Distributed loose coupling automatic driving vehicle calibration system

Abstract

The invention relates to a distributed loosely-coupled automatic driving vehicle calibration system, and the system comprises a core algorithm module which comprises a standard interface unit and a calibration algorithm unit, the standard interface unit is used for connecting a vehicle-mounted sensor, obtaining laser radar point cloud data and camera image data, and transmitting the data to the calibration algorithm unit; the calibration algorithm unit is connected with the standard interface unit and is used for receiving point cloud and image data for calculation and outputting calibration result data of the vehicle sensor; and the peripheral service module is connected with the core algorithm module and comprises a scene recognition unit, a task scheduling unit and a result arrangement and writing unit. The problems of high coupling, low efficiency, poor compatibility and the like of a traditional calibration system are solved, automation, high precision and high reliability of automatic driving vehicle calibration are achieved, and key technical support is provided for large-scale mass production and continuous iteration.

Description

Technical Field

[0001] The present invention relates to the technical field of autonomous driving vehicle calibration, and in particular to a distributed loosely coupled autonomous driving vehicle calibration system. Background Art

[0002] From an application perspective, calibration software must be deployed on multiple platforms, including onboard domain controllers, diagnostic PADs, diagnostic PCs, and cloud servers. Calibration scenarios include production line calibration (software system deployment within domain controllers, on PCs, or in the cloud), after-sales calibration (diagnostic PAD deployment, controller internal deployment, PC deployment), and online calibration (deployment within domain controllers, in the cloud). This poses challenges to the portability and interface versatility of the software system, requiring universal interfaces to adapt to peripheral software on multiple platforms. It also requires that all communication interfaces be predefined and standardized to meet the needs of rapid multi-platform adaptation.

[0003] The existing coupling of software and hardware hinders development. Sensor calibration algorithms are deeply tied to hardware drivers, requiring a complete rebuild of the communication protocol stack for every sensor model change, increasing development cycles by over 30%. Existing non-distributed software architectures rely on fixed topologies, preventing rapid iteration and replacement during algorithm updates. Existing non-distributed software relies on specific operating environments, making it difficult to adapt simultaneously to diverse environments, such as vehicle-side controllers, diagnostic pads, and host PCs, making it difficult to meet diverse calibration requirements. Summary of the Invention

[0004] In view of the above problems, the present invention provides a distributed loosely coupled autonomous driving vehicle calibration system, which not only solves the problems of high coupling, low efficiency, and poor compatibility existing in traditional calibration systems, but also realizes the automation, high precision and high reliability of autonomous driving vehicle calibration, providing key technical support for large-scale mass production and continuous iteration.

[0005] In order to achieve the above-mentioned and other related purposes, the present invention provides the following technical solutions:

[0006] A distributed loosely coupled autonomous driving vehicle calibration system, comprising:

[0007] The core algorithm module includes a standard interface unit and a calibration algorithm unit. The standard interface unit is used to connect to the vehicle sensor to obtain lidar point cloud data and camera image data. The calibration algorithm unit is connected to the standard interface unit to receive point cloud and image data for calculation and output the calibration result data of the vehicle sensor.

[0008] A peripheral business module, connected to the core algorithm module, includes a scene recognition unit, a task scheduling unit, and a result collation and writing unit. The scene recognition unit is used to receive lidar point cloud data and camera image data, and use a scene recognition algorithm based on a variant Gaussian distribution to identify and classify the scene to obtain data information of the classified scene. The task scheduling unit is used to schedule different core algorithm modules based on the classified scene data information to meet the autonomous driving sensor calibration requirements in different scenarios. The result collation and writing unit is used to process the calibration results of the autonomous driving sensors in different scenarios, convert the various results into different formats, and then send or write them to the user.

[0009] The communication interface conversion module is used to connect the business module, core algorithm module and external sensor signal input, and is used for data interface standardization and data format conversion.

[0010] Furthermore, the communication structure conversion module includes a hardware driving unit, a middleware communication unit and a data serialization unit. The hardware driving unit is used as a channel for external communication to convert internal data into actual physical signal data of the hardware.

[0011] Furthermore, the middleware communication unit is used as a bridge for data interaction between various parts within the system, connecting the hardware driver and the specific variable values ​​shared by the system's internal data memory.

