Video injection-to-recharge full-automatic perception algorithm test method and system

Through video injection into the fully automated perception algorithm testing method and system, the problems of low efficiency and high cost in the existing testing methods are solved, automated batch testing and data reuse are realized, and the efficiency of intelligent driving perception algorithm testing is improved. and quality.

CN118890464APending Publication Date: 2024-11-01CHANGZHOU XINGYU AUTOMOTIVE LIGHTING SYST CO LTD
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202411088825.3
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2024-08-09
Publication Date
2024-11-01

AI Technical Summary

Technical Problem

The existing video injection and data re-injection testing methods have problems such as low efficiency, cumbersome manual operations, high cost, inability to test multiple people simultaneously and low data reuse rate, especially in the testing of intelligent driving perception algorithms.

Method used

Using fully automated perception algorithm testing methods and systems for video injection to refille, we realize automated batch video injection and data refille through cloud servers and resource pools, automatically convert pack files and video files, and combine data sources of video injection and refille , realize multi-line testing and efficient data reuse.

Benefits of technology

It improves testing efficiency and quality, reduces R&D testing costs, realizes multi-line sensing algorithm testing, reduces testing time and manpower, supports multiple people to test simultaneously, and enhances the data utilization rate.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN118890464A_ABST
    Figure CN118890464A_ABST
Patent Text Reader

Abstract

The invention provides a full-automatic perception algorithm test method and system from video injection to recharge, and the method comprises the steps: enabling a video injection system to execute an automatic batch video injection program instruction, obtaining selected scene data video stream files from a cloud server, and injecting the scene data video stream files into a perception algorithm board end system in batches, the generated pack video data and the identified perception log data are stored in a resource pool; the automatic recharging system executes an automatic data recharging program instruction, acquires specified pack video data from a resource pool to perform batch recharging on at least one sensing algorithm board end system, and stores recognized sensing log data in the resource pool; and comparing the recognized perception log data with the original data to complete the full-automatic perception algorithm test from video injection to recharge, so that the technical problem that the existing test method is low in test efficiency and the like can be effectively solved.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present invention relates to the field of testing technology, and in particular to a fully automated perception algorithm testing method and system for video injection to re-injection. Background Art

[0002] With the development and updating of intelligent driving technology, more and more sensors are used. The method of testing intelligent driving products entirely by actual road testing has been gradually abandoned by intelligent driving product companies due to defects such as long cycle and high cost. Bench testing combined with real road testing has become a commonly used testing method in the industry.

[0003] In bench testing, the test methods commonly used are data re-injection or video injection. Data re-injection is usually an offline mode in which the real scene data collected by the actual vehicle is played back and injected into the intelligent driving board by reading files. It is used to test the image processing capabilities of the intelligent driving board; video injection is an online mode that synchronously injects images and vehicle body signals into the intelligent driving board based on road scene or simulation scene data. Image injection simulates camera input, and vehicle body signal injection simulates the input of vehicle body sensor signals.

[0004] The video injection method tests a longer board-side link, and because it can use a simulated scene as the input source, it is more efficient for corner cases (special scenes) in actual road tests. However, it has the following defects:

[0005] 1) During video injection, the FPGA board parses the video file to form frame-by-frame image data, which enters the software processing link through the video acquisition path of the intelligent driving board; CAN information is injected into the simulated vehicle body signal through another board. During injection, manual operation is generally required to select a video file and a vehicle body signal file, and enter commands to achieve synchronous injection. After the injection is completed, the corresponding command is executed to continue the injection of the next scene. This operation method is prone to errors and takes up a lot of time for testers.

[0006] 2) If you only use video injection to test the algorithm, when the algorithm is iterated and the same scene needs to be tested again, you need to inject it again.

[0007] 3) Video injection relies on HIL (hardware-in-the-loop simulation) benches and corresponding software tools, which are relatively expensive and require FPGA boards to simulate sensors. A set of FPGA boards can only be connected to one smart driving board, that is, a HIL bench can only be used by one tester to perform video injection testing at the same time. It is not easy for multiple people to test at the same time, and it is difficult to deploy on a large scale.

