Visual perception algorithm precision optimization method and system based on simulation scene generation

By generating simulated scene video streams and analyzing perception results, the visual perception algorithm is optimized, and the problem of insufficient simulation rendering distortion and automatic optimization capabilities in the existing technology is solved, and the effect of high-quality simulation scenarios and reducing test costs is achieved.

CN118968126APending Publication Date: 2024-11-15CHANGZHOU XINGYU AUTOMOTIVE LIGHTING SYST CO LTD
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202410959178.2
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2024-07-17
Publication Date
2024-11-15

AI Technical Summary

Technical Problem

The existing visual perception algorithm evaluation technology has severe distortion in the rendering effect of simulated scenes, and the adaptability and optimization recognition performance of the perception algorithm are unable to automatically adjust the actual test cost and low efficiency, making it difficult to achieve repeatability and offline iterative testing.

Method used

By generating simulated scene video streams, visual perception algorithms are used to identify and detect traffic elements, analyze and compare perception results, and generate parameter adjustment signals to optimize the algorithm to achieve automatic adjustment and improve recognition accuracy.

Benefits of technology

Generate more realistic and high-quality simulated traffic scenes, reduce the cost of testing and optimization of visual perception algorithms, and improve the algorithm's adaptability and recognition accuracy.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN118968126A_ABST
    Figure CN118968126A_ABST
Patent Text Reader

Abstract

The invention belongs to the technical field of traffic, and particularly relates to a visual perception algorithm precision optimization method and system based on simulation scene generation, and the method comprises the steps: generating a simulation scene video stream; identifying and detecting traffic elements in the simulation scene video stream through a visual perception algorithm, and outputting a perception result of the traffic elements; analyzing, comparing and correcting the output perception results of the traffic elements to generate corresponding parameter adjustment signals; optimizing a visual perception algorithm through the parameter adjustment signal; by analyzing, comparing and correcting the output perception results of the traffic elements, the corresponding parameter adjustment signals are generated to optimize the visual perception algorithm, the adaptive capacity of the visual perception algorithm can be automatically adjusted, the recognition accuracy of the visual perception algorithm can be automatically improved, the output is finer and more real, and the visual perception algorithm is more accurate. A more vivid high-quality simulation traffic scene can be generated, and the test and optimization cost of a visual perception algorithm is reduced.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present invention belongs to the field of traffic technology, and in particular relates to a method and system for optimizing the accuracy of a visual perception algorithm based on simulation scene generation. Background Art

[0002] After the development of different forward visual perception algorithms used in intelligent driving, the recognition ability of the algorithms needs to be evaluated and optimized. At present, the method of collecting real road traffic information by equipping the road test vehicle with a road traffic information truth value collection system is used to evaluate the forward visual perception algorithm. By comparing the recognition results of the forward visual perception algorithm for each traffic element with the true value results of the traffic information collected by the road test vehicle using the truth value system, the recognition level of the algorithm is evaluated and the recognition ability of the algorithm is improved through corresponding optimization methods.

[0003] The existing visual perception algorithm evaluation technology solutions have the following shortcomings: the rendering effect of the simulation scene constructed by the simulation platform is general, there is a relatively serious distortion, and the simulation platform does not restore the details of the real world (such as the effect of rain hitting the glass) in place; for different working conditions, the existing technical solutions cannot evaluate the performance of the visual perception system in real time, and realize the real-time dynamic automatic adjustment of the adaptability of the perception algorithm; for the same working conditions, the existing technical solutions can only verify the effectiveness of the perception algorithm, and cannot automatically optimize the recognition performance of the algorithm to improve the recognition accuracy of the perception algorithm; for the traditional road actual test method, it places the autonomous driving vehicle in an actual specific environment for a long mileage road test, collects the results required by the perception system, and tests and verifies the system performance. Because the actual road test method requires a specific road condition environment and specific true value collection equipment, the cost of system construction is high, the test efficiency is low, it is difficult to repeat the iterative test offline, and the test repeatability is poor; for the simulation test method, the construction cost of the commonly used system HIL bench test is also high.

