Road end sensing assembly applied to vehicle-road cooperation pose calibration and calibration method

Abstract

The invention discloses a road end sensing assembly applied to vehicle-road cooperation pose calibration and a calibration method. The road end sensing assembly comprises a movable base, a visual calibration system, an initial pose calibration projection tool and a computing platform. Deploying a road end sensing assembly to a target position, and starting an initial pose calibration projection tool to project a calibration frame; carrying out observation and closed-loop calibration on the projected calibration frame by using a visual calibration system; the target vehicle is guided to enter the calibration frame and is aligned, and rapid calibration of the initial pose is completed; after the target vehicle is driven out of the calibration frame, the calculation platform adopts a road end dominant tracking mode or a vehicle-road cooperation matching mode according to the vehicle distance and the environment feature richness condition so as to carry out continuous calibration; the roadside sensing assembly has the advantages of being convenient to operate, flexible in deployment, low in cost and high in precision, and can solve the problems that in the prior art, roadside equipment is high in deployment cost, small in coverage area and complex in initial pose calibration operation, and the precision is difficult to guarantee.

Description

Technical Field

[0001] This invention relates to the field of intelligent vehicles and smart transportation, and in particular to a roadside sensing assembly and calibration method for vehicle-road cooperative pose calibration. Background Technology

[0002] In the field of intelligent vehicles and smart transportation, the rapid development of vehicle-to-infrastructure (V2I) technology has greatly promoted the intelligence of transportation systems. However, technological advancements in this field have also brought new challenges, particularly in V2I pose calibration. The positioning and navigation accuracy of intelligent vehicles in the road environment directly affects their driving safety and efficiency. To achieve high-precision V2I, accurate pose calibration between vehicles and road infrastructure is required. This involves not only the vehicle's own sensor data but also effective integration with sensing data from the roadside.

[0003] Traditional static calibration methods, such as manual measurement, while offering high accuracy, are cumbersome, inefficient, and ill-suited to the demands of real-time changes in dynamic environments. Therefore, vehicle-road cooperative online calibration has become a key technology for improving system performance. Visual online calibration technology, due to its non-contact and real-time advantages, demonstrates significant superiority in the calibration process. Compared to traditional hardware calibration methods, visual online calibration can acquire the relative position and attitude between the vehicle and the road in real time through cameras and image processing technology. This method not only improves calibration flexibility but also allows for dynamic adjustment of calibration parameters in practical applications, thus adapting to different environmental changes and vehicle motion states.

[0004] Despite the significant advantages of visual online calibration technology, its implementation, particularly algorithm verification and evaluation, still faces numerous challenges. First, data acquisition and ground truth extraction are key difficulties. Existing roadside equipment (such as fixed cameras) is costly to deploy, and the limited coverage of a single device results in numerous blind spots, hindering continuous vehicle tracking. Second, obtaining high-precision initial pose ground truth is crucial for ensuring the accuracy of subsequent continuous calibration results in algorithm verification experiments. However, traditional ground truth extraction methods (such as manual measurement or reliance on inertial navigation systems) are either inefficient or lack sufficient accuracy. Finally, achieving high-precision continuous calibration at a low cost and effectively evaluating the performance of calibration algorithms are pressing technical problems that need to be solved.

[0005] In summary, existing technologies for vehicle-road cooperative pose calibration suffer from problems such as high deployment costs, small coverage, difficulty and low efficiency in obtaining the initial pose truth value, and the need to improve continuous calibration accuracy. Summary of the Invention

[0006] To address the shortcomings of existing technologies, this application proposes a roadside sensing assembly and calibration method for vehicle-road cooperative pose calibration. It provides a roadside sensing assembly and corresponding vehicle-road cooperative pose calibration method that are easy to operate, flexible to deploy, low in cost, and high in accuracy. The aim is to solve the problems of high deployment cost, small coverage area, complex initial pose calibration operation, and difficulty in guaranteeing accuracy of roadside equipment in existing technologies.

[0007] The technical solution adopted in this invention is as follows: A roadside sensing assembly for vehicle-road cooperative pose calibration includes a movable base, a vision calibration system, an initial pose calibration projection fixture, and a computing platform. The movable base is used to support and move the entire assembly; The visual calibration system is fixed on a movable base and consists of a surround-view array of multiple cameras to capture image information of the surrounding environment of the assembly and the target vehicle, achieving wide-area coverage without blind spots. The initial pose calibration projection fixture is fixed on a movable base and has a pre-calibrated relative pose relationship with the visual calibration system; it includes a laser projection device for projecting a calibration frame with a predetermined size and position on the ground for the target vehicle to park and quickly obtain the initial relative pose. The computing platform receives and processes image data from the vision calibration system, performs pose calibration by running a pose estimation algorithm, and communicates with the vehicle-mounted equipment.