[0008] The data re-injection method can make up for the above defects of the video injection method. It can be used by multiple people for simultaneous testing, with a shorter cycle and higher efficiency. However, it also has the following defects:

[0009] 1) When recharging data, you need to manually select a .pack file through the client, replay the data into the smart driving board, save the results of the algorithm output through the log, and then select another .pack file for recharging. This operation method requires manual repeated selection, start recharging, and save the log test efficiency is very low.

[0010] 2) The data re-injection method cannot cover all testing requirements due to the complexity of road scenarios and the difficulty of real vehicle data collection to cover all scenarios. Therefore, it is not sufficient to use only data re-injection to evaluate perception algorithms.

[0011] On this basis, the two test methods are efficient and complementary test methods in the intelligent driving perception algorithm test, but when implemented, the two are usually carried out independently. In addition, data re-injection generally uses information such as sensor data collected by the actual vehicle and the processing results of the intelligent driving board algorithm to record and package them into a storage format ending with pack. The input source of video injection is a separate video file and a vehicle body signal file. The pack file collected by the vehicle body cannot be directly used for video injection, and correspondingly, the video injection file cannot be directly used for data re-injection. Summary of the invention

[0012] In order to overcome the above shortcomings, the present invention provides a fully automated perception algorithm testing method and system from video injection to re-injection, which effectively solves technical problems such as low testing efficiency of existing testing methods.

[0013] The technical solution provided by the present invention is:

[0014] On the one hand, the present invention provides a fully automated perception algorithm testing method for video injection to re-injection, comprising:

[0015] The video injection system executes the automated batch video injection program instructions, obtains the selected scene data video stream files in the cloud server, batch-injects them into the perception algorithm board system, and stores the generated pack video data and recognized perception log data in the resource pool;

[0016] The automated recharging system executes automated data recharging program instructions, obtains designated pack video data in the resource pool, performs batch recharging on at least one perception algorithm board system, and stores the identified perception log data in the resource pool;

[0017] The perceived log data obtained through recognition is compared with the original data to complete the fully automated perception algorithm test from video injection to re-injection.

[0018] On the other hand, the present invention provides a fully automated perception algorithm testing system for video injection to re-injection, comprising:

[0019] Cloud server, used to store scene data video stream files in mp4 format;

[0020] Resource pool, used to store pack video data and perception log data;

[0021] The video injection system is used to execute the automated batch video injection program instructions, obtain the selected scene data video stream files in the cloud server and batch inject them into the perception algorithm board system, and store the generated pack video data and recognized perception log data in the resource pool;

[0022] An automated recharging system, used to execute automated data recharging program instructions, obtain designated pack video data in the resource pool, perform batch recharging on at least one perception algorithm board system, and store the identified perception log data in the resource pool;

[0023] The comparison module is used to compare the perceived log data obtained through recognition with the original data, and complete the fully automated perception algorithm test from video injection to re-injection.

[0024] The video injection to re-injection fully automated perception algorithm testing method and system provided by the present invention can at least bring the following beneficial effects:

[0025] (1) The present invention adopts an automated batch video injection method to achieve automatic and continuous injection of scene data including real scenes and simulated scenes, which greatly improves the test efficiency and quality.

[0026] (2) The present invention realizes the automatic conversion between pack files, video files and vehicle body signal files, and satisfies the data source requirements of video injection and data re-injection, thereby improving the data reuse rate.

[0027] (3) The present invention realizes an automated method for injecting video files and vehicle body signal files in simulation scenes and synchronously recording and storing them as pack files. It solves the problem of being able to continuously convert multiple simulation scenes into real vehicle acquisition pack storage formats in a short time, greatly improving work efficiency and saving human resources.

[0028] (4) The integrated automated testing method for video injection and re-injection coupling in the present invention solves the dependence on hardware equipment and environment through the resource pool. Developers and testers can perform multi-line algorithm requirement tests such as performance, function, and stability of the perception algorithm at the same time, which greatly improves the efficiency and quality of perception algorithm testing, thereby accelerating the development and deployment process of the intelligent driving system.

