A method, system, and medium for evaluating rendering quality of a SR interface

Abstract

The present application relates to a kind of methods for evaluating the rendering quality of SR interface, the method comprises: step 1, implantation lightweight burying point logic, to obtain the common feature of different rendering engine, and obtain perception data by intelligent driving domain controller;Step 2, the common feature and perception data are spatio-temporal alignment;Step 3, based on the data after alignment, calculate the core evaluation index, including: real-time, geometric accuracy, stability, model expressiveness and safety;Step 4, set the weight coefficient vector of each index W=[w1,w2,...,w5], the comprehensive score S is calculated by normalization processing and weighted summation: S=∑(Metrici×Wi), according to the comprehensive score S, SR rendering quality is divided into several grades, and the visual evaluation report containing radar chart and score loss reason analysis is generated.Solve the technical problems such as the existing technical evaluation scheme being limited by rendering engine, spatio-temporal benchmark being difficult to unify, safety evaluation dimension missing and the like.

Description

Technical Field

[0001] This invention relates to the field of intelligent driving and automotive software testing technology, and in particular to a method, system, and medium for evaluating the rendering quality of SR interfaces. Background Technology

[0002] Currently, advanced autonomous driving vehicles commonly feature SR (Radio Reflection) rendering capabilities. This function typically relies on high-performance in-vehicle chips (such as the Qualcomm 8295) and utilizes commercial game engines or automakers' self-developed rendering engines to present environmental elements perceived by the autonomous driving system, such as lane lines, surrounding vehicles, pedestrians, and traffic facilities, in 3D model form on the central control screen or dashboard. This aims to enhance the driver's trust in the autonomous driving system's status. However, significant technical bottlenecks exist in the acceptance testing of SR rendering effects: evaluation schemes are limited by single rendering architectures; existing automated testing methods heavily rely on script systems specific to certain engines (e.g., only supporting Unity). With the development of autonomous driving technology, some automakers are beginning to adopt Unreal Engine or self-developed C++ rendering engines to pursue ultimate performance. Existing evaluation tools are difficult to adapt across architectures, resulting in inconsistent testing standards and an inability to cope with industry technological iterations. Spatiotemporal benchmarks are also difficult to unify: from the autonomous driving domain controller (ADC) issuing the perceived target to the rendering engine completing the drawing and submitting it to the GPU, heterogeneous systems and complex software stacks are involved. Existing technologies struggle to achieve high-precision time alignment (microsecond level) across systems, resulting in large end-to-end latency measurement errors and an inability to accurately pinpoint performance bottlenecks. Safety evaluation dimensions are lacking: SR rendering is not only a visual effect but also a crucial interactive interface for human-machine interface (HMI). If the rendered image shows misrepresented distances (displayed distances greater than actual distances) or critical hazardous targets (such as cut-in vehicles) are missing from the image, it can easily mislead the driver into making incorrect takeover decisions. Existing technologies lack a quantitative safety risk assessment system for the rendered data itself, making it difficult to support ISO 26262 functional safety verification. Summary of the Invention

[0003] In view of the above problems, the present invention provides a method, system and medium for evaluating the rendering quality of SR interface, so as to solve the technical problems of existing evaluation schemes being limited by rendering engine, difficult to unify spatiotemporal benchmarks, and lack of security evaluation dimensions.

[0004] This invention provides a method for evaluating the rendering quality of an SR (Rendering Simulation) interface. The method includes: Step 1, embedding lightweight tracking logic to obtain common features of different rendering engines and acquiring perception data through an intelligent driving domain controller; Step 2, aligning the common features and perception data spatiotemporally; Step 3, calculating core evaluation indicators based on the aligned data, including: real-time performance, geometric accuracy, stability, model expressiveness, and security; Step 4, setting the weight coefficient vector W=[w1,w2,...,w5] for each indicator, calculating the comprehensive score S through normalization and weighted summation: S=∑(Metrici×Wi), dividing the SR rendering quality into several levels according to the comprehensive score S, and generating a visual evaluation report including a radar chart and analysis of the reasons for the score loss, where is the normalized score of the core evaluation indicator, i represents all evaluation dimensions, and Wi is the preset weight coefficient corresponding to the i-th indicator.