[0008] Furthermore, the visual calibration system consists of 6 to 8 industrial cameras arranged in a circle, which simultaneously acquire multiple images through a data synchronization device and stitch them together into a 360-degree panoramic image.

[0009] Furthermore, the initial pose calibration projection fixture includes an adjustable two-dimensional gimbal, which is used to adjust the angle of the laser projection device in the horizontal and pitch directions to adapt to road scenes of different widths.

[0010] Furthermore, the two-dimensional gimbal is equipped with a scale, which allows for the rapid determination of the pose parameters corresponding to the calibration frame based on the scale readings.

[0011] Furthermore, by comparing the actual image position of the calibration box with the theoretical projection position, closed-loop calibration of the initial pose parameters is performed, denoted as: ;in, To determine the theoretical corner points of the calibration frame in the world coordinate system, The coordinates are the actual pixel coordinates obtained through an image recognition algorithm, and π is the camera projection function. To determine the theoretical pixel coordinates of the calibration box on the camera image plane, Let be the transformation matrix that describes the camera's pose relative to the world coordinate system.

[0012] A vehicle-road cooperative pose calibration method includes the following steps: Step 1: Deploy the roadside sensing assembly to the target location and start the initial pose calibration projection fixture to project the calibration frame; use the vision calibration system to observe and perform closed-loop calibration on the projected calibration frame; guide the target vehicle into the calibration frame and align it to complete the rapid calibration of the initial pose. Step 2: After the target vehicle leaves the calibration frame, the computing platform adaptively switches or fuses at least two of the following modes based on vehicle distance and environmental feature richness conditions for continuous calibration; the modes include: Roadside-dominated tracking mode: The visual calibration system is used to perform three-dimensional detection and tracking of the target vehicle to obtain the vehicle pose; Vehicle-road cooperative matching mode: Real-time reception and matching of common environmental features in roadside panoramic surround view images and vehicle-side camera images, and high-precision vehicle-road relative pose is obtained by solving geometric constraints.

[0013] Furthermore, in the vehicle-road cooperative matching mode, geometric constraints are established by matching N pairs of common feature points in the roadside panoramic surround view image and the vehicle-side camera image; let the coordinate system of the roadside visual calibration system be... The coordinate system of the vehicle-mounted camera is For the i-th pair of matching points, it is in The three-dimensional coordinates below are ,exist The three-dimensional coordinates below are The core of continuous calibration is solving... arrive rigid body transformation matrix This process is achieved by minimizing the registration error of the following 3D point pairs, denoted as: ;in, Let be the rotation matrix from the vehicle-side camera coordinate system to the road-side camera coordinate system. This is the translation vector from the vehicle-side camera coordinate system to the road-side camera coordinate system.

[0014] Furthermore, during continuous calibration, when the target vehicle is occluded, a kinematic model-based prediction mode is activated to ensure the continuity of the output pose. This kinematic model-based prediction mode is implemented using an extended Kalman filter. Let the vehicle's state vector at time k be... The state transition equation is: ;in, To control the input, This is process noise; When visual observations are lost, the system predicts the state using state transition equations: ;in, This is the final result of the previous moment; This is a preliminary prediction for the current moment; when visual observation is restored, the observation model will be... ;in, For the observed values, To reduce observation noise; EKF uses Kalman gain. Update the predicted state: This prediction and update cycle ensures the continuity of pose output.

[0015] Furthermore, by utilizing the high-quality data generated during continuous calibration, parameters in the calibration model, such as the installation extrinsic parameters of the vehicle-mounted camera, are refined online. This is achieved as follows: First, the system automatically selects the calibration results with the highest confidence based on preset evaluation criteria such as the number of feature matches and reprojection error, and uses these results as high-precision observation samples. Next, after accumulating a sufficient number of samples, the system constructs a global optimization problem. The core of this problem is to simultaneously adjust the vehicle's trajectory and the internal parameters, such as the extrinsic parameters of the camera to be optimized, so that the consistency error between the theoretical predictions and actual observations of all high-precision observation samples is minimized. This optimization problem is typically expressed as a nonlinear least squares problem. The optimal parameters obtained from the solution are used to update the initial settings in the system. Through this "data-driven" self-calibration cycle, the overall calibration accuracy of the system can be continuously improved during use.