[0029] (5) The present invention combines an automated testing method for video injection and re-injection, and the advantages of a large amount of video data of different scenes generated by video injection and the low cost and high efficiency of data re-injection, to achieve multi-line perception algorithm testing, reduce R&D testing costs, and improve the efficiency of perception algorithm testing.

[0030] (6) The fully automated intelligent driving perception algorithm testing method from video injection to data re-injection provided by the present invention includes automated batch injection of real scene data collected by a real vehicle and virtual scene data generated by simulation software, automated pack video recording and storage mechanism, automated re-injection, synchronous scene classification storage, etc., which realizes a cross-testing method in which real scene data collected by a real vehicle can be injected into the video and virtual scene data generated by simulation software can be re-injected into the data, greatly improving the data availability and being able to perform multi-line perception algorithm testing without relying on expensive and limited hardware equipment, greatly reducing testing time, manpower and cost, and accelerating large-scale verification and iterative optimization of intelligent driving perception algorithms. BRIEF DESCRIPTION OF THE DRAWINGS

[0031] Figure 1 The overall structural block diagram of the fully automated perception algorithm test from video injection to re-injection in the present invention;

[0032] Figure 2 It is a flow chart of the method for automatic batch video injection and pack data synchronous recording and storage in the present invention;

[0033] Figure 3 It is a flow chart of the automatic recharging and pack scene classification storage method in the present invention;

[0034] Figure 4 A flow chart of the method for automatic conversion of pack-to-mp4 format in the present invention;

[0035] Figure 5 The present invention is a flow chart of the method for converting video stream to pack format data. DETAILED DESCRIPTION

[0036] In order to more clearly illustrate the implementation cases of the present invention or the technical solutions in the prior art, the specific implementation methods of the present invention will be described below with reference to the accompanying drawings. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other accompanying drawings and other implementation methods can be obtained based on these drawings without creative work.

[0037] The first embodiment of the present invention is a fully automated perception algorithm testing method for video injection to re-injection, such as Figure 1 As shown, including:

[0038] The S10 video injection system executes the automated batch video injection program instructions, obtains the selected scene data video stream files in the cloud server and batch-injects them into the perception algorithm board system, and stores the generated pack video data and recognized perception log data in the resource pool;

[0039] S20 The automated recharging system executes automated data recharging program instructions, obtains the specified pack video data in the resource pool, performs batch recharging on at least one perception algorithm board system, and stores the identified perception log data in the resource pool;

[0040] S30 compares the perceived log data obtained through recognition with the original data, and completes the fully automated perception algorithm test from video injection to re-injection.

[0041] The fully automated perception algorithm testing method from video injection to re-injection provided in this embodiment connects the two testing methods of video injection and data re-injection to form an integrated process. A large amount of pack video data and recognized perception log data are obtained through video injection of real scene data and virtual scene data of large mileage. The pack video data is stored in the resource pool through the pack synchronous recording and storage mechanism. The automated data re-injection system can select relevant pack video data for re-injection according to the test requirements, and complete multi-line test items and test work by injecting it into different board ends. The perception log data obtained by video injection back into the perception algorithm and the perception log data obtained by data re-injection into the perception system board end can be used for data analysis and mining. Specifically:

[0042] Step S10: The video injection system executes the automated batch video injection program instructions, obtains the selected scene data video stream files in the cloud server and batch injects them into the perception algorithm board system, and stores the generated pack video data and the recognized perception log data in the resource pool, such as Figure 2 As shown in FIG. 1 , the video injection process of a single scene data video stream file includes:

[0043] S11 obtains the scene data video stream file and the vehicle body signal file collected by the bus.

[0044] Before obtaining the scene data video stream file and the body signal file collected by the bus, the hardware, software and server are connected, and the video injection process involves the function startup and data interaction of the intelligent driving injection bench, the perception algorithm board system and the software tools.

[0045] After the automated batch video injection program is started, the program starts and enters the injection script directory. After that, the scene data video stream file and the bus-collected body signal file (CAN bus data file) in the specified folder on the cloud server are obtained. The data sources include real scene data and virtual scene data. The real scene data is obtained through real vehicle collection, and the virtual scene data is generated by simulation software. When performing video injection, they are all in standard video injection format.

