Autonomous driving data processing and tracking method, system and vehicle

By establishing a data storage management cluster and distributed storage system in the processing of autonomous driving data, recording and tracking the information of each data processing link, the repeated calculation problems caused by errors are solved, rapid positioning and repairing are achieved, and computing resources are optimized. use of.

CN112115173BActive Publication Date: 2024-12-06ZHEJIANG GEELY AUTOMOBILE RES INST CO LTD +1
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202010912098.3
Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
Filing Date
2020-09-02
Publication Date
2024-12-06
Estimated Expiration
2040-09-02

AI Technical Summary

Technical Problem

During the process of autonomous driving data processing, repeated calculations are caused by errors in a certain link in the processing chain of massive data, which wastes time and computing resources, and it is difficult to quickly locate and repair errors.

Method used

Establish an autonomous driving data storage management cluster, including a relational database and a distributed storage system, record and track the input and output files and operation relationships of each data processing link, record error information through the database table, and allow continued operation from the error steps.

Benefits of technology

It realizes rapid positioning and repair of each link in the data processing chain, avoids repeated calculations, optimizes the utilization of computing resources, and reduces processing time.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN112115173B_ABST
    Figure CN112115173B_ABST
Patent Text Reader

Abstract

The present invention provides an autonomous driving data processing tracking method, system and vehicle, wherein the autonomous driving data processing tracking method, through the application of a relational database on distributed storage, can effectively locate, classify, track and search each file that appears in the data processing process, thereby effectively tracking each link in the data processing chain, directly locating the erroneous link and continuing to run data processing from the erroneous link, thereby solving the problem of wasting time and computing power in the process of processing massive data.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present invention relates to the field of big data technology, and more specifically, to an autonomous driving data processing and tracking method and system, and a vehicle having the system. Background Art

[0002] In autonomous driving technology, cameras and lidar replace human eyes, while machine learning and neural networks replace human brains. Whether it is testing various simulation scenarios or training the brain of autonomous driving technology - neural networks, massive amounts of data are required, including camera data, lidar data, map data, vehicle bus data, etc. These data include data collected from actual roads, and more are data produced by scene simulation systems, which can reach hundreds of PB levels. These original scene data files are often processed through multiple steps, such as cutting, cleaning, and conversion of various data formats, and finally generate data format files that are valuable to the R&D process. In the entire data processing chain, if there is an error in data processing, the R&D personnel need to find the error in the data processing error log, solve the defects of the data processing program here, and then rerun the entire data processing process until few or no errors occur. This will be a very long process and will result in a large waste of computing resources.

[0003] There are several problems in the current autonomous driving data processing process:

[0004] Autonomous driving R&D requires a series of processing of collected or simulated data files, such as cutting, cleaning, and various format conversions. Various result data files will be generated during the processing. Therefore, how to track these original data files, how to know which operations have been performed on a certain original data file, and which result files have been generated is a problem that must be solved in autonomous driving data process management.

[0005] The amount of data for autonomous driving is often huge, often in the tens or even hundreds of PB. It takes a long time and a lot of computing resources to complete large-scale data processing. Therefore, how to avoid repeated calculations due to errors in one link of the data processing chain is a problem that needs to be solved in massive data processing.

[0006] An autonomous driving data processing chain contains many links, and errors may occur in any link. How to quickly locate and fix errors is also a problem that needs to be solved in the autonomous driving data processing process. Summary of the invention

[0007] One purpose of the first aspect of the present invention is to solve the problem of how to avoid repeated calculations due to errors in a certain link in the data processing chain of massive data, thereby wasting processing time and computing resources.

[0008] A further purpose of the first aspect of the present invention is to track the status and quickly locate each link in the data processing chain of autonomous driving, so as to achieve the purpose of optimizing computing power and computing process.