[0005] Furthermore, step 1 includes: Step 11, pre-setting the logical node at the end of the rendering pipeline and before GPU submission as the timing for data acquisition, ensuring that the acquired data is the deterministic state data finally presented on the screen, and eliminating jitter caused by asynchronous calculation; Step 12, generating a rendering submission timestamp through a high-precision timer, and synchronizing it with the master clock via the vehicle Ethernet to ensure that the time base of all electronic control units is consistent; Step 13, traversing the active nodes of the current SR scene and extracting standardized attributes that are independent of the rendering engine, including: model identifier, world coordinates, orientation angle, and visibility status; Step 14, serializing the structured data of the standardized attributes into a lightweight binary format independent of the engine, and sending it as a common feature to the in-vehicle log server via shared memory or UDP protocol to avoid blocking the main rendering thread.

[0006] Furthermore, step 2 includes: step 21, acquiring perception data containing perception timestamps and high-precision RTK-GPS / inertial measurement unit true data containing ground truth timestamps through the intelligent driving domain controller; step 22, using the rendering submission timestamp as the reference timestamp, aligning the perception data and ground truth data to the same moment using a linear interpolation algorithm; step 23, calibrating the extrinsic parameter matrix, establishing a universal coordinate transformation model, transforming the world coordinates of the rendering engine to the UTM coordinate system or vehicle coordinate system consistent with the RTK ground truth, eliminating errors caused by different installation locations and coordinate system definitions.

[0007] Furthermore, step 3 includes: Step 31, calculating the end-to-end rendering latency and statistically analyzing the standard deviation of the rendering submission time error interval between consecutive frames to reflect the system response speed and the stability of the rendering thread; Step 32, calculating the deviation between the converted rendering coordinates / heading and the ground truth data, as well as the deviation between the rendered relative distance between the main vehicle and the target vehicle and the ground truth relative distance; Step 33, statistically analyzing the number of times the target position or heading angle change exceeds a preset threshold per unit time to evaluate the smoothness of the perception fusion output; Step 34, statistically analyzing the ratio of the number of different Model IDs rendered within the test period to the total number of preset model libraries to reflect the richness of rendering resources, comparing the consistency between the target classification in the perception data and the rendered Model ID to reflect the accuracy of the mapping from the perception result to the rendered object; Step 35, in the AEB / FCW trigger critical zone, calculating the critical target loss rate = number of lost frames / total number of frames in the dangerous scene, and determining it as a high-risk error when the relative distance Dsr displayed by SR rendering is greater than the actual relative distance Dreal and the difference exceeds a preset safety threshold.

[0008] Furthermore, among the weighting coefficients, the security index has the highest weighting coefficient.

[0009] Furthermore, the grades include five levels: excellent, good, satisfactory, and unsatisfactory.

[0010] This invention also provides a system for evaluating the rendering quality of an SR interface. The system includes: a data acquisition module for embedding lightweight data tracking logic to obtain common features of different rendering engines and for acquiring perception data through an intelligent driving domain controller; a synchronization module connected to the data acquisition module for spatiotemporal alignment of common features and perception data; and a processing module connected to the synchronization module for calculating core evaluation indicators based on the aligned data, setting a weight coefficient vector W=[w1,w2,...,w5] for each indicator, and calculating a comprehensive score S through normalization and weighted summation: S=∑(Metrici×Wi). Based on the comprehensive score S, the SR rendering quality is divided into several levels, and a visual evaluation report containing a radar chart and analysis of the reasons for the score loss is generated. Here, Metrici is the normalized score of the core evaluation indicator, i represents all evaluation dimensions, and Wi is the preset weight coefficient corresponding to the i-th indicator.

[0011] Furthermore, among the weighting coefficients, the security index has the highest weighting coefficient.