[0046] S12 obtains a video injection command, executes the injection of the scene data video stream file, and updates the state file at the time of executing the video injection command, and configures the state in the state file to a first preset value.

[0047] The scene data video stream file is processed through the video injection command and converted into the video protocol information required by the camera controller image processing unit, and injected into the perception algorithm board system for a series of processing such as perception recognition, post-processing, and analysis, to obtain the online pack video stream and perception log data. Among them, the pack video stream can be viewed through the client.

[0048] When the injection command is executed, the video injection system will send out an injection signal. When the injection start signal is obtained, the state in the injection state file will be set to a first preset value, usually 1.

[0049] Perception log data generally includes image data such as lane lines, dynamic / static obstacles, traffic signs, signal lights, etc.; radar data such as distance and speed, obstacle type, etc.; lidar data such as point cloud data, three-dimensional spatial information, etc.; camera data such as color, lighting, etc.; GPS data such as location information, etc.; IMU data, etc.

[0050] S13: The monitoring program script monitors that the status configuration in the status file is the first preset value, and executes the client recording control instruction.

[0051] The monitoring script is used to capture the start and end signals of video injection in real time.

[0052] S14 determines whether the scene data video stream file has been injected.

[0053] When the program obtains the video injection end signal, it indicates that the video injection is completed.

[0054] If yes, S15 updates the status file, configures the status in the status file to the second preset value, and executes the client stop recording control instruction to obtain and store the pack video data converted from the scene data video stream file.

[0055] When a single scene data video stream file is injected and processed, the state in the state file is updated to a second preset value, usually 0, and the program executes the client stop recording control instruction. The state file generally includes information on the start and end time of video injection.

[0056] It is worth noting that if a single scene data video stream injection process is interrupted or an error occurs, resulting in incomplete injection, the injection program will terminate, restart the execution of the video injection program instructions, and perform video injection from the current incomplete scene data video stream.

[0057] In this process, the purpose of executing the client recording instruction in step S13 is to store video data in pack format (corresponding to pack video data), and the client stops recording in step S15 to obtain the pack video data generated by the recording.

[0058] S16 downloads the perception log data recognized by the perception algorithm board system and stores it in a specified folder.

[0059] When the status in the status file is updated to a second preset value, usually 0, the perception log data is packaged and saved separately, and the program executes to save the perception log data corresponding to each video in the scene data video stream to a specified folder. At this point, the injection, recording and storage of perception recognition data of a scene data video stream are completed.

[0060] Steps S11 to S16 are the injection process of a single scene data video stream. The program executes steps S11 to S16 in a loop according to the command until all video data in the folder are injected and processed, that is, the automatic batch injection of videos is completed. And through automatic batch video injection and synchronous pack recording, a large amount of pack video data of different scenes and working conditions and perception log data files stored in packets are stored in the resource pool.

[0061] Step S20: The automatic recharging system executes the automatic data recharging program instruction, obtains the specified pack video data in the resource pool, performs batch recharging on at least one intelligent driving board, and stores the recognized perception log data in the resource pool. Figure 3 As shown in the figure, the process of recharging a single pack of video data includes:

[0062] S21 obtains pack video data.

[0063] Before obtaining the pack video data, the connection between hardware and software devices must be completed, which mainly involves the connection of the perception algorithm board system hardware devices and the implementation of the remote terminal command line.

[0064] After starting the automated data re-injection program, first obtain the pack video data in the specified folder and open it using the client. When the pack video screen appears, the program executes the control command to start the video re-injection button.

[0065] The S22 client opens and completes the parameter configuration for refilling.

[0066] The program automatically completes the perception algorithm parameter configuration of the relevant version, generally including network port configuration, network model re-injection parameter configuration, re-injection segment duration configuration, path saving parameter configuration, etc.

[0067] S23 executes the start re-injection control command and packs the video data back into the perception algorithm board system for perception and recognition.