[0012] Furthermore, the data serialization unit is used to package various types of data during the operation of the system according to a predefined protocol and convert them into a protocol standard data format, and convert different types of data into standard data to provide support for the replaceability and portability of various modules of the system.

[0013] Furthermore, the scene recognition algorithm based on the variant Gaussian distribution is used to identify and classify the scene, including:

[0014] M1. Based on the camera image data, extract the pixel value of each pixel in the image and construct the probability density function Q of the variant Gaussian distribution of the pixel.

[0015] ,

[0016] Where x is the pixel value of each pixel in the image, ɑ1, ɑ2 and ɑ3 are weight coefficients, which characterize the distribution state of the pixels of the image and obtain data information of the distribution state of the pixels of the image;

[0017] M2. Based on the laser radar point cloud data, extract the feature values ​​of the point cloud and construct the probability density function W of the variation Gaussian distribution of the point cloud.

[0018] ,

[0019] Among them, y is the eigenvalue of the point cloud, β1, β2 and β3 are penalty factors, which characterize the distribution state of the point cloud and obtain the data information of the distribution state of the point cloud;

[0020] M3. Establishing a scene category function R based on the data information of the distribution state of the point cloud and the data information of the distribution state of the pixels of the image,

[0021] ,

[0022] Among them, z1 is the data information of the distribution state of the point cloud, z2 is the data information of the distribution state of the pixel points of the image, δ1, δ2 and δ3 are any constant parameters between 0 and 1, and the scene is identified and classified to obtain the data information of the classified scene.

[0023] Furthermore, the weight coefficients ɑ1, ɑ2 and ɑ3 are,

[0024] ,

[0025] ,

[0026] ,

[0027] Among them, x max is the maximum value of the pixel value of each pixel in the image, x min The minimum value of the pixel values ​​of each pixel in the image.

[0028] Furthermore, the penalty factors β1, β2 and β3 are,

[0029] ,

[0030] ,

[0031] ,

[0032] Among them, y max is the maximum value of the eigenvalue of the point cloud, y min is the minimum eigenvalue of the point cloud.

[0033] Furthermore, the constraint function f of the constant parameters δ1, δ2 and δ3 is,

[0034] ,

[0035] Among them, the value range of the constraint function f is (1,2).

[0036] Furthermore, the hardware driver unit includes a CAN communication card driver subunit, a network card driver subunit and a serial port driver subunit.

[0037] Furthermore, the scene recognition unit is communicatively connected to the task scheduling unit, and the task scheduling unit is communicatively connected to the result collating and writing unit.

[0038] The present invention has the following positive effects:

[0039] 1. This invention solves the problems of high coupling, low efficiency, and poor compatibility existing in traditional calibration systems through core technologies such as distributed architecture decoupling, standardized interfaces, and online collaborative optimization. It realizes the automation, high precision, and high reliability of autonomous vehicle calibration, providing key technical support for large-scale mass production and continuous iteration.

[0040] 2. This invention decouples hardware physical signals from software data by pre-installing universal interface protocols such as CAN communication card drivers and network card drivers. For example, deploying a standardized communication protocol stack between the vehicle controller and the diagnostic instrument PAD allows the same algorithm module to be adapted to different hardware platforms, reducing the interface development workload.

[0041] 3. This invention utilizes a middleware architecture similar to CyberRT to define unified data exchange channels between modules (e.g., shared memory variable mapping). For example, LiDAR point cloud data is directly transmitted to the core algorithm module via the middleware, eliminating the need for fixed topology. Switching between hardware environments requires only adjusting middleware parameters, shortening development cycles. Furthermore, by converting sensor data (PCD point clouds, video streams) into standard protocol formats (e.g., Protobuf), multi-platform data interoperability is enabled. For example, cloud servers and onboard controllers interact directly via serialized data packets, reducing cross-platform data format conversion overhead. BRIEF DESCRIPTION OF THE DRAWINGS

[0042] Figure 1 Schematic diagram of the system framework of the present invention;

[0043] Figure 2 Schematic diagram of the process of the scene recognition algorithm based on variant Gaussian distribution of the present invention. DETAILED DESCRIPTION

[0044] The following description of exemplary embodiments of the present disclosure is made in conjunction with the accompanying drawings, including various details of the embodiments of the present disclosure to facilitate understanding. These details should be considered as merely exemplary. Therefore, those skilled in the art will recognize that various changes and modifications may be made to the embodiments described herein without departing from the scope and spirit of the present disclosure. Similarly, for the sake of clarity and conciseness, descriptions of well-known functions and structures are omitted in the following description.

