that: The computer device includes a processor and a memory, wherein the memory stores at least one program, and the at least one program is loaded and executed by the processor to implement the target recognition model detection method as described in any one of claims 1 to 6.

9. A computer-readable storage medium, characterized in that: The storage medium stores at least one program, and the at least one program is loaded and executed by the processor to implement the target recognition model detection method as described in any one of claims 1 to 6.

10. A computer program product, characterized in that It includes a computer program, which, when executed by a processor, implements the detection method of the target recognition model as described in any one of claims 1 to 6.

Citation Information

Patent Citations

  • Adversarial sample defense method and device based on data disturbance

    CN113537463A

  • Automatic driving simulation test scene generation method, system and related device

    CN115187742A

  • Model robustness detection method, related device and storage medium

    CN115471495A

  • Model robustness detection method, related device and storage medium

    CN115588131A

  • Automatic driving simulation scene test method and device, electronic equipment and readable medium

    CN116010289A

Cited By

  • Automatic driving algorithm training and verification method based on virtual simulation scene

    CN119206404A

  • A method for training and validating an autonomous driving algorithm based on a virtual simulation scenario

    CN119206404B

  • Anti-intelligent camouflage simulation test scene generation method and system and medium

    CN119339000A

  • Adversarial sample generation method based on automatic driving multi-sensor fusion

    CN119723251A

  • Adversarial Sample Generation Method Based on Multi-Sensor Fusion for Autonomous Driving

    CN119723251B