[0068] Click the Start Video Replay button to start the replay operation and replay the data to the perception algorithm board system for perception and recognition.

[0069] S24 determines whether the single pack video data is refilled.

[0070] When the video end signal is recognized and the video replay button is pressed to pause, the program will execute the control command to exit the replay mode.

[0071] S25 executes the pack video data scene classification and storage program, and stores the pack video data and the recognized perception log data by naming the classified scenes.

[0072] During the refill process, the perception log data will be generated at the same time, and the pack video data scene classification and storage program will be executed. At the same time, the log perception data and pack video data will be named and stored according to the classification scene. At this point, the refill of a pack video data is completed. Figure 1 As shown, n board ends (including board end 1 recharge, board end 2 recharge, etc.) can be recharged at the same time. Pack video scene classification is generally divided according to development and test requirements, and can be divided according to different scene conditions, such as day and night, sunny, rainy, snowy, foggy, etc., and no specific restrictions are made here.

[0073] When executing the pack video data scene classification and saving program, the pack video data and the recognized perception log data are stored in the classified scene names, further comprising:

[0074] S251 executes the control instruction to intercept the header file information picture in the pack video data, identify the specified position in the picture and output the key scene data.

[0075] The program executes control instructions to intercept the header file information image in the pack video data, and performs preprocessing such as data enhancement on the image to more accurately identify the relevant text or characters in the image, and then identifies and outputs the key scene data in the header file according to the specified position in the image.

[0076] S252 converts the key scene data into an editable text string and performs unified formatting processing.

[0077] The key scene data in the header file is converted into an editable text string and formatted uniformly. The processed text naming can clearly and accurately describe the key scenes of the pack video.

[0078] S253 replaces the file name of the pack video data and the file name of the recognized perception log data with a text string containing key scene data and stores them in a designated folder to complete the scene classification and preservation.

[0079] Replace the pack video data file name and the obtained perception log data file name with the text string name containing the key scene and store them in the specified folder to complete the scene classification and preservation.

[0080] After completing the above-mentioned fully automated perception algorithm test from video injection to re-injection, the perception log data obtained by re-injection and the perception log data obtained by previous video injection will be used as data comparison files for algorithm iterative optimization to verify the functions and performance of the intelligent driving perception algorithm system after optimization and iteration.

[0081] Since the input of the video injection system is a video stream file and a bus data file, the output is a pack format file and perception log data, and the standard data format for video injection is a video stream format file. The real scene data generated by the vehicle body used for video injection is obtained by real vehicle data collection and is a pack format file; the virtual scene data is generated by simulation software and is in mp4 format. Therefore, in other embodiments, in order to realize the automatic injection of batch videos, before the video injection, it also includes a step of automatically converting the data file in pack format into a data file in mp4 format, which is used to automatically convert the data file in pack format collected by the real vehicle into a standard injection file in video stream format. It is then stored in the cloud server, and the corresponding data can be read from it according to demand during video injection.

[0082] The pack-to-mp4 format automatic conversion process mainly involves parsing the pack format data and converting it into video frames, and then encoding these video frames into mp4 format video streams. Based on open source technologies such as FFmpeg and Horizon Technologies, corresponding improvements and perfection are made according to development and testing requirements. Figure 4 As shown, including:

[0083] S41 obtains the data file in pack format and creates a new mp4 file;

[0084] S42 creates a decoder context and sets decoder context parameters;

[0085] Decoder context refers to creating and configuring a decoding environment for the decoder to correctly process and decode data. It contains all configuration information about how the decoder processes data. Decoder context parameters include settings such as video width, video height, pixel format, sampling frequency, channel layout, etc.

[0086] S43 cyclically reads each frame in the data file in pack format and performs operations such as decoding and display;

[0087] S44 parses and processes the decoded data, and encodes the valid frame data into a data frame in mp4 format;

[0088] S45 formats the encoded data frame in mp4 format and writes it into the mp4 file;

[0089] S46 determines whether the end of the data file has been reached;

[0090] If yes, S47 completes the automatic conversion of the data file in pack format to the data file in mp4 format, and stores the generated data file in mp4 format in the cloud server as a scene data video stream file.