[0009] In particular, the present invention provides an autonomous driving data processing and tracking method, comprising the following steps:

[0010] S1. Establish an autonomous driving data storage management cluster;

[0011] S2. Adding a distributed storage system to the storage management cluster;

[0012] S3, adding a relational database to the storage management cluster;

[0013] S4, road data collection: record the raw data collected by various sensors, roof boxes and bus data on the test vehicle, add data tags and metadata tags, write them into data files, and store the data files in SSD flash memory;

[0014] S5, data upload: uploading the data in the SSD flash memory to the distributed storage system;

[0015] S6. Run the data text import script: write the file path, file name and corresponding tag data of the original data in the distributed storage system into the relational database;

[0016] S7, running the tag data and metadata script: writing the mapping relationship between the tag data and the original data file location and the mapping relationship between the metadata device tag and the original data file location into the relational database;

[0017] S8, run the file processing script: process the scene file, write the original file, operation name and processed file mapping relationship into the relational data, write the name of the error node and the input file into the relational database for storage;

[0018] S9. Develop data storage management applications and front-end UI.

[0019] Furthermore, the storage management cluster in step S1 includes at least three servers, and the at least three servers include:

[0020] A relational database server, on which a MySQL relational database is deployed, is used to store location information, metadata information, label data information, and various mapping relationship information of data files;

[0021] Data management server, on which data storage management applications are deployed to classify, store, accurately locate and continuously track data;

[0022] Data application server, on which various applications such as data processing and format conversion are deployed.

[0023] Furthermore, the mapping relationship information at least includes mapping relationship information between tag data and data files and mapping relationship information between data files and data file processing operations.

[0024] Furthermore, the distributed storage system in step S2 stores real scene file data collected during autonomous driving, simulation scene data, and result data obtained by conversion of various data formats.

[0025] Further, in step S4, a Vigem data recorder acquisition device is used to record data generated by the sensor, the roof box and the vehicle bus;

[0026] Tags include scene tags, location tags, and time tags.

[0027] Furthermore, in step S4, the data file is stored in the SSD flash memory in VPCAP format.

[0028] Furthermore, in step S5, the data files in the SSD flash memory are uploaded to the distributed storage using the Copy Station.

[0029] Furthermore, in step S8, the collected raw data is processed in sequence to generate a final file, each step of the processing generates an intermediate file, and the generated final file and the intermediate files generated by each step of the processing are recorded and tracked using the relational database;

[0030] Establish a table for recording errors in the relational database, and write the task ID, the name of the error step and the input file of the error step running during the processing into the error table;

[0031] When an error occurs at any node, the system reports an error and the step stops running. After the developer fixes the error step, the system will continue running from the error step.

[0032] The present invention also provides an autonomous driving data processing and tracking system, comprising:

[0033] An autonomous driving data storage management cluster; comprising at least three servers, wherein the at least three servers include a database server deployed with a relational database, a data management server deployed with an autonomous driving data management platform application, and a data processing server deployed with a tool chain script application for various data processing and format conversion;

[0034] A distributed storage system having a plurality of storages, each storage being assigned an IP address within the autonomous driving data storage management cluster, and the storage being suitable for integration into the autonomous driving data storage management cluster via the corresponding IP address;

[0035] A relational database, suitable for storing the storage path and file name of the collected data files, data tags and metadata related to the data files, mapping relationships between data files and data tags, mapping relationships between data files and metadata, and mapping relationships between data file conversions and specific operations, and added to the autonomous driving data storage management cluster;

[0036] Data recorder, suitable for recording raw data collected by various sensors, roof boxes and bus data on the test vehicle;

[0037] Data management applications, for data management; and

[0038] Front-end UI.

[0039] The present invention also provides a vehicle, comprising the above-mentioned automatic driving data processing and tracking system.

[0040] The autonomous driving data processing tracking method according to the embodiment of the present invention can effectively locate, classify, track and search each file that appears in the data processing process, effectively track each link in the data processing chain, directly locate the erroneous link, and continue to run data processing from the erroneous link, thereby solving the problem of wasted time and computing power in the process of massive data processing.