[0012] Furthermore, the grades include five levels: excellent, good, satisfactory, and unsatisfactory.

[0013] The present invention also provides a computer-readable storage medium storing a computer program, which, when executed by a processor, implements the steps of the above-described method.

[0014] This invention provides a method, system, and medium for evaluating the rendering quality of SR interfaces. It can obtain structured data from the underlying layer of the rendering pipeline and perform high-precision evaluation in conjunction with the spatiotemporal reference of the whole vehicle, so as to solve the compatibility problem and the lack of safety assessment of existing evaluation tools. Attached Figure Description

[0015] Figure 1 A flowchart of a method for evaluating the rendering quality of an SR interface provided by the present invention; Figure 2 A flowchart of a data acquisition method provided by the present invention; Figure 3 This is a flowchart of a spatiotemporal alignment method provided by the present invention; Figure 4 This is a flowchart of a method for calculating core evaluation indicators provided by the present invention. Detailed Implementation

[0016] The exemplary embodiments of this disclosure are described below with reference to the accompanying drawings, including various details of the embodiments to aid understanding, and should be considered merely exemplary. Therefore, those skilled in the art will recognize that various changes and modifications can be made to the embodiments described herein without departing from the scope and spirit of this disclosure. Similarly, for clarity and brevity, descriptions of well-known functions and structures are omitted in the following description.

[0017] Example 1: This invention provides a method, system, and medium for evaluating the rendering quality of an SR (Rendering Style Sheet) interface, applicable to intelligent driving vehicles. The system includes a data acquisition module, a synchronization module, and a processing module. Figure 1 As shown, the method includes: Step 1: Implant lightweight event tracking logic to obtain common features of different rendering engines, and obtain perception data through the intelligent driving domain controller; The data acquisition module is used to embed lightweight tracking logic to obtain common features of different rendering engines, and to acquire perception data through the intelligent driving domain controller.

[0018] Step 2: Spatiotemporally align common features and perceived data; The synchronization module, connected to the acquisition module, is used to align common features and perceived data in time and space.

[0019] Step 3: Based on the aligned data, calculate the core evaluation metrics, including: real-time performance, geometric accuracy, stability, model expressiveness, and security. Step 4: Set the weight coefficient vector W=[w1,w2,...,w5] for each indicator, and calculate the comprehensive score S through normalization and weighted summation: S=∑(Metrici×Wi). Based on the comprehensive score S, divide the SR rendering quality into several levels and generate a visual evaluation report containing radar charts and analysis of the reasons for the failures. Here, is the normalized score of the core evaluation indicator, i is all evaluation dimensions, and Wi is the preset weight coefficient corresponding to the i-th indicator.

[0020] The processing module, connected to the synchronization module, calculates the core evaluation indicators based on the aligned data, sets the weight coefficient vector W=[w1,w2,...,w5] for each indicator, and calculates the comprehensive score S through normalization and weighted summation: S=∑(Metrici×Wi). Based on the comprehensive score S, the SR rendering quality is divided into several levels, and a visual evaluation report containing radar charts and analysis of the reasons for the score loss is generated. Here, Metrici is the normalized score of the core evaluation indicator, i is all evaluation dimensions, and Wi is the preset weight coefficient corresponding to the i-th indicator.

[0021] Based on the same inventive concept as the above method, this application also provides a computer-readable storage medium storing a computer program, which, when executed by a processor, implements the steps of the method for evaluating the rendering quality of an SR interface.

[0022] This invention provides a method, system, and medium for evaluating the rendering quality of the SR interface. This technical solution defines a general interception timing and supporting data structure before the rendering command is submitted to the GPU, and combines it with the vehicle's PTP / gPTP high-precision clock synchronization to achieve a full-link, quantitative evaluation of the real-time performance, geometric accuracy, stability, and safety of SR rendering, thereby unifying industry standards and improving the safety and reliability of intelligent driving systems.