[0004] Therefore, there is an urgent need to develop a new method and system for optimizing the accuracy of visual perception algorithms based on simulation scene generation, in order to solve the problems of general and severe distortion of traffic scene rendering effects constructed by existing simulation platforms, the inability of existing technical solutions to automatically adjust the adaptability of perception algorithms and the inability to automatically optimize the recognition ability of perception algorithms, and the high cost of actual testing methods and simulation testing methods based on simulation systems. Summary of the invention

[0005] The purpose of the present invention is to provide a method and system for optimizing the accuracy of a visual perception algorithm based on simulation scene generation.

[0006] In order to solve the above technical problems, the present invention provides a method for optimizing the accuracy of a visual perception algorithm based on simulation scene generation, which includes: generating a simulation scene video stream; identifying and detecting traffic elements in the simulation scene video stream through a visual perception algorithm, and outputting perception results of the traffic elements; analyzing, comparing and correcting the output perception results of the traffic elements to generate corresponding parameter adjustment signals; and optimizing the visual perception algorithm through the parameter adjustment signal.

[0007] Specifically, the method for generating a simulation scene video stream includes: generating traffic simulation scenes under different working conditions through a simulation scene generation module, and outputting scene information corresponding to each traffic simulation scene; the simulation scene generation module generates a corresponding simulation scene video stream according to each scene information.

[0008] Specifically, the method of generating traffic simulation scenarios under different working conditions through a simulation scenario generation module and outputting scenario information corresponding to each traffic simulation scenario includes: receiving a customized scenario description to construct a corresponding traffic simulation scenario.

[0009] Specifically, the method of generating traffic simulation scenarios under different working conditions by means of a simulation scenario generation module and outputting scenario information corresponding to each traffic simulation scenario includes: randomly constructing traffic simulation scenarios.

[0010] Specifically, the method for the simulation scene generation module to generate the corresponding simulation scene video stream according to each scene information includes: setting parameters of each traffic element in the traffic simulation scene, and combining, rendering and interacting each traffic element to form a simulation scene video stream.

[0011] Specifically, the method for generating a simulation scene video stream also includes: outputting the simulation scene video stream online, or saving the simulation scene video stream offline to a local device.

[0012] Specifically, the method of identifying and detecting traffic elements in a simulation scene video stream through a visual perception algorithm and outputting perception results of the traffic elements includes: receiving a simulation scene video stream output by a simulation scene generation module through a visual perception algorithm module, and identifying and detecting traffic elements in the simulation scene video stream; the visual perception algorithm module outputs the perception results of the traffic elements in an SPI manner.

[0013] Specifically, the method for analyzing, comparing and correcting the output perception results of traffic elements to generate corresponding parameter adjustment signals includes: receiving the perception results of traffic elements output by the visual perception algorithm module through the SPI analysis module, and performing SPI analysis on them; receiving the analysis data output by the SPI analysis module through the SPI analysis perception result data processing module, and classifying them; receiving the scene truth data output by the simulation scene generation module through the scene truth data processing module, and processing the scene truth data; comparing the classified scene truth data with the SPI analysis perception result data through the automatic data comparison module, and outputting the comparison results to the evaluation module; marking the data that does not meet the standards in the comparison results through the evaluation module, so that the correction and optimization module generates the corresponding parameter adjustment signal according to the marking results of the evaluation module.

[0014] Specifically, the frame information analyzed by SPI is paired with the corresponding simulation frame information in time synchronization; after the corresponding frame synchronization is completed, the scene true value target in the synchronized frame and the board-side perception target analyzed by SPI are matched; after the target matching is completed, the parameter values ​​of the two successfully matched targets are compared, and the corresponding degree of proximity is calculated.