[0041] Based on the following detailed description of specific embodiments of the present invention in conjunction with the accompanying drawings, those skilled in the art will become more aware of the above and other objects, advantages and features of the present invention. BRIEF DESCRIPTION OF THE DRAWINGS

[0042] Hereinafter, some specific embodiments of the present invention will be described in detail in an exemplary and non-limiting manner with reference to the accompanying drawings. The same reference numerals in the accompanying drawings indicate the same or similar components or parts. It should be understood by those skilled in the art that these drawings are not necessarily drawn to scale. In the accompanying drawings:

[0043] Figure 1 is a structural schematic diagram of an autonomous driving data processing and tracking method according to an embodiment of the present invention;

[0044] Figure 2 A design diagram of a database business table in an autonomous driving data processing and tracking method according to an embodiment of the present invention;

[0045] Figure 3 This is a schematic diagram of the raw data processing chain flow of step S8 of the autonomous driving data processing and tracking method according to an embodiment of the present invention;

[0046] Figure 4 This is a diagram of the architecture of an autonomous driving storage management cluster of an autonomous driving data processing and tracking system according to an embodiment of the present invention.

[0047] Reference numerals:

[0048] 10-Database server;

[0049] 20-Data management server;

[0050] 30-data processing server;

[0051] 40-Distributed storage system. DETAILED DESCRIPTION

[0052] See also Figures 1 to 4 As shown, the autonomous driving data processing and tracking method according to an embodiment of the present invention includes the following steps:

[0053] S1. Establish an autonomous driving data storage management cluster;

[0054] S2. Adding a distributed storage system to the storage management cluster;

[0055] S3, adding a relational database to the storage management cluster;

[0056] S4, road data collection: record the raw data collected by various sensors, roof boxes and bus data on the test vehicle, add data tags and metadata tags, write them into data files, and store the data files in SSD flash memory;

[0057] S5, data upload: uploading the data in the SSD flash memory to the distributed storage system;

[0058] S6. Run the data text import script: write the file path, file name and corresponding tag data of the original data in the distributed storage system into the relational database;

[0059] S7, running the tag data and metadata script: writing the mapping relationship between the tag data and the original data file location and the mapping relationship between the metadata device tag and the original data file location into the relational database;

[0060] S8, run the file processing script: process the scene file, write the original file, operation name and processed file mapping relationship into the relational data, write the name of the error node and the input file into the relational database for storage;

[0061] S9. Develop data storage management applications and front-end UI.

[0062] In step S1, since most of the data in the process of autonomous driving development is file data, the main task of the autonomous driving data storage management cluster is to manage and process massive data files. To achieve this task, the data storage management cluster includes at least three servers, a database server 10, on which a MySQL relational database is deployed, storing the location information, metadata information, label data information and various mapping relationship information of data files, wherein the mapping information at least includes the mapping between label data and data files and the mapping between data files and data file processing operations. A data processing server 30, on which various applications such as tool chain script applications for data cleaning and format conversion are deployed. A data management server 20, on which a data storage management platform application is deployed, is used for classified storage, precise positioning, continuous tracking and other tasks of data.

[0063] In step S2, since storing hundreds of PB of autonomous driving data requires corresponding storage hardware, the distributed storage system has multiple storages, and each storage needs to be assigned an IP address within the cluster, and the storage needs to be integrated into the storage management cluster through the corresponding IP address. The shared storage of the cluster stores the real scene files collected by the road test team, the simulation scene files generated based on the real scene, and various result data files generated after scene data cropping, cleaning and format conversion.

[0064] In step S3, what is stored in the relational database is not the specific content of the collected data file, but the location of the data file on the distributed storage, which specifically includes the storage path and file name, data tags and metadata related to the data file, the mapping relationship between the data file and the data tag, the mapping relationship between the data file and the collection device information, i.e., the metadata, the mapping relationship between the data file conversion and the specific operation, etc. In this way, each data file on the distributed storage, whether the original file or the result file, can be effectively classified, located and tracked. The specific design of the relational database business table is as follows: Figure 2 shown.