[0023] Example 2: This invention provides a method for evaluating the rendering quality of an SR (Simultaneous Rendering) interface. The core technical concept of this method lies in intercepting and extracting structured data of the SR scene at a predetermined time before the graphics processing unit (GPU) receives rendering instructions at the end of the rendering pipeline, and then performing cross-domain spatiotemporal alignment and calculation with the perception data and high-precision ground truth data of the intelligent driving domain controller. For example... Figure 1 As shown, the method includes: Step 1: Implant lightweight event tracking logic to obtain common features of different rendering engines, and obtain perception data through the intelligent driving domain controller; In the rendering process of the in-vehicle intelligent driving app, lightweight event tracking logic is embedded. This logic abstracts the common features of different rendering engines and does not rely on specific script syntax. For example... Figure 2 As shown, step 1 includes: Step 11: Preset the logical node at the end of the rendering pipeline and before GPU submission as the time to collect data, so as to ensure that the collected data is the deterministic state data that is finally presented on the screen, and eliminate the jitter caused by asynchronous calculation. Define a logical node where data acquisition occurs at the end of the rendering pipeline and before GPU submission. This node is characterized by the completion of physics calculations, animation blending, and scene graph traversal, and is about to undergo rasterization. For Unity-based implementations: utilize the `LateUpdate()` or `OnRenderObject()` lifecycle functions. For Unreal Engine-based implementations: utilize the `PreRender` proxy in the Render Thread or callback hooks before `DrawCall` submission. For self-developed C++ rendering engines: inject acquisition code at the node after scene graph traversal ends and before memory buffer swapping. Purpose: To ensure that the acquired data is the deterministic state data ultimately presented on the screen, eliminating jitter caused by asynchronous calculations.

[0024] Step 12: Generate a rendering submission timestamp using a high-precision timer, and synchronize it with the master clock via the vehicle Ethernet to ensure that the time base of all electronic control units is consistent. The system invokes a high-precision timer at the operating system or engine level (such as Unity's Time.realtimeSinceStartupAsDouble, Windows' QPC, or Linux's Clock_gettime) to generate a render submission timestamp (Trender). This timestamp is synchronized with the vehicle's PTP / gPTP (IEEE 802.1AS) master clock via the vehicle's Ethernet to ensure that all electronic control units (ECUs) have a consistent time base.

[0025] Step 13: Traverse the active nodes of the current SR scene and extract standardized attributes that are independent of the rendering engine, including: model identifier, world coordinates, orientation angle and visibility status. Traverse the active nodes in the current SR scene and extract standardized attributes independent of the rendering engine: Model ID: Extract the unique identifier or resource name of the node (e.g., Car_Sedan_01) for model diversity evaluation; World Position: Extract the 3D coordinates (Vector3) of the node in the rendering engine's world coordinate system for position accuracy calculation; Orientation / Yaw: Extract the rotation angle (Eulerian angle Yaw) of the node about the vertical axis of the world coordinate system for heading accuracy calculation; Visibility: Determine whether the target should be rendered based on the frustum culling results or the node's activation state.

[0026] Step 14: Serialize the structured data of standardized attributes into an engine-independent lightweight binary format, and send it as a common feature to the in-vehicle log server via shared memory or UDP protocol to avoid blocking the main rendering thread.

[0027] The standardized attributes are structured data, which is serialized into a lightweight binary format (such as Protobuf or FlatBuffers) that is independent of the engine, and sent to the in-vehicle log server via shared memory or UDP protocol to avoid blocking the main rendering thread.