[0015] On the other hand, the present invention provides a visual perception algorithm accuracy optimization system, which includes: a simulation scene generation module, which generates a simulation scene video stream; a visual perception algorithm module, which receives the simulation scene video stream output by the simulation scene generation module, identifies and detects traffic elements in the simulation scene video stream, and outputs the perception results of the traffic elements in SPI mode; an SPI analysis module, which receives the perception results of the traffic elements output by the visual perception algorithm module, and performs SPI analysis on them; an SPI analysis perception result data processing module, which receives the analysis data output by the SPI analysis module and classifies them; a scene truth value data processing module, which receives the scene truth value data output by the simulation scene generation module and processes the scene truth value data; an automatic data comparison module, which compares the scene truth value data after classification with the SPI analysis perception result data, and outputs the comparison results to an evaluation module; an evaluation module, which marks the data that does not meet the standards in the comparison results, so that the correction optimization module generates a corresponding parameter adjustment signal according to the marking result of the evaluation module, and outputs the parameter adjustment signal to the visual perception algorithm module.

[0016] The beneficial effect of the present invention is that the present invention analyzes, compares and corrects the perception results of the output traffic elements, generates corresponding parameter adjustment signals to optimize the visual perception algorithm, can automatically adjust the adaptability of the visual perception algorithm and automatically improve the recognition accuracy of the visual perception algorithm, so that the output is more delicate and more realistic, can generate more realistic high-quality simulated traffic scenes, and reduce the testing and optimization costs of the visual perception algorithm.

[0017] Other features and advantages of the present invention will be set forth in the description which follows, and in part will be apparent from the description, or may be learned by practicing the present invention.

[0018] In order to make the above-mentioned objects, features and advantages of the present invention more obvious and easy to understand, preferred embodiments are given below and described in detail with reference to the accompanying drawings. BRIEF DESCRIPTION OF THE DRAWINGS

[0019] In order to more clearly illustrate the specific implementation methods of the present invention or the technical solutions in the prior art, the drawings required for use in the specific implementation methods or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are some implementation methods of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0020] Figure 1 It is a workflow diagram of the method for optimizing the accuracy of a visual perception algorithm based on simulation scene generation of the present invention;

[0021] Figure 2 It is a principle block diagram of the method for optimizing the accuracy of a visual perception algorithm based on simulation scene generation of the present invention;

[0022] Figure 3 It is a principle block diagram of the data automatic comparison module of the present invention;

[0023] Figure 4 It is a principle block diagram of the evaluation module of the present invention;

[0024] Figure 5 It is a principle block diagram of the correction and optimization module of the present invention. DETAILED DESCRIPTION

[0025] In order to make the purpose, technical solution and advantages of the embodiments of the present invention clearer, the technical solution of the present invention will be clearly and completely described below in conjunction with the accompanying drawings. Obviously, the described embodiments are 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 are within the scope of protection of the present invention.

[0026] In some embodiments, Figures 1 to 5 As shown, at least one embodiment provides a method for optimizing the accuracy of a visual perception algorithm generated based on a simulation scene, which includes: generating a simulation scene video stream; identifying and detecting traffic elements in the simulation scene video stream through a visual perception algorithm, and outputting perception results of the traffic elements; analyzing, comparing and correcting the output perception results of the traffic elements to generate corresponding parameter adjustment signals; and optimizing the visual perception algorithm through the parameter adjustment signal.

[0027] In some embodiments, by analyzing, comparing and correcting the perception results of the output traffic elements, corresponding parameter adjustment signals are generated to optimize the visual perception algorithm. This can automatically adjust the adaptability of the visual perception algorithm and automatically improve the recognition accuracy of the visual perception algorithm, making the output more delicate and realistic, and being able to generate more realistic high-quality simulated traffic scenes, thereby reducing the testing and optimization costs of the visual perception algorithm.