[0091] In the above process, in addition to converting the data file in the pack format into the data file in the mp4 format, it also includes the conversion of the data files in the pack format compatible with different sampling frequencies, supports the output of the mp4 files with different sampling frequencies, and can realize the automatic batch processing of data files.

[0092] In addition to automatically converting data files in pack format to data files in mp4 format, since the input of the data reinjection system is a pack file and the output is perception log data, and the standard data input format of data reinjection is a pack file. Therefore, before conducting a fully automated perception algorithm test from video injection to reinjection, the video stream needs to be converted to pack format data. Store it in the resource pool. Specifically, the video stream file is converted to pack format data through the intelligent driving injection test bench, perception algorithm board system and client software, such as Figure 5 As shown, including:

[0093] S51 obtains a data file in mp4 format from a cloud server;

[0094] The S52 remote terminal switches to the injection directory and executes the video injection command line;

[0095] S53 opens the client software, sets the pack video storage path and clicks the record button;

[0096] S54 determines whether the video injection is completed;

[0097] S55If yes, click the Stop Recording button;

[0098] S56 downloads the recognized perception log data for storage, completes the pack format data conversion of an mp4 format data file, and obtains the pack video data and stores it in the resource pool.

[0099] It should be understood that the above-mentioned steps of automatic conversion between pack files and video files and vehicle body signal files are completed in advance and then stored in a cloud server or resource pool. In subsequent automated tests, relevant data can be screened and read according to different test requirements.

[0100] Corresponding to the above-mentioned video injection to re-injection fully automated perception algorithm testing method, the present invention also provides a video injection to re-injection fully automated perception algorithm testing system, comprising:

[0101] Cloud server, used to store scene data video stream files in mp4 format;

[0102] Resource pool, used to store pack video data and perception log data;

[0103] The video injection system is used to execute the automated batch video injection program instructions, obtain the selected scene data video stream files in the cloud server and batch inject them into the perception algorithm board system, and store the generated pack video data and recognized perception log data in the resource pool;

[0104] An automated refill system, used to execute automated data refill program instructions, obtain designated pack video data in a resource pool, batch refill at least one perception algorithm board system, and store the identified perception log data in the resource pool;

[0105] The comparison module is used to compare the perceived log data obtained through recognition with the original data, and complete the fully automated perception algorithm test from video injection to re-injection.

[0106] The video injection and re-injection fully automated perception algorithm test system provided in this embodiment connects the two test methods of video injection and data re-injection to form an integrated process, and obtains a large amount of pack video data and recognized perception log data through video injection of real scene data and virtual scene data of large mileage. The pack video data is stored in the resource pool through the pack synchronous recording and storage mechanism. The automated data re-injection system can select relevant pack video data for re-injection according to the test requirements, and complete multi-line test items and test work by injecting into different board ends. The perception log data obtained by video injection back into the perception algorithm and the perception log data obtained by data re-injection into the perception system board end can be analyzed and mined.

[0107] The video injection system includes: a first data acquisition module, which is used to obtain the scene data video stream file and the vehicle body signal file collected by the bus; a video injection module, which is used to obtain the video injection command, execute the injection of the scene data video stream file, and update the status file at the time of the video injection command execution, and configure the status in the status file to the first preset value; a monitoring module, if the status in the status file is configured as the first preset value, executes the client recording control instruction; a video injection judgment module, which is used to judge whether the injection of the scene data video stream file is completed; a first data storage module, which is used to update the status file when the video injection judgment module judges that the injection is completed, configure the status in the status file to the second preset value, and execute the client stop recording control instruction, obtain the pack video data converted from the scene data video stream file and store it; and is used to download the perception log data recognized by the perception algorithm board system and store it in a specified folder.

[0108] The automated re-injection system includes: a first data acquisition module for acquiring pack video data; a parameter configuration module for completing parameter configuration during re-injection; a data re-injection module for executing a start re-injection control command, and re-injecting the pack video data into the perception algorithm board system for perception and recognition; a data re-injection judgment module for judging whether the re-injection of a single pack video data is completed; and a second data storage module for executing a pack video data scene classification and saving program, and storing the pack video data and the recognized perception log data in the classified scene name.