[0065] In step S4, a Vigem data recorder is used to record the raw data collected by various sensors, roof boxes and bus data on the test vehicle, wherein the sensors include radar, lidar, camera, GPS / IMU, etc. Specifically, the Vigem data recorder packages all the raw data and generates a file with a duration of about 1 minute and a suffix of VPCAP. According to the calculation of 8 hours of road sampling per day, each vehicle generates about 480 VPCAP files per day. At the same time, the collected data is tagged with data labels and metadata labels by road test followers and automatic scripts. The data labels are labels attached to these data to facilitate the classification and storage of data files. For example, scene labels include lane line status, weather, road conditions, road construction, etc., location labels include cities, suburbs, highways, rural areas, etc., and time labels include daytime, night, dusk, etc. The metadata label is to link the data recorded by the Vigem recorder with the hardware and software information that generates these data, mainly the hardware and software information of vehicles, sensors and ECUs. The final raw data and label data will be stored in the SSD flash memory.

[0066] In step S5, the Copy Station is used to upload the data files in the SSD flash memory to the distributed storage of the autonomous driving storage management cluster.

[0067] In step S6, the main task of the file import script is to write the location of the original data VPCAP file in the distributed storage, including the path and file name of the file, into the relational database. At the same time, the script will parse the original data file, extract the metadata and label data in the original data file and write them into the database.

[0068] In step S7, the main task of the script is to write the mapping relationship between the tag data and the original data file location and the mapping relationship between the metadata device tag and the original data file location into the relational database. After the data files are processed by the tag data and metadata scripts, the data files are effectively classified, and R&D personnel can quickly search for specific types of data files for processing.

[0069] In step S8, the original data files stored in the distributed storage system need to undergo various operations, such as data cutting, cleaning, format conversion, etc., to meet the needs of research and development work. Specifically, the above operations can be understood as an abstract processing chain, see Figure 3As shown. F(0) in the figure represents the original VPCAP file collected. After F(0) is processed by S(1) (representing step 1), an intermediate file F(1) is generated. After F(1) is processed by the next step S(2), an intermediate file F(2) is generated, ..., and so on, until the final file F(n) is generated. The key is that each step in the data processing chain and the intermediate files generated in each step are recorded and tracked using a relational database. Since the most common problem in the operation of the big data processing chain is that a step in it goes wrong, the usual solution is to solve the error in this step, and then start again from the beginning until each link is error-free. In this way, the steps calculated before must be recalculated, resulting in a huge waste of time and computing power. The present invention creatively writes the mapping relationship between the input path of each node data file in the processing chain, the operation name number and the output path of the processed intermediate result data file into the database, so that the input and output of each processing node in the data processing chain can be continuously tracked. At the same time, the distributed storage system saves the intermediate result data file of each processing node, and the corresponding relationship between the data processing chain number, the processing node number and the error that occurred is written into the database. Specifically, a table that records the errors in the processing chain is established in the relational database, and then the task ID of the processing chain operation, the name of the error step and the input file location of the error step, that is, the path and file name of the file generated by the previous successful operation in the data processing chain are written into the error table, and then the operation of the processing chain is stopped. In this way, after the developer repairs the step that reports an error in the processing chain, the record saved in the processing chain error table in the database can be directly used to continue running the remaining processing flow from the step that ran the error last time, until no error occurs in the entire processing chain. In this way, the time and computing power spent on each step running before the error in the data processing chain will not be wasted, and the results of each step are saved in the relational database and distributed storage.

[0070] In step S9, after all data valuable to the storage and management of autonomous driving data files are stored in the relational database, it is also necessary to develop corresponding data storage services and UI interfaces according to the business of autonomous driving research and development, so that various data required in the entire autonomous driving research and development process can be effectively stored and managed. If an error occurs in any link in the data processing chain, the user can continue to process the data directly from the incorrect link.