[0028] During the working process, traffic simulation scenes under different working conditions will be automatically generated in batches, and scene information corresponding to each traffic scene will be output. This information will be used as the true value data of the entire system. The generated simulation scene video stream will be used as the perception object of the visual perception algorithm. The visual perception algorithm receives the injection of the simulation video stream, and recognizes and detects the traffic elements in the simulation video in real time, and outputs the results of perceiving the traffic elements in SPI mode. At the same time, it receives the correction signal and the correction operation, adjusts and corrects the parameters of the modules that need to be corrected to improve the algorithm's recognition accuracy of the scene elements. Through the recognition results from the visual perception algorithm, SPI analysis and classification processing are performed on them. For different traffic elements, the relevant information of the elements is classified and counted. Vehicle information is classified into vehicle category, pedestrian information is divided into pedestrian category, traffic sign information is divided into sign category, etc., and the type, number, location coordinates, distance from the vehicle, etc. corresponding to each type of element are counted. Based on the classified scene true value data and SPI analysis perception result data, an automatic comparison is performed, and the comparison results are evaluated for recognition effectiveness and accuracy, so that the substandard data can be marked, and the corresponding recognition optimization of the visual perception algorithm is performed based on the marking results, so as to achieve automatic optimization of the recognition ability of the entire system.

[0029] In some embodiments, the method for generating a simulation scene video stream includes: generating traffic simulation scenes under different working conditions through a simulation scene generation module, and outputting scene information corresponding to each traffic simulation scene; the simulation scene generation module generates a corresponding simulation scene video stream according to each scene information.

[0030] Specifically, the simulation scenario generation module is based on the relevant parameters that can be set by the system (weather, vehicles, roads, traffic lights, signs, obstacles, VRU, etc.), freely and randomly combines different parameters, and automatically generates various types of traffic scenarios in batches.

[0031] In some embodiments, the method of generating traffic simulation scenarios under different working conditions through a simulation scenario generation module and outputting scenario information corresponding to each traffic simulation scenario includes: receiving a customized scenario description to construct a corresponding traffic simulation scenario.

[0032] Specifically, the scene description reading module: This module is used to receive user-defined scene descriptions and quickly build the traffic scene the user wants. The user's description of the traffic simulation scene he wants to build is divided into a specific description and a simple description. For example, the specific description can be: "At night, it is raining heavily, there are many vehicles on the city road, the speed is slow, and there are many pedestrians walking with umbrellas on the roadside. The car drove 300m in the congested section. After driving out of the congested section, the car continued to drive on the city section, and drove a total of 20km." The simple description corresponding to the specific description describes the scene through a combination of adjectives and nouns, that is, "night, heavy rain, city roads, many vehicles, slow speed, many pedestrians with umbrellas, 300m congested section, total mileage 20km".

[0033] In some embodiments, the method of generating traffic simulation scenarios under different working conditions through a simulation scenario generation module and outputting scenario information corresponding to each traffic simulation scenario includes: randomly constructing traffic simulation scenarios.

[0034] Specifically, traffic scene random generation module: When the user does not input any scene construction information, the system automatically randomly generates a standard simple description of the traffic scene based on the traffic scene random generation module, and sends the traffic element signal corresponding to the standard simple description to the traffic element module.

[0035] In some embodiments, the method in which the simulation scene generation module generates a corresponding simulation scene video stream according to each scene information includes: setting parameters of each traffic element in the traffic simulation scene, and combining, rendering and interacting each traffic element to form a simulation scene video stream.

[0036] Specifically, the scene description information processing module: This module receives input information from the scene description reading module and performs data processing on the description information. For specific descriptions, the system extracts adjectives and nouns from the specific descriptions, converts them into simple descriptions, and then applies them; for simple descriptions, the system standardizes non-standard simple descriptions (using the same method as converting specific descriptions into simple descriptions), and matches the standard simple descriptions converted from specific descriptions and non-standard simple descriptions with the characteristic signals corresponding to each traffic element signal and the traffic element, outputs the traffic element signal to the traffic element module, and outputs the characteristic signals corresponding to the traffic element to the traffic element parameter setting module.