[0016] Furthermore, before deploying the roadside sensing assembly to the target location, a deployment location simulation is performed in a digital twin environment constructed with a high-precision 3D map. During actual deployment, the roadside panoramic view image collected by the roadside visual calibration system is registered with the digital twin model in this environment to achieve rapid positioning of the roadside sensing assembly in the global coordinate system.

[0017] The beneficial effects of this invention are: (1) Convenient and efficient deployment, with high accuracy and reliability in calibration starting point. This invention significantly improves deployment efficiency through an integrated movable assembly and "digital twin" guided deployment. Its original initial pose calibration method, which combines laser projection with visual closed-loop calibration, transforms complex on-site measurements into an automated "alignment and parking" operation. This not only greatly shortens preparation time but also overcomes errors caused by uncertainties in the on-site environment, providing an extremely accurate and reliable initial benchmark for subsequent calibration.

[0018] (2) Intelligent and robust calibration process with industry-leading dynamic accuracy. This invention proposes a multi-mode adaptive fusion continuous calibration strategy that can intelligently switch between modes such as road-side dominant tracking, vehicle-road cooperative matching, and kinematic prediction based on vehicle distance, environmental characteristics, and occlusion conditions. This mechanism ensures that the calibration results remain continuous and highly accurate under various complex dynamic conditions. Combined with online parameter self-optimization capabilities, the system can continuously learn and iterate during use, achieving industry-leading dynamic accuracy and environmental adaptability.

[0019] (3) High system integration, combining low cost and high scalability. This invention replaces the traditional multi-point fixed deployment scheme with a single mobile assembly, which greatly reduces hardware and deployment costs, while expanding the coverage of a single point through a surround view system. This scheme can not only achieve high-precision vehicle-road relative pose calibration, but also extend to global absolute pose positioning by combining with digital twin technology. Its technical architecture has the potential to smoothly evolve to higher-level autonomous driving applications, with significant comprehensive benefits. Attached Figure Description

[0020] Figure 1 This is a schematic diagram of the roadside sensing assembly of the present invention.

[0021] Figure 2 A schematic diagram of the structure of a two-dimensional gimbal on the projection fixture for initial pose calibration.

[0022] Figure 3 This is a schematic diagram illustrating the use of a two-dimensional gimbal to adjust the position of a laser projection device.

[0023] Figure 4 A schematic diagram of the initial pose calibration projection fixture projecting a calibration frame onto the ground. Detailed Implementation

[0024] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are for illustrative purposes only and are not intended to limit the invention.

[0025] like Figure 1 , 2 As shown in Figures 3 and 4, a roadside sensing assembly for vehicle-road cooperative pose calibration is described. The main structure of this assembly is designed as an integrated, movable unit. Specifically, the roadside sensing assembly includes: a movable base, a vision calibration system, an initial pose calibration projection fixture, and a computing platform; as detailed below: The movable base is located at the bottom and is used to support and move the entire assembly. Therefore, the movable base is equipped with lockable casters and a built-in high-capacity power bank to ensure the convenience of moving the equipment between different testing scenarios and the stability after deployment.

[0026] An adjustable-length column is vertically installed on the upper part of the base, which can raise the visual calibration system to a predetermined height to obtain a wider field of view and reduce obstruction from obstacles such as vehicles. In this embodiment, the column is a telescopic rod.

[0027] The vision calibration system is mounted on top of the column and consists of six industrial cameras arranged in a circular array to capture image information of the surrounding environment and the target vehicle. Each camera is equipped with a wide-angle lens, and a hardware synchronization device ensures frame synchronization of images acquired by all cameras. This design enables 360-degree image coverage of the surrounding environment without blind spots.

[0028] The initial pose calibration projection fixture is fixed on a movable base and has a pre-calibrated relative pose relationship with the vision calibration system. This fixture includes a laser projection device for projecting a calibration frame of predetermined size and position onto the ground, allowing the target vehicle to park and quickly acquire a high-precision initial relative pose. The computing platform integrated inside the base is the "brain" of the entire assembly. This platform uses a high-performance embedded industrial computer with a built-in wireless communication module, responsible for receiving and processing all image data acquired by the vision calibration system, running complex calibration algorithms, and interacting with the test vehicle in real time.

[0029] More specifically, the vision calibration system includes multiple data synchronization devices to ensure the synchronous acquisition of multiple image frames and to stitch the multiple images into a 360-degree panoramic image.