[0045] Example 1: Figure 1 As shown, a distributed loosely coupled autonomous driving vehicle calibration system includes:

[0046] The core algorithm module includes a standard interface unit and a calibration algorithm unit. The standard interface unit is used to connect to the vehicle sensor to obtain lidar point cloud data and camera image data. The calibration algorithm unit is connected to the standard interface unit to receive point cloud and image data for calculation and output the calibration result data of the vehicle sensor.

[0047] A peripheral business module, connected to the core algorithm module, includes a scene recognition unit, a task scheduling unit, and a result collation and writing unit. The scene recognition unit is used to receive lidar point cloud data and camera image data, and use a scene recognition algorithm based on a variant Gaussian distribution to identify and classify the scene to obtain data information of the classified scene. The task scheduling unit is used to schedule different core algorithm modules based on the classified scene data information to meet the autonomous driving sensor calibration requirements in different scenarios. The result collation and writing unit is used to process the calibration results of the autonomous driving sensors in different scenarios, convert the various results into different formats, and then send or write them to the user.

[0048] The communication interface conversion module is used to connect the business module, core algorithm module and external sensor signal input, and is used for data interface standardization and data format conversion.

[0049] In this embodiment, the communication structure conversion module includes a hardware driver unit, a middleware communication unit and a data serialization unit. The hardware driver unit is used as a channel for external communication and converts internal data into actual physical signal data of the hardware.

[0050] In this embodiment, the middleware communication unit is used as a bridge for data interaction between various parts within the system, connecting the hardware driver and the specific variable values ​​shared by the system's internal data memory.

[0051] In this embodiment, the data serialization unit is used to package various types of data during the operation of the system according to a predefined protocol, convert them into a protocol standard data format, and convert different types of data into standard data to provide support for the replaceability and portability of various modules of the system.

[0052] In this embodiment, the hardware driver unit includes a CAN communication card driver subunit, a network card driver subunit, and a serial port driver subunit.

[0053] In this embodiment, the scene recognition unit is communicatively connected to the task scheduling unit, and the task scheduling unit is communicatively connected to the result collating and writing unit.

[0054] Data serialization. Data serialization refers to the process of packaging various types of data during the operation of a software system according to a predefined protocol and converting them into a standard data format. This process converts different types of data into standard data, supporting the modular interchangeability and portability of the software system.

[0055] Scene recognition. Scene recognition is a component of the outsourced business module. It is used to identify the current calibration scenario and select different core algorithm modules to enable the software system to adapt to different calibration scenarios such as EOL and after-sales.

[0056] Task scheduling. Task scheduling schedules the core algorithm modules based on the scene recognition results and is an important part of connecting business needs and algorithm processing.

[0057] Result collation and writing: Result collation and writing is the process of collating the calibration results received by the business module after the core algorithm module receives them. The various calibration results are collated into the required format and written into the designated file path and file.

[0058] Standard interfaces. These encapsulate the core algorithm module's external input and output interfaces and embody data serialization. They serialize and normalize the data of different sub-functions within the core algorithm module into a common external interface, enabling the use of the same input and output interfaces across multiple algorithm modules and avoiding frequent data interface changes after module replacement.

[0059] Calibration algorithm. The calibration algorithm refers to the core module in the core algorithm module that implements the calibration process. Different calibration algorithms correspond to different scenarios and calibration requirements.

[0060] Example 2: Based on the distributed coupled autonomous driving vehicle calibration system of Example 1, the present invention is further illustrated and described below.

[0061] like Figure 1 As shown, a distributed loosely coupled autonomous driving vehicle calibration system includes:

[0062] The core algorithm module includes a standard interface unit and a calibration algorithm unit. The standard interface unit is used to connect to the vehicle sensor to obtain lidar point cloud data and camera image data. The calibration algorithm unit is connected to the standard interface unit to receive point cloud and image data for calculation and output the calibration result data of the vehicle sensor.