[0037] Specifically, the traffic element module: This module contains all parameter information of various traffic element modules such as the weather module, obstacle module, road module, etc. It receives signals from the scene description information processing module, and calls the target elements corresponding to the signals from each traffic element module. For example, based on the simple description of the above standard, the system randomly calls rainy scenes from the weather library of the weather module, randomly calls urban roads from the database of the road module, randomly calls multiple vehicles of various types from the database of the vehicle module, and calls various types of pedestrians with umbrellas from the database of the pedestrian module.

[0038] Specifically, the traffic element parameter setting module: This module receives the traffic element characteristic signal input by the scene description information processing module, and sets the corresponding characteristic parameters of the target elements selected by the traffic element module. For the traffic scene described by the above scene description information processing module, the light intensity is set to a lower value, the rainfall is set to a larger value, the road friction coefficient is set to a smaller value, the vehicle density on a random 300m long section is set to a large value, the speed of each vehicle is set to a lower value, and the vehicle density on the remaining sections is randomly set to a normal value. Based on the pedestrian dynamics parameter module, the movement speed of each pedestrian is set to a random value. The system sets the path of the vehicle in the path planning module to travel according to the planned route, and travels a total of 20km in the constructed urban road scene. The paths of the remaining vehicles are random.

[0039] Specifically, the traffic element combination module: This module is used to reasonably combine the various traffic elements with set characteristic parameters so that the operation of each traffic element conforms to the real logic, such as vehicles driving on the road, pedestrians walking on the road, rain or snowflakes falling from the sky to the ground at a certain angle, etc.

[0040] Specifically, the scene rendering module: after forming a complete traffic scene through the traffic element combination module, the scene rendering module renders the constructed traffic scene. It achieves fine processing of traffic scene details and fast rendering effect presentation by setting features such as lighting, materials, and textures. At the same time, it automatically adjusts parameters according to the characteristics of different scenes, adapts to the rendering requirements of various scenes, and then enhances and improves the rendering quality of different scenes, outputting more realistic traffic scene images that are more conducive to recognition by the perception system.

[0041] Specifically, traffic element interaction module: This module controls the interaction between various moving traffic elements and the interaction between moving traffic elements and stationary traffic elements based on interaction rules, such as the interaction between vehicles, the interaction between pedestrians, the interaction between vehicles and pedestrians, the movement of vehicles, pedestrians, etc. according to traffic lights, etc.

[0042] In some embodiments, the method for generating a simulation scene video stream further includes: outputting the simulation scene video stream online, or saving the simulation scene video stream offline to a local computer.

[0043] Specifically, output simulation scene video stream: This module is used to output the simulation video obtained by simulation in online or offline modes. When outputting in online mode, the simulation screen data can be transmitted to the visual perception algorithm module in real time. The offline mode can be used to save the generated simulation video as a local file for subsequent testing.

[0044] In some embodiments, the method of identifying and detecting traffic elements in a simulation scene video stream through a visual perception algorithm and outputting the perception results of the traffic elements includes: receiving the simulation scene video stream output by a simulation scene generation module through a visual perception algorithm module, and identifying and detecting traffic elements in the simulation scene video stream; the visual perception algorithm module outputs the perception results of the traffic elements in an SPI manner.

[0045] In some embodiments, the method for analyzing, comparing and correcting the output perception results of traffic elements to generate corresponding parameter adjustment signals includes: receiving the perception results of traffic elements output by the visual perception algorithm module through the SPI analysis module, and performing SPI analysis on them; receiving the analysis data output by the SPI analysis module through the SPI analysis perception result data processing module, and classifying them; receiving the scene truth data output by the simulation scene generation module through the scene truth data processing module, and processing the scene truth data; comparing the classified scene truth data with the SPI analysis perception result data through the automatic data comparison module, and outputting the comparison results to the evaluation module; marking the data that does not meet the standards in the comparison results through the evaluation module, so that the correction and optimization module generates the corresponding parameter adjustment signal according to the marking results of the evaluation module.