[0109] In this embodiment, the process of the video injection system and the automatic re-injection system to perform the perception algorithm test of the video injection to the re-injection is the same as the aforementioned perception algorithm test method of the re-injection, which will not be repeated here. And correspondingly, before the video injection to the re-injection fully automated perception algorithm test is performed, it also includes the automatic conversion process between the pack file and the video file and the vehicle body signal file, wherein the pack-to-mp4 format automatic conversion is used to automatically convert the pack format data collected by the actual vehicle into a standard injection file in the video stream format, so that the video injection operation can be performed; the method of converting the video stream to the pack format data is used to convert the video stream format file into the pack format data and save it, so that the data re-injection operation can be performed. The specific process is also the same as the aforementioned perception algorithm test method of the re-injection, which will not be repeated here.

[0110] It should be noted that the above embodiments can be freely combined as needed. The above are only preferred embodiments of the present invention. It should be pointed out that for ordinary technicians in this technical field, several improvements and modifications can be made without departing from the principle of the present invention, and these improvements and modifications should also be regarded as the protection scope of the present invention.

Claims

1. A fully automated perception algorithm testing method for video injection to re-injection, characterized in that: include: The video injection system executes the automated batch video injection program instructions, obtains the selected scene data video stream files in the cloud server, batch-injects them into the perception algorithm board system, and stores the generated pack video data and recognized perception log data in the resource pool; The automated recharging system executes automated data recharging program instructions, obtains designated pack video data in the resource pool, performs batch recharging on at least one perception algorithm board system, and stores the identified perception log data in the resource pool; The perceived log data obtained through recognition is compared with the original data to complete the fully automated perception algorithm test from video injection to re-injection.

2. The video injection to re-injection fully automated perception algorithm testing method as claimed in claim 1, characterized in that: The video injection system executes the automatic batch video injection program instructions, obtains the selected scene data video stream files in the cloud server and injects them into the perception algorithm board system in batches, and stores the generated pack video data and the recognized perception log data in the resource pool. The video injection process of a single scene data video stream file includes: Obtain scene data video stream files and vehicle body signal files collected by the bus; Obtain a video injection command, execute injection of a scene data video stream file, and at the moment when the video injection command is executed, update a state file, and configure the state in the state file to a first preset value; The monitoring program script monitors that the state configuration in the state file is a first preset value, and executes the client recording control instruction; Determine whether the scene data video stream file has been injected; If yes, update the state file, configure the state in the state file to the second preset value, execute the client stop recording control instruction, obtain the pack video data converted from the scene data video stream file and store it; Download the perception log data recognized by the perception algorithm board system and store it in the specified folder.

3. The video injection to re-injection fully automated perception algorithm testing method as claimed in claim 2, characterized in that: The method obtains the video injection command, executes the injection of the scene data video stream file, and updates the status file at the time of execution of the video injection command, and configures the status in the status file to the first preset value, including: processing the scene data video stream file into the video protocol information required by the camera controller image processing unit, and injecting it into the perception algorithm board end system together with the vehicle body signal file for processing to obtain pack video data and perception log data.

4. The video injection to re-injection fully automated perception algorithm testing method as claimed in claim 1, characterized in that: The automatic recharging system executes the automatic data recharging program instruction, obtains the specified pack video data in the resource pool, performs batch recharging on at least one intelligent driving board, and stores the recognized perception log data in the resource pool. The recharging process of a single pack video data includes: Get pack video data; The client opens and completes parameter configuration when refilling; Execute the start re-injection control command and pack the video data back into the perception algorithm board system for perception and recognition; Determine whether the re-injection of a single pack video data is completed; Execute the pack video data scene classification and saving program to store the pack video data and the recognized perception log data by naming them according to the classification scenes.