[0028] Step 2: Spatiotemporally align common features and perceived data; like Figure 3 As shown, step 2 includes: Step 21: Obtain perception data containing perception timestamps and high-precision RTK-GPS / inertial measurement unit true data containing true timestamps through the intelligent driving domain controller; Step 22: Using the rendering submission timestamp as the base timestamp, the perceptual data and ground truth data are aligned to the same time using a linear interpolation algorithm; Steps 21 and 22 involve time alignment of multi-source data: using the Trender as the reference timestamp, a linear interpolation algorithm is employed to align the perception data (including the perception timestamp Tperception) output by the intelligent driving domain controller and the high-precision RTK-GPS / IMU (Inertial Measurement Unit) ground truth data (including the ground truth timestamp Tgt) to the same moment. Here, Trender represents the rendering submission timestamp, i.e., the timestamp generated by the tracking script at a predetermined time before the rendering pipeline submits the rendering instructions to the GPU, representing the submission time of the final state of the current frame data in the SR image; Tperception represents the perception output timestamp, i.e., the timestamp of the intelligent driving domain controller (ADC) outputting the corresponding target perception result, representing the generation time of the environmental perception data at the source end. Both Trender and Tperception are synchronized with the vehicle's master clock via the PTP / gPTP protocol, ensuring they are on the same time reference, thus enabling precise alignment of the perception data and rendering data to the same moment through the interpolation algorithm. Furthermore, RTK positioning technology is a real-time dynamic positioning technology based on carrier phase observations. It can provide real-time three-dimensional positioning results of the measurement station in a specified coordinate system, achieving centimeter-level accuracy. In RTK operation mode, the base station transmits its observations and station coordinates to the rover via a data link. Therefore, high-precision GPS measurements must utilize carrier phase observations.

[0029] Step 23: Calibrate the extrinsic parameter matrix, establish a universal coordinate transformation model, and transform the world coordinates of the rendering engine to the UTM coordinate system or vehicle coordinate system consistent with the RTK true value, eliminating errors caused by different installation positions and coordinate system definitions.

[0030] This step involves spatial alignment of multi-source data: establishing a universal coordinate transformation model. The world coordinates of the rendering engine are transformed to the UTM coordinate system or vehicle coordinate system (VCS) consistent with the RTK truth using a pre-calibrated extrinsic matrix (including rotation and translation), eliminating errors caused by different installation locations and coordinate system definitions.

[0031] Step 3: Based on the aligned data, calculate the core evaluation metrics, including: real-time performance, geometric accuracy, stability, model expressiveness, and security. Based on the aligned data, the following core evaluation metrics were calculated, as shown in Table 1:

[0032] Table 1 Core Evaluation Indicators like Figure 4 As shown, step 3 includes: Step 31: Calculate the end-to-end rendering latency and calculate the standard deviation of the rendering submission time error interval between consecutive frames to reflect the system response speed and the stability of the rendering thread. Step 32: Calculate the deviation between the converted rendered coordinates / heading and the ground truth data, as well as the deviation between the rendered relative distance between the main vehicle and the target vehicle and the ground truth relative distance; Step 33: Count the number of times the change in target position or heading angle exceeds a preset threshold per unit time, and evaluate the smoothness of the perception fusion output. Step 34: Calculate the ratio of the number of different Model IDs rendered within the test period to the total number of preset model libraries. This reflects the richness of rendering resources. Compare the consistency between the target classification in the perception data and the rendered Model IDs to reflect the accuracy of the mapping from perception results to rendering objects. Step 35: In the AEB / FCW trigger critical zone, calculate the critical target loss rate = number of lost frames / total number of frames in the dangerous scene. When the relative distance Dsr displayed by SR rendering is greater than the actual relative distance Dreal, and the difference exceeds the preset safety threshold, it is judged as a high-risk error.

[0033] AEB and FCW are two active safety technologies in automobiles. FCW (Forward Collision Warning) uses radar or cameras to monitor the distance and speed of vehicles ahead, assess the risk of a pre-collision, and warn or alert the driver. AEB (Autonomous Emergency Braking) is an upgraded version of FCW. When a collision is detected as imminent and the driver does not react, the system automatically triggers the brakes, even bringing the vehicle to a complete stop, to avoid or mitigate the collision. AEB comes in several types, including partial automatic emergency braking (standard and enhanced) and medium-speed automatic emergency braking (AEB-M). FCW and AEB are typically used in tandem; FCW provides a warning first, and if the driver does not react, AEB automatically applies the brakes. It is important to note that these systems are driver assistance tools and cannot completely replace the driver's attention and judgment. The driver remains fully responsible for maintaining control of the vehicle and for controlling its speed and distance from other vehicles.