[0030] More specifically, the initial pose calibration projection fixture also includes an adjustable two-dimensional gimbal with a scale dial, which allows the laser projection device to make precise angle adjustments in the horizontal and pitch directions to adapt to road scenes of different widths, and can quickly determine the pose parameters corresponding to the calibration frame based on the scale readings.

[0031] More specifically, the visual calibration system of this invention is also used to observe the calibration frame projected onto the ground by the initial pose calibration projection fixture. The computing platform performs closed-loop calibration of the initial pose parameters by comparing the actual image position of the calibration frame with the theoretical projection position. Specifically, let the theoretical corner point of the calibration frame in the world coordinate system (ground) be... The actual pixel coordinates obtained through the image recognition algorithm are The goal of closed-loop calibration is to find an optimal initial pose transformation matrix from the world coordinate system to the camera coordinate system. , making The theoretical pixel coordinates after this matrix transformation and projection onto the camera image plane , and the actual observed pixel coordinates The reprojection error between them is minimized. This optimization problem can be expressed by the following formula: Where π is the camera projection function, Let be the transformation matrix that describes the camera's pose relative to the world coordinate system.

[0032] Based on the aforementioned roadside sensing assembly for vehicle-road cooperative pose calibration, this invention can also implement a vehicle-road cooperative pose calibration method, which is specifically executed according to the following steps: Step 1: Deployment and Initial Pose Calibration This process can be referenced. Figure 2 As shown. First, the entire roadside sensing assembly is moved to the predetermined test location, such as a corner of an intersection or next to a specific road section, the base is locked, and the column is raised. Then, the operator activates the initial pose calibration projection fixture. The laser projector on the fixture projects a clear, standard-sized rectangular calibration frame onto the designated lane surface. (See diagram.) Figure 2 As shown, the test vehicle simply needs to slowly drive into the calibration frame composed of light rays and precisely align the vehicle with the frame lines. Since the relative positional relationship between the projection fixture and the top-mounted visual calibration system is pre-calibrated with high precision, when the vehicle is parked within this calibration frame, its initial pose relative to the roadside sensing assembly is a precise and known preset value. After the operator confirms the alignment via the onboard terminal, the computing platform records this high-precision initial pose true value. The entire process requires no manual measurement, is fast, simple, and highly repeatable. Furthermore, by adjusting the two-dimensional gimbal on the fixture, the projection position of the calibration frame can be easily changed to adapt to lanes of different widths.

[0033] Step 2: Continuous Collaborative Calibration Once the initial pose calibration is complete, the test vehicle leaves the calibration frame and begins traveling along a predetermined trajectory within the coverage area of ​​the roadside sensing assembly. At this point, the visual calibration system located on top of the assembly begins continuous operation. Its surround-view cameras continuously capture images containing the test vehicle. The computing platform stitches and processes multiple images in real time, continuously estimating the vehicle's pose in the roadside coordinate system using object detection and tracking algorithms. Simultaneously, the test vehicle's built-in forward-facing camera captures its forward field of view, including the roadside sensing assembly itself and surrounding environmental features such as traffic lights and buildings. The vehicle wirelessly transmits these images or extracted visual features to the computing platform of the roadside sensing assembly. Upon receiving the vehicle-side data, the computing platform performs the crucial collaborative calibration step: it matches the environmental features captured by the roadside cameras from its own perspective with the environmental features sent by the vehicle-side cameras in real time. Using these feature points observed from two different perspectives, and employing visual geometry principles such as the PnP algorithm, the extremely precise relative pose between the vehicle and the roadside assembly can be accurately deduced. Finally, the computing platform employs fusion algorithms such as Kalman filtering to combine the pose results tracked by the roadside with the high-precision pose results calculated through vehicle-road cooperative matching, outputting a smooth, continuous, and highly accurate vehicle trajectory. This trajectory can then serve as a benchmark for evaluating the performance of other calibration algorithms.

[0034] In this embodiment, the continuous calibration mode includes: Roadside-dominated tracking mode: The visual calibration system is used to perform three-dimensional detection and tracking of the target vehicle to obtain the vehicle pose; Vehicle-road cooperative matching mode: Real-time reception and matching of common environmental features in roadside panoramic surround view images and vehicle-side camera images, and high-precision vehicle-road relative pose is obtained by solving geometric constraints.