[0046] In some embodiments, the frame information analyzed by SPI is paired with the corresponding simulation frame information for time synchronization; after the corresponding frame synchronization is completed, the scene true value target in the synchronized frame and the board-side perception target analyzed by SPI are matched; after the target matching is completed, the parameter values ​​of the two successfully matched targets are compared, and the corresponding degree of proximity is calculated.

[0047] Specifically, the degree of closeness p refers to the number of parameters n for which the SPI analytical data results completely correspond to the simulation true value results (or are within the allowable error range). SPI The total number of parameters compared with n total The ratio of is as follows:

[0048]

[0049] Specifically, the specific process of the evaluation module is as follows: based on the data comparison results of each module generated by the data automation comparison module, the evaluation module analyzes and judges the comparison results; the threshold value of the closeness between the parsed data and the true value data is set to 0.95. When the closeness between the parsed data and the true value data in the comparison results obtained by the data automation comparison module is less than the set threshold, the evaluation module counts and determines which modules have a closeness less than the threshold, and the data of these modules are marked as unqualified through the data unqualified marking module. The marking content includes target type, target number, target location information, target ID, target distance, etc., and the unqualified marking results of the corresponding modules are sent to the correction optimization module for correction. When the closeness between the parsed data and the true value data of all modules is greater than the set threshold, a signal that no optimization is required is sent to the correction optimization module, so that the correction optimization module does not perform any operation.

[0050] Specifically, the correction optimization module: This module receives the signal output by the evaluation module indicating whether the visual perception algorithm module needs to be corrected, and performs algorithm correction on the corresponding module of the visual perception algorithm module. When the visual perception algorithm module performs visual perception on the simulation scene, it cannot achieve 100% recognition accuracy in all recognition processes due to algorithm accuracy issues. If there are phenomena such as incorrect identification of the sign type during the recognition process, failure to identify the obstacle ahead, and incorrect calculation of the distance between the two vehicles, the correction module automatically generates relevant adjustment parameters for module recognition based on the inaccurate identification type, and sends the generated parameter adjustment signal to the visual perception algorithm module.

[0051] Specifically, the types of inaccurate recognition include: misrecognition, missed recognition and repeated recognition.

[0052] Specifically, misidentification: for erroneous identification situations, the error identification correction module corrects the erroneous identification results according to the simulation true value results, and generates a misidentification parameter adjustment signal;

[0053] Specifically, missed identification: in the case of missed identification, the missed identification correction module adds and corrects the missed identification result accordingly according to the simulation true value result, and generates a missed identification parameter adjustment signal;

[0054] Specifically, repeated recognition: in the case of repeated recognition, the repeated recognition correction module deletes the results of repeated recognition according to the simulation true value results, and generates a repeated recognition parameter adjustment signal.

[0055] Specifically, the correction and optimization module first locates the sub-modules (obstacle recognition module, sign recognition module, etc.) corresponding to the perception algorithm module to be adjusted by each parameter adjustment signal, and then makes corresponding parameter adjustments to the sub-modules according to the parameter adjustment signal sent to the visual perception algorithm module; for each type of parameter adjustment signal, it will update the characteristic parameters in the corresponding module of the perception algorithm to achieve the purpose of optimizing the perception of traffic elements by each sub-module, thereby realizing the improvement of the recognition accuracy of the visual perception algorithm being tested.