5. The video injection to re-injection fully automated perception algorithm testing method as claimed in claim 1, characterized in that: The execution of the pack video data scene classification and storage program stores the pack video data and the recognized perception log data in the classified scene names, including: Execute control instructions to intercept the header file information picture in the pack video data, identify the specified position in the picture and output key scene data; Convert the key scene data into an editable text string and perform unified formatting processing; The file name of the pack video data and the file name of the recognized perception log data are replaced with the text string containing the key scene data and stored in a designated folder to complete the scene classification and preservation.

6. The method for testing a perception algorithm from video injection to re-injection fully automated as claimed in any one of claims 1 to 5, characterized in that: The video injection system executes the automatic batch video injection program instructions, obtains the selected scene data video stream files in the cloud server and batch injects them into the perception algorithm board end system, and before storing the generated pack video data and the recognized perception log data in the resource pool, it also includes the step of automatically converting the data files in the pack format into the data files in the mp4 format, including: Get the data file in pack format; Create a decoder context and set decoder context parameters; Loop through each frame in the data file in the pack format and perform decoding operations; Parse and process the decoded data, and encode the valid frame data into data frames in mp4 format; The encoded data frame in mp4 format is formatted and written into the mp4 file; Determine whether the end of the data file has been reached; If so, complete the automatic conversion of the data file in the pack format to the data file in the mp4 format, and store the generated data file in the mp4 format in the cloud server as a scene data video stream file.

7. The method for testing a perception algorithm from video injection to re-injection fully automated as claimed in any one of claims 1 to 5, characterized in that: The video injection system executes the automatic batch video injection program instructions, obtains the selected scene data video stream files in the cloud server and batch injects them into the perception algorithm board system, and before storing the generated pack video data and the recognized perception log data in the resource pool, it also includes the step of automatically converting the data files in mp4 format into data files in pack format, including: Get the data file in mp4 format; The remote terminal switches to the injection directory and executes the video injection command line; Open the client software and click the record button; Determine whether video injection is completed; If yes, click the Stop Recording button; Download the recognized perception log data for storage, complete the pack format data conversion of an mp4 format data file, and obtain the pack video data and store it in the resource pool.

8. A fully automated video injection and re-injection perception algorithm testing system, characterized in that: include: Cloud server, used to store scene data video stream files in mp4 format; Resource pool, used to store pack video data and perception log data; The video injection system is used to execute the automated batch video injection program instructions, obtain the selected scene data video stream files in the cloud server and batch inject them into the perception algorithm board system, and store the generated pack video data and recognized perception log data in the resource pool; An automated recharging system, used to execute automated data recharging program instructions, obtain designated pack video data in the resource pool, perform batch recharging on at least one perception algorithm board system, and store the identified perception log data in the resource pool; The comparison module is used to compare the perceived log data obtained through recognition with the original data, and complete the fully automated perception algorithm test from video injection to re-injection.

9. The perception algorithm testing system according to claim 8, characterized in that: The video injection system comprises: A first data acquisition module, used to acquire scene data video stream files and vehicle body signal files collected by the bus; The video injection module is used to obtain a video injection command, execute the injection of the scene data video stream file, and update the state file at the time of executing the video injection command, and configure the state in the state file to a first preset value; The monitoring module executes the client recording control instruction if the status configuration in the status file is monitored to be a first preset value; The video injection judgment module is used to judge whether the injection of the scene data video stream file is completed; The first data storage module is used to update the status file when the video injection judgment module determines that the injection is completed, configure the status in the status file to the second preset value, and execute the client's stop recording control instruction to obtain the pack video data converted from the scene data video stream file and store it; and is used to download the perception log data recognized by the perception algorithm board system and store it in a designated folder.

10. The perception algorithm testing system according to claim 8, characterized in that: The automatic recharging system comprises: A first data acquisition module, used to acquire pack video data; Parameter configuration module, used to complete the parameter configuration during recharging; The data re-injection module is used to execute the start re-injection control command and pack the video data back into the perception algorithm board system for perception and recognition; The data re-injection judgment module is used to judge whether the re-injection of a single pack video data is completed; The second data storage module is used to execute the pack video data scene classification and preservation program, and store the pack video data and the recognized perception log data by naming the classified scenes.