[0071] After the above steps S1-S8, the original data files stored in the distributed storage can be effectively located, classified and stored. The various processing operations and various result files generated by the original data will also be effectively located, classified and stored by storing them in the relational database. The steps that failed in the data processing chain are also stored in the relational database to track the errors in the data processing.

[0072] In short, through the application of relational databases in distributed storage, every file that appears in the data processing process can be effectively located, classified, tracked and searched. It can effectively track every link in the data processing chain, directly locate the wrong link, and continue to run data processing from the wrong link, solving the problem of wasted time and computing power in the process of massive data processing.

[0073] The autonomous driving data processing and tracking system of the present invention includes an autonomous driving data storage management cluster, a distributed storage system, a relational database, a data recorder, a data management application and a front-end UI. Figure 4 The autonomous driving data storage management cluster includes at least three servers, including a database server 10 with a relational database, a data management server 20 with an autonomous driving data management platform application, and a data processing server 30 with a tool chain script application for various data processing and format conversion. The data management server 20 is connected to the data management client.

[0074] The distributed storage system 40 has multiple storages, each of which is assigned an IP address within the autonomous driving data storage management cluster, and the storage is suitable for integration into the autonomous driving data storage management cluster via the IP address.

[0075] A relational database is suitable for storing the storage path and file name of the collected data files, the data tags and metadata related to the data files, the mapping relationship between data files and data tags, the mapping relationship between data files and metadata, and the mapping relationship between data file conversion and specific operations, and is added to the autonomous driving data storage management cluster.

[0076] The data logger is a Vigem data logger, which is suitable for recording the raw data collected by various sensors, roof boxes and bus data on the test vehicle.

[0077] In summary, the autonomous driving data processing and management system of the present invention can effectively locate, classify, track and search each file that appears in the data processing process through the application of relational database on distributed storage. It can effectively track each link in the data processing chain, directly locate the wrong link and continue to run data processing from the wrong link, solving the problem of time and computing power waste in the process of massive data processing.

[0078] The present invention also provides a vehicle, including the system of the above embodiment. Since the autonomous driving data processing tracking system according to the embodiment of the present invention has the above technical effects, the vehicle according to the embodiment of the present invention also has the corresponding technical effects, that is, the vehicle of the present invention, by setting the autonomous driving data processing tracking system based on the relational database, can effectively locate, classify, track and search each file that appears in the data processing process, effectively track each link in the data processing chain, directly locate the erroneous link, and continue to run data processing from the erroneous link, solving the problem of time and computing power waste in the process of massive data processing.

[0079] Other structures and operations of the vehicle according to the embodiment of the present invention are understandable and easily implemented by those skilled in the art, and thus will not be described in detail.

[0080] At this point, those skilled in the art should recognize that, although multiple exemplary embodiments of the present invention have been shown and described in detail herein, many other variations or modifications that conform to the principles of the present invention can still be directly determined or derived based on the content disclosed in the present invention without departing from the spirit and scope of the present invention. Therefore, the scope of the present invention should be understood and recognized as covering all these other variations or modifications.