[0056] In some embodiments, at least one embodiment provides a visual perception algorithm accuracy optimization system, which includes: a simulation scene generation module, which generates a simulation scene video stream; a visual perception algorithm module, which receives the simulation scene video stream output by the simulation scene generation module, and identifies and detects the traffic elements in the simulation scene video stream, and outputs the perception results of the traffic elements in SPI mode; an SPI analysis module, which receives the perception results of the traffic elements output by the visual perception algorithm module, and performs SPI analysis on them; an SPI analysis perception result data processing module, which receives the analysis data output by the SPI analysis module and classifies them; a scene truth value data processing module, which receives the scene truth value data output by the simulation scene generation module and processes the scene truth value data; an automatic data comparison module, which compares the scene truth value data after classification with the SPI analysis perception result data, and outputs the comparison results to an evaluation module; an evaluation module, which marks the data that does not meet the standards in the comparison results, so that the correction optimization module generates a corresponding parameter adjustment signal according to the marking result of the evaluation module, and outputs the parameter adjustment signal to the visual perception algorithm module.

[0057] To summarize, the present invention analyzes, compares and corrects the perception results of the output traffic elements, generates corresponding parameter adjustment signals to optimize the visual perception algorithm, can automatically adjust the adaptability of the visual perception algorithm and automatically improve the recognition accuracy of the visual perception algorithm, making the output more delicate and realistic, and can generate more realistic high-quality simulated traffic scenes, thereby reducing the testing and optimization costs of the visual perception algorithm.

[0058] The various devices selected in this application (components whose specific structures are not described) are all universal standard parts or components known to technical personnel in this field, and their structures and principles can be known to technical personnel through technical manuals or through conventional experimental methods.

[0059] In the description of the embodiments of the present invention, unless otherwise clearly specified and limited, the terms "installed", "connected", and "connected" should be understood in a broad sense, for example, it can be a fixed connection, a detachable connection, or an integral connection; it can be a mechanical connection or an electrical connection; it can be a direct connection or an indirect connection through an intermediate medium, or it can be the internal communication of two components. For ordinary technicians in this field, the specific meanings of the above terms in the present invention can be understood according to specific circumstances.

[0060] In the description of the present invention, it should be noted that the terms "center", "upper", "lower", "left", "right", "vertical", "horizontal", "inner", "outer", etc., indicating the orientation or positional relationship, are based on the orientation or positional relationship shown in the drawings, and are only for the convenience of describing the present invention and simplifying the description, rather than indicating or implying that the device or element referred to must have a specific orientation, be constructed and operated in a specific orientation, and therefore cannot be understood as limiting the present invention. In addition, the terms "first", "second", and "third" are used for descriptive purposes only, and cannot be understood as indicating or implying relative importance.

[0061] In the several embodiments provided in the present application, it should be understood that the disclosed systems, devices and methods can be implemented in other ways. The device embodiments described above are merely schematic. For example, the division of the units is only 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 communication interfaces, and the indirect coupling or communication connection of devices or units can be electrical, mechanical or other forms.

[0062] 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 network units. Some or all of the units may be selected according to actual needs to achieve the purpose of the solution of this embodiment.

[0063] 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.

[0064] Based on the above ideal embodiments of the present invention, the relevant staff can make various changes and modifications without departing from the technical concept of the present invention through the above description. The technical scope of the present invention is not limited to the contents of the specification, and its technical scope must be determined according to the scope of the claims.

Claims

1. A method for optimizing the accuracy of a visual perception algorithm based on simulation scene generation, characterized in that: include: Generate simulation scene video stream; Identify and detect traffic elements in the video stream of the simulation scene through visual perception algorithms, and output the perception results of the traffic elements; Analyze, compare and correct the output traffic element perception results to generate corresponding parameter adjustment signals; The visual perception algorithm is optimized by adjusting the signal parameters.

2. The method for optimizing the accuracy of a visual perception algorithm based on simulation scene generation according to claim 1, characterized in that: The method for generating a simulation scene video stream comprises: Generate traffic simulation scenarios under different working conditions through the simulation scenario generation module, and output the scenario information corresponding to each traffic simulation scenario; The simulation scene generation module generates corresponding simulation scene video streams according to each scene information.