[0034] Step 4: Set the weight coefficient vector W=[w1,w2,...,w5] for each indicator, and calculate the comprehensive score S through normalization and weighted summation: S=∑(Metrici×Wi). Based on the comprehensive score S, divide the SR rendering quality into several levels and generate a visual evaluation report containing radar charts and analysis of the reasons for the failures. Here, is the normalized score of the core evaluation indicator, i is all evaluation dimensions, and Wi is the preset weight coefficient corresponding to the i-th indicator.

[0035] A weighted coefficient vector W = [w1, w2, ..., w5] is set for each indicator, with the safety indicator w5 having the highest weight. The comprehensive score S is calculated through normalization and weighted summation: S = ∑(Metrici × Wi), where Metrici represents the normalized score of the i-th quantitative evaluation indicator calculated in step S3, and i iterates through all evaluation dimensions, including but not limited to: end-to-end rendering latency score, position accuracy score, heading accuracy score, position jump rate score, heading jump rate score, model diversity score, and safety assessment indicator score (including key target loss rate score and misleading safety error score); Wi represents the preset weight coefficient corresponding to the i-th indicator, where the weight W of the safety assessment indicator is higher than the weights of other technical indicators to reflect the safety-oriented evaluation principle. The larger the comprehensive score S, the higher the SR rendering quality. Based on the comprehensive score S, the SR rendering quality is divided into five levels: S (Excellent), A (Good), B (Good), C (Acceptable), and D (Unacceptable), and a visual evaluation report including a radar chart and analysis of the reasons for the score loss is generated.

[0036] For example, Example 1: Implementation based on the Unity engine (preferred embodiment). Create an SRQualityMonitor.cs script in the Unity project and attach it to the SR Camera. Collect data using the LateUpdate() event. Code logic: Iterate through the list of dynamic objects maintained by SRManager, reading transform.position and transform.eulerAngles.y. Use GeometryUtility.TestPlanesAABB to determine if the target is within the view frustum. Timestamp: Use Time.realtimeSinceStartupAsDouble to obtain high-precision time and synchronize with the PTP master clock via the Socket interface. Transmission: Use UDPClient to send Protobuf format data to the log server to avoid blocking the main thread. Example 2: Implementation based on the Unreal Engine. In an Unreal Engine C++ project, obtain the scene Actor list before the rendering thread refreshes by inheriting UGameViewportClient or registering the FSlateApplication::Get().OnPreTick() hook. World coordinates and orientation are obtained using GetActorLocation() and GetActorRotation(), and high-precision timestamps are obtained using FPlatformTime::Seconds(). The data structure and transmission method are consistent with Example 1 to ensure the universality of the backend analysis logic. Example 3: A general alternative solution based on external acquisition (non-intrusive). When the source code cannot be obtained (such as when a third-party evaluation agency evaluates a competitor's car infotainment system), the alternative implementation method of this solution is adopted: HDMI / DisplayPort Frame Grabber (such as AUMO A16) is used to bypass and acquire the rendered output frame, and the frame arrival time is marked using the hardware PTP timestamp of the acquisition card. The target box in the image is identified by combining computer vision algorithms (YOLO+DeepSORT), and the approximate position is calculated. Although the accuracy is lower than that of Examples 1 and 2, it still strictly follows the indicator system (latency, jump rate, security) and calculation method proposed in this invention, realizing the unification of evaluation standards.

[0037] In summary, this invention boasts a highly adaptable architecture: by abstracting the universal interception timing of the rendering pipeline, this solution is no longer limited to the Unity engine and can seamlessly adapt to Unreal Engine and automotive manufacturers' self-developed rendering engines, solving the problem of poor universality of evaluation tools and expanding the scope of patent protection. High data fidelity: By directly acquiring the engine's underlying Transform data, it avoids perspective and image recognition errors caused by traditional screen photography, achieving centimeter-level positional accuracy assessment. Clear safety orientation: It pioneers the "critical target loss rate" and "misleading safety error" indicators, elevating SR rendering evaluation from the UI experience level to the level of functional safety auxiliary verification, filling an industry gap. End-to-end traceability: Microsecond-level PTP alignment enables precise separation of the time consumed in each stage of perception, transmission, and rendering, facilitating agile development and performance optimization.