Claims

1. A method for tracking autonomous driving data processing, It is characterized in that The following steps are involved: S1. Establish an autonomous driving data storage management cluster; S2. Adding a distributed storage system to the storage management cluster; S3, adding a relational database to the storage management cluster; S4, road data collection: record the raw data collected by various sensors, roof boxes and bus data on the test vehicle, add data tags and metadata tags, write them into data files, and store the data files in SSD flash memory; S5, data upload: uploading the data in the SSD flash memory to the distributed storage system; S6. Run the data text import script: write the file path, file name and corresponding tag data of the original data in the distributed storage system into the relational database; S7, running the tag data and metadata script: writing the mapping relationship between the tag data and the original data file location and the mapping relationship between the metadata device tag and the original data file location into the relational database; S8, run the file processing script: process the scene file, write the original file, operation name and processed file mapping relationship into the relational data, write the name of the error node and the input file into the relational database for storage; S9, developing a data storage management application and a front-end UI; in step S8, the collected raw data is processed in sequence to generate a final file, each step of the processing generates an intermediate file, and the generated final file and the intermediate files generated by each step of the processing are recorded and tracked using the relational database; Establish a table for recording errors in the relational database, and write the task ID, the name of the error step and the input file of the error step running during the processing into the error table; When an error occurs at any node, the system reports an error and the step stops running. After the developer fixes the error step, the system will continue running from the error step. The storage management cluster in step S1 includes at least three servers, and the at least three servers include: A relational database server, on which a MySQL relational database is deployed, is used to store location information, metadata information, label data information, and various mapping relationship information of data files; Data management server, on which data storage management applications are deployed to classify, store, accurately locate and continuously track data; A data application server on which applications for data processing and format conversion are deployed; The mapping relationship information at least includes mapping relationship information between tag data and data files and mapping relationship information between data files and data file processing operations.

2. The autonomous driving data processing and tracking method according to claim 1, It is characterized in that The distributed storage system in step S2 stores real scene file data collected during autonomous driving, simulation scene data, and result data obtained by conversion of various data formats.

3. The autonomous driving data processing and tracking method according to claim 1, It is characterized in that In step S4, the data generated by the sensor, the roof box and the vehicle bus are recorded using a Vigem data recorder acquisition device; Tags include scene tags, location tags, and time tags.

4. The autonomous driving data processing and tracking method according to claim 1 or 3, It is characterized in that In step S4, the data file is stored in the SSD flash memory in VPCAP format.

5. The autonomous driving data processing and tracking method according to claim 1, It is characterized in that In step S5, CopyStation is used to upload the data files in the SSD flash memory to the distributed storage.

6. An autonomous driving data processing and tracking system, It is characterized in that include: Autonomous driving data storage management cluster; The system comprises at least three servers, wherein the at least three servers include a database server with a relational database deployed, a data management server with an autonomous driving data management platform application deployed, and a data processing server with a tool chain script application for various data processing and format conversion deployed; A distributed storage system having a plurality of storages, each storage being assigned an IP address within the autonomous driving data storage management cluster, and the storage being suitable for integration into the autonomous driving data storage management cluster via the corresponding IP address; A relational database, suitable for storing the storage path and file name of the collected data files, data tags and metadata related to the data files, mapping relationships between data files and data tags, mapping relationships between data files and metadata, and mapping relationships between data file conversions and specific operations, and added to the autonomous driving data storage management cluster; Data recorder, suitable for recording raw data collected by various sensors, roof boxes and bus data on the test vehicle; Data management applications, used for data management; and Front-end UI; The autonomous driving data storage management cluster is configured to process the collected raw data in sequence to generate a final file, each step of the processing generates an intermediate file, and the generated final file and the intermediate file generated by each step of the processing are recorded and tracked by the relational database; The autonomous driving data storage management cluster is further configured to establish a table for recording errors in the relational database, and the task ID, the name of the error step, and the input file of the error step running during the processing are written into the error table; The autonomous driving data storage management cluster is also configured such that when an error occurs in any node, the system reports an error and the operation of the step stops. After the developer repairs the error reporting step, the system continues to run from the error step; The relational database server has a MysQL relational database deployed thereon, which is used to store location information, metadata information, label data information and various mapping relationship information of data files; The data management server has a data storage management application deployed thereon for classifying, storing, accurately locating and continuously tracking data; The data application server has data processing and format conversion applications deployed thereon; The mapping relationship information at least includes mapping relationship information between tag data and data files and mapping relationship information between data files and data file processing operations.

7. A vehicle, It is characterized in that Including an autonomous driving data processing and tracking system as described in claim 6.