[0063] A peripheral business module, connected to the core algorithm module, includes a scene recognition unit, a task scheduling unit, and a result collation and writing unit. The scene recognition unit is used to receive lidar point cloud data and camera image data, and use a scene recognition algorithm based on a variant Gaussian distribution to identify and classify the scene to obtain data information of the classified scene. The task scheduling unit is used to schedule different core algorithm modules based on the classified scene data information to meet the autonomous driving sensor calibration requirements in different scenarios. The result collation and writing unit is used to process the calibration results of the autonomous driving sensors in different scenarios, convert the various results into different formats, and then send or write them to the user.

[0064] The communication interface conversion module is used to connect the business module, core algorithm module and external sensor signal input, and is used for data interface standardization and data format conversion.

[0065] In this embodiment, if Figure 2 As shown, the scene recognition algorithm based on variant Gaussian distribution is used to identify and classify the scene, including:

[0066] M1. Based on the camera image data, extract the pixel value of each pixel in the image and construct the probability density function Q of the variant Gaussian distribution of the pixel.

[0067] ,

[0068] Where x is the pixel value of each pixel in the image, ɑ1, ɑ2 and ɑ3 are weight coefficients, which characterize the distribution state of the pixels of the image and obtain data information of the distribution state of the pixels of the image;

[0069] M2. Based on the laser radar point cloud data, extract the feature values ​​of the point cloud and construct the probability density function W of the variation Gaussian distribution of the point cloud.

[0070] ,

[0071] Among them, y is the eigenvalue of the point cloud, β1, β2 and β3 are penalty factors, which characterize the distribution state of the point cloud and obtain the data information of the distribution state of the point cloud;

[0072] M3. Establishing a scene category function R based on the data information of the distribution state of the point cloud and the data information of the distribution state of the pixels of the image,

[0073] ,

[0074] Among them, z1 is the data information of the distribution state of the point cloud, z2 is the data information of the distribution state of the pixel points of the image, δ1, δ2 and δ3 are any constant parameters between 0 and 1, and the scene is identified and classified to obtain the data information of the classified scene.

[0075] In this embodiment, the weight coefficients ɑ1, ɑ2 and ɑ3 are,

[0076] ,

[0077] ,

[0078] ,

[0079] Among them, x max is the maximum value of the pixel value of each pixel in the image, x min The minimum value of the pixel values ​​of each pixel in the image.

[0080] In this embodiment, the penalty factors β1, β2 and β3 are,

[0081] ,

[0082] ,

[0083] ,

[0084] Among them, y max is the maximum value of the eigenvalue of the point cloud, y min is the minimum eigenvalue of the point cloud.

[0085] In this embodiment, the constraint function f of the constant parameters δ1, δ2 and δ3 is,

[0086] ,

[0087] Among them, the value range of the constraint function f is (1,2).

[0088] Modular layered structure: including core algorithm module (computing layer), peripheral business module (control layer), and communication interface conversion module (transmission layer). The three are loosely coupled through standardized interfaces, supporting multi-platform deployment and heterogeneous hardware collaboration.

[0089] Hardware driver adaptation rules: Covers the definition and implementation methods of common interface protocols such as CAN communication card drivers and network card drivers, ensuring seamless porting of algorithm modules on platforms such as vehicle controllers and diagnostic instrument PADs.

[0090] Data serialization protocol: Based on the standardized data packaging rules of Protobuf, including sensor data field definition, verification mechanism and version compatibility processing method.

[0091] Middleware communication decoupling: A CyberRT-like middleware architecture and a shared memory mapping mechanism enable sensor data (such as camera video streams) to be transmitted directly to the core algorithm module through the middleware, eliminating the need for a fixed topology. Hardware switching requires only adjusting the middleware parameters.

[0092] In summary, the present invention not only solves the problems of high coupling, low efficiency, and poor compatibility existing in traditional calibration systems, but also realizes the automation, high precision, and high reliability of autonomous vehicle calibration, providing key technical support for large-scale mass production and continuous iteration.

[0093] The above specific embodiments do not constitute a limitation on the scope of protection of this disclosure. Those skilled in the art will appreciate that various modifications, combinations, sub-combinations, and substitutions may be made based on design requirements and other factors. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of this disclosure shall be included within the scope of protection of this disclosure.