[0035] More specifically, in the vehicle-road cooperative matching mode, geometric constraints are established by matching N pairs of common feature points in the roadside panoramic surround view image and the vehicle-side camera image; let the coordinate system of the roadside visual calibration system be... The coordinate system of the vehicle-mounted camera is For the i-th pair of matching points, it is in The three-dimensional coordinates below are ;exist The three-dimensional coordinates below are The core of continuous calibration is solving... arrive rigid body transformation matrix This process is achieved by minimizing the registration error of the following 3D point pairs, denoted as: ;in, Let be the rotation matrix from the vehicle-side camera coordinate system to the road-side camera coordinate system. This is the translation vector from the vehicle-side camera coordinate system to the road-side camera coordinate system.

[0036] More specifically, during continuous calibration, when the target vehicle is occluded, a kinematic model-based prediction mode is activated to ensure the continuity of the output pose. This kinematic model-based prediction mode is implemented using an extended Kalman filter. Let the vehicle's state vector at time k be: ;in, To control the input, This is process noise; When visual observations are lost, the system predicts the state using state transition equations: ;in, This is the final result of the previous moment; This is a preliminary prediction for the current moment; when visual observation is restored, the observation model will be... ;in, For the observed values, To reduce observation noise; EKF uses Kalman gain. Update the predicted state: This prediction and update cycle ensures the continuity of pose output.

[0037] More specifically, this method also includes an online parameter self-optimization step. This step aims to refine parameters in the calibration model, such as the extrinsic parameters of the vehicle-mounted camera, online using high-quality data generated during continuous calibration. The implementation is as follows: First, the system automatically selects the calibration results with the highest confidence based on preset evaluation criteria such as the number of feature matches and reprojection error, and uses these results as high-precision observation samples. Next, after accumulating a sufficient number of samples, the system constructs a global optimization problem. The core of this problem is to simultaneously adjust the vehicle's trajectory and the internal parameters, such as the extrinsic parameters of the camera to be optimized, so that the consistency error between the theoretical predictions and actual observations of all high-precision observation samples is minimized. This optimization problem is typically expressed as a nonlinear least squares problem. The optimal parameters obtained from the solution are used to update the initial settings in the system. Through this "data-driven" self-calibration loop, the overall calibration accuracy of the system can be continuously improved during use.

[0038] More specifically, before deploying the roadside sensing assembly to the target location, a deployment location simulation is performed in a digital twin environment constructed with a high-precision 3D map. During actual deployment, the roadside panoramic view image collected by the roadside visual calibration system is registered with the digital twin model in this environment to achieve rapid positioning of the roadside sensing assembly in the global coordinate system.

[0039] The above embodiments are only used to illustrate the design concept and features of the present invention, and their purpose is to enable those skilled in the art to understand the content of the present invention and implement it accordingly. The protection scope of the present invention is not limited to the above embodiments. Therefore, all equivalent changes or modifications made based on the principles and design ideas disclosed in the present invention are within the protection scope of the present invention.

Claims

1. A road-side sensing assembly applied to vehicle-road cooperative pose calibration, characterized in that, The application relates to a road-side sensing assembly for vehicle-road cooperative pose calibration, which comprises a movable base, a visual calibration system, an initial pose calibration projection tool and a computing platform. The movable base is used for carrying and moving the whole assembly. The visual calibration system is fixed to the movable base and comprises a ring-shaped array of multiple cameras, which are used for capturing image information of the surrounding environment of the assembly and a target vehicle, so as to realize wide-area blind-free coverage. The initial pose calibration projection tool is fixed to the movable base and has a pre-calibrated relative pose relationship with the visual calibration system. The computing platform receives and processes image data of the visual calibration system, realizes pose calibration by running a pose estimation algorithm and communicates with a vehicle-side device.

2. The road-side sensing assembly for vehicle-to-everything pose calibration of claim 1, wherein, The visual calibration system is composed of 6-8 industrial cameras, which are arranged in a ring shape, and the multiple images are synchronously collected by a data synchronization device and spliced into a 360-degree panoramic ring-view image.

3. The road-side sensing assembly for vehicle-to-everything pose calibration of claim 1, wherein, The initial pose calibration projection tool comprises an adjustable two-dimensional holder, which is used for adjusting the angle of the laser projection device in the horizontal and pitching directions by the two-dimensional holder, so as to adapt to different road scenes.

4. The road-side sensing assembly for vehicle-to-everything pose calibration of claim 3, wherein, The two-dimensional holder is provided with a scale dial, and the pose parameters corresponding to the calibration frame are quickly determined according to the scale readings on the scale dial.