3. The method for optimizing the accuracy of a visual perception algorithm based on simulation scene generation according to claim 2, characterized in that: The method of generating traffic simulation scenarios under different working conditions by means of a simulation scenario generation module and outputting scenario information corresponding to each traffic simulation scenario comprises: Receive customized scenario descriptions to build corresponding traffic simulation scenarios.

4. The method for optimizing the accuracy of a visual perception algorithm based on simulation scene generation according to claim 2, characterized in that: The method of generating traffic simulation scenarios under different working conditions by means of a simulation scenario generation module and outputting scenario information corresponding to each traffic simulation scenario comprises: Randomly construct traffic simulation scenarios.

5. The method for optimizing the accuracy of a visual perception algorithm based on simulation scene generation according to claim 3 or 4, characterized in that: The method in which the simulation scene generation module generates a corresponding simulation scene video stream according to each scene information includes: Set the parameters of each traffic element in the traffic simulation scene, and combine, render and interact the traffic elements to form a simulation scene video stream.

6. The method for optimizing the accuracy of a visual perception algorithm based on simulation scene generation according to claim 2, characterized in that: The method for generating a simulation scene video stream also includes: Output the simulation scene video stream online, or Save the simulation scene video stream offline to local computer.

7. The method for optimizing the accuracy of a visual perception algorithm based on simulation scene generation according to claim 2, characterized in that: The method of identifying and detecting traffic elements in a simulation scene video stream by a visual perception algorithm and outputting perception results of the traffic elements comprises: The simulation scene video stream output by the simulation scene generation module is received through the visual perception algorithm module, and the traffic elements in the simulation scene video stream are identified and detected; The visual perception algorithm module outputs the perception results of traffic elements in SPI mode.

8. The method for optimizing the accuracy of a visual perception algorithm based on simulation scene generation according to claim 7, characterized in that: The method of analyzing, comparing and correcting the outputted perception results of traffic elements to generate corresponding parameter adjustment signals includes: The SPI analysis module receives the perception results of traffic elements output by the visual perception algorithm module and performs SPI analysis on them; The SPI analysis perception result data processing module receives the analysis data output by the SPI analysis module and classifies and processes the analysis data; The scene truth data processing module receives the scene truth data output by the simulation scene generation module and processes the scene truth data; The classified scene truth data and the SPI analysis perception result data are compared through the automatic data comparison module to output the comparison results to the evaluation module; The evaluation module marks the data that does not meet the standards in the comparison results, so that the correction and optimization module generates corresponding parameter adjustment signals according to the marking results of the evaluation module.

9. The method for optimizing the accuracy of a visual perception algorithm based on simulation scene generation according to claim 8, characterized in that: Time-synchronize and pair the frame information analyzed by SPI with the corresponding simulation frame information; After the corresponding frame synchronization is completed, the scene true value target in the synchronized frame is matched with the board-side perception target analyzed by SPI; After the target matching is completed, the parameter values ​​of the two successfully matched targets are compared and the corresponding degree of proximity is calculated.

10. A visual perception algorithm accuracy optimization system, characterized in that: include: A simulation scene generation module generates a simulation scene video stream; The visual perception algorithm module receives the simulation scene video stream output by the simulation scene generation module, identifies and detects the traffic elements in the simulation scene video stream, and outputs the perception results of the traffic elements in the SPI mode; The SPI analysis module receives the perception results of traffic elements output by the visual perception algorithm module and performs SPI analysis on them; The SPI analysis perception result data processing module receives the analysis data output by the SPI analysis module and classifies and processes it; A scene truth data processing module receives the scene truth data output by the simulation scene generation module and processes the scene truth data; The automated data comparison module compares the classified scene truth data with the SPI analysis perception result data, and outputs the comparison results to the evaluation module; The evaluation module marks the data that does not meet the standards in the comparison results, so that the correction and optimization module generates corresponding parameter adjustment signals according to the marking results of the evaluation module, and outputs the parameter adjustment signals to the visual perception algorithm module.