Claims

1. A distributed loosely coupled autonomous driving vehicle calibration system, characterized in that: The system comprises: The core algorithm module includes a standard interface unit and a calibration algorithm unit. The standard interface unit is used to connect to the vehicle sensor to obtain lidar point cloud data and camera image data. The calibration algorithm unit is connected to the standard interface unit to receive point cloud and image data for calculation and output the calibration result data of the vehicle sensor. A peripheral business module, connected to the core algorithm module, includes a scene recognition unit, a task scheduling unit, and a result collation and writing unit. The scene recognition unit is used to receive lidar point cloud data and camera image data, and use a scene recognition algorithm based on a variant Gaussian distribution to identify and classify the scene to obtain data information of the classified scene. The task scheduling unit is used to schedule different core algorithm modules based on the classified scene data information to meet the autonomous driving sensor calibration requirements in different scenarios. The result collation and writing unit is used to process the calibration results of the autonomous driving sensors in different scenarios, convert the various results into different formats, and then send or write them to the user. The communication interface conversion module is used to connect the business module, core algorithm module and external sensor signal input, and is used for data interface standardization and data format conversion.

2. The distributed loosely coupled autonomous driving vehicle calibration system according to claim 1, characterized in that: The communication structure conversion module includes a hardware driving unit, a middleware communication unit and a data serialization unit. The hardware driving unit is used as a channel for external communication and converts internal data into actual physical signal data of the hardware.

3. The distributed loosely coupled autonomous driving vehicle calibration system according to claim 2, characterized in that: The middleware communication unit is used as a bridge for data interaction between various parts of the system, connecting the hardware driver and the specific variable values ​​shared by the system's internal data memory.

4. The distributed loosely coupled autonomous driving vehicle calibration system according to claim 2, characterized in that: The data serialization unit is used to package various types of data during system operation according to a pre-defined protocol and convert them into a protocol standard data format, and convert different types of data into standard data to provide support for the replaceability and portability of various modules of the system.

5. The distributed loosely coupled autonomous driving vehicle calibration system according to claim 1, characterized in that: The method of identifying and classifying scenes using a scene recognition algorithm based on variant Gaussian distribution includes: M1. Based on the camera image data, extract the pixel value of each pixel in the image and construct the probability density function Q of the variant Gaussian distribution of the pixel. ， Where x is the pixel value of each pixel in the image, ɑ1, ɑ2 and ɑ3 are weight coefficients, which characterize the distribution state of the pixels of the image and obtain data information of the distribution state of the pixels of the image; M2. Based on the laser radar point cloud data, extract the feature values ​​of the point cloud and construct the probability density function W of the variation Gaussian distribution of the point cloud. ， Among them, y is the eigenvalue of the point cloud, β1, β2 and β3 are penalty factors, which characterize the distribution state of the point cloud and obtain the data information of the distribution state of the point cloud; M3. Establishing a scene category function R based on the data information of the distribution state of the point cloud and the data information of the distribution state of the pixels of the image, ， Among them, z1 is the data information of the distribution state of the point cloud, z2 is the data information of the distribution state of the pixel points of the image, δ1, δ2 and δ3 are any constant parameters between 0 and 1, and the scene is identified and classified to obtain the data information of the classified scene.

6. The distributed loosely coupled autonomous driving vehicle calibration system according to claim 5, characterized in that: The weight coefficients ɑ1, ɑ2 and ɑ3 are, ， ， ， Among them, x max is the maximum value of the pixel value of each pixel in the image, x min The minimum value of the pixel values ​​of each pixel in the image.

7. The distributed loosely coupled autonomous driving vehicle calibration system according to claim 5, characterized in that: The penalty factors β1, β2 and β3 are, ， ， ， Among them, y max is the maximum value of the eigenvalue of the point cloud, y min is the minimum eigenvalue of the point cloud.

8. The distributed loosely coupled autonomous driving vehicle calibration system according to claim 5, characterized in that: The constraint function f of the constant parameters δ1, δ2 and δ3 is, ， Among them, the value range of the constraint function f is (1,2).

9. The distributed loosely coupled autonomous driving vehicle calibration system according to claim 2, characterized in that: The hardware drive unit includes a CAN communication card drive subunit, a network card drive subunit and a serial port drive subunit.

10. The distributed loosely coupled autonomous driving vehicle calibration system according to claim 1, characterized in that: The scene recognition unit is in communication with the task scheduling unit, and the task scheduling unit is in communication with the result collating and writing unit.

Images