5. The road-side sensing assembly for vehicle-to-everything pose estimation of claim 1, wherein, The initial pose parameters are closed-loop calibrated by comparing the actual image position of the calibration frame with the theoretical projection position, denoted as: ; wherein, is the theoretical corner of the calibration frame in the world coordinate system, is the actual pixel coordinate obtained by the image recognition algorithm, and π is the camera projection function, is the theoretical pixel coordinate of the calibration frame on the camera image plane, is the transformation matrix to be solved, which describes the pose of the camera relative to the world coordinate system.

6. A vehicle-road cooperation pose calibration method, characterized in that, The application relates to a road-side sensing assembly for vehicle-road cooperative pose calibration, which comprises a movable base, a visual calibration system, an initial pose calibration projection tool and a computing platform. Step 1: the road-side sensing assembly is deployed to a target position, the initial pose calibration projection tool is started to project a calibration frame, the visual calibration system is used for observing and closed-loop calibrating the projected calibration frame, the target vehicle is guided to drive into the calibration frame and is aligned, and the initial pose is quickly calibrated. Step 2: after the target vehicle drives out of the calibration frame, the computing platform is used for adaptively switching or fusing in at least two modes according to the vehicle distance and the environmental feature richness to continuously calibrate. The modes include: A road-side leading tracking mode, in which the visual calibration system is used for three-dimensional detection and tracking of the target vehicle to obtain the vehicle pose. A vehicle-road cooperative matching mode, in which common environmental features in the road-side panoramic ring-view image and the vehicle-side camera image are matched in real time, and high-precision vehicle-road relative poses are obtained by solving geometric constraints.

7. The vehicle-road cooperative pose calibration method according to claim 6, characterized in that, In the vehicle-road cooperative matching mode, a geometric constraint is established by matching N pairs of common feature points in the road-end panoramic view image and the vehicle-end camera image; a road-end vision calibration system coordinate system is ; and a vehicle-end camera coordinate system is . For the i-th pair of matching points, it is in The three-dimensional coordinates below are ,exist The three-dimensional coordinates below are The core of continuous calibration is solving... arrive rigid body transformation matrix This process is achieved by minimizing the registration error of the following 3D point pairs, denoted as: ;in, Let be the rotation matrix from the vehicle-side camera coordinate system to the road-side camera coordinate system. This is the translation vector from the vehicle-side camera coordinate system to the road-side camera coordinate system.

8. The vehicle-road cooperation pose calibration method according to claim 6, characterized in that, In the continuous calibration process, when the target vehicle is blocked, a prediction mode based on a kinematic model is enabled to ensure the continuity of the output pose, and the prediction mode based on the kinematic model is implemented by using an extended Kalman filter; assuming that the state vector of the vehicle at time k is , and the state transition equation is: ; wherein, is the control input, is the process noise; When visual observation is lost, the system makes state prediction through state transition equation: ; where, is the final result at the last moment; is the preliminary prediction at the current moment; when visual observation is restored, the observation model is ; where, is the observation value, is the observation noise; EKF updates the predicted state through Kalman gain : K = PpH (HPpH + R)-1 ; through the prediction and update cycle, the continuity of the pose output is ensured.

9. The vehicle-road cooperative pose calibration method according to claim 6, characterized in that, High-quality data generated in the continuous calibration process are used for online refining of parameters such as the installation external parameters of the vehicle-side camera in the calibration model. Firstly, the system automatically selects the calibration results with the highest confidence according to preset feature matching quantity, re-projection error and other evaluation standards, and the results are used as high-precision observation samples. Then, when enough samples are accumulated, the system constructs a global optimization problem. The core of the problem is to simultaneously adjust the driving track of the vehicle and the internal parameters such as the camera external parameters to be optimized, so that the consistency error between the theoretical prediction and the actual observation of all high-precision observation samples is minimized. The optimization problem is usually expressed as a nonlinear least square problem. The optimal parameters obtained by solving the problem are used for updating the initial set values in the system. Through this "data-driven" self-calibration cycle, the overall calibration accuracy of the system can be continuously improved during use.

10. The vehicle-road cooperation pose calibration method according to claim 6, characterized in that, Before deploying the road end sensing assembly to the target position, a simulation of the deployment position is carried out in the digital twin environment of high-precision three-dimensional map construction, and when actually deployed, the panoramic look-around image of the road end captured by the road end visual calibration system is matched with the digital twin model in the environment to realize rapid positioning of the road end sensing assembly in the global coordinate system.

Images