[0038] The specific embodiments described above do not constitute a limitation on the scope of protection of this disclosure. Those skilled in the art should understand that various modifications, combinations, sub-combinations, and substitutions can be made according to design requirements and other factors. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of this disclosure should be included within the scope of protection of this disclosure.

Claims

1. A method for evaluating the rendering quality of an SR (Rendering Style) interface, applied to intelligent driving vehicles, characterized in that, The method includes: Step 1: Implant lightweight event tracking logic to obtain common features of different rendering engines, and obtain perception data through the intelligent driving domain controller; Step 2: Spatiotemporally align common features and perceived data; Step 3: Based on the aligned data, calculate the core evaluation metrics, including: real-time performance, geometric accuracy, stability, model expressiveness, and security. Step 4: Set the weight coefficient vector W=[w1,w2,...,w5] for each indicator, and calculate the comprehensive score S through normalization and weighted summation: S=∑(Metrici×Wi). Based on the comprehensive score S, divide the SR rendering quality into several levels and generate a visual evaluation report containing radar charts and analysis of the reasons for the failures. Here, is the normalized score of the core evaluation indicator, i is all evaluation dimensions, and Wi is the preset weight coefficient corresponding to the i-th indicator.

2. The method for evaluating the rendering quality of an SR interface according to claim 1, characterized in that, Step 1 includes: Step 11: Preset the logical node at the end of the rendering pipeline and before GPU submission as the time to collect data, so as to ensure that the collected data is the deterministic state data that is finally presented on the screen, and eliminate the jitter caused by asynchronous calculation. Step 12: Generate a rendering submission timestamp using a high-precision timer, and synchronize it with the master clock via the vehicle Ethernet to ensure that the time base of all electronic control units is consistent. Step 13: Traverse the active nodes of the current SR scene and extract standardized attributes that are independent of the rendering engine, including: model identifier, world coordinates, orientation angle and visibility status. Step 14: Serialize the structured data of standardized attributes into an engine-independent lightweight binary format, and send it as a common feature to the in-vehicle log server via shared memory or UDP protocol to avoid blocking the main rendering thread.

3. The method for evaluating the rendering quality of an SR interface according to claim 2, characterized in that, Step 2 includes: Step 21: Obtain perception data containing perception timestamps and high-precision RTK-GPS / inertial measurement unit true data containing true timestamps through the intelligent driving domain controller; Step 22: Using the rendering submission timestamp as the base timestamp, the perceptual data and ground truth data are aligned to the same time using a linear interpolation algorithm; Step 23: Calibrate the extrinsic parameter matrix, establish a universal coordinate transformation model, and transform the world coordinates of the rendering engine to the UTM coordinate system or vehicle coordinate system consistent with the RTK true value, eliminating errors caused by different installation positions and coordinate system definitions.

4. The method for evaluating the rendering quality of an SR interface according to claim 1, characterized in that, Step 3 includes: Step 31: Calculate the end-to-end rendering latency and calculate the standard deviation of the rendering submission time error interval between consecutive frames to reflect the system response speed and the stability of the rendering thread. Step 32: Calculate the deviation between the converted rendered coordinates / heading and the ground truth data, as well as the deviation between the rendered relative distance between the main vehicle and the target vehicle and the ground truth relative distance; Step 33: Count the number of times the change in target position or heading angle exceeds a preset threshold per unit time, and evaluate the smoothness of the perception fusion output. Step 34: Calculate the ratio of the number of different Model IDs rendered within the test period to the total number of preset model libraries. This reflects the richness of rendering resources. Compare the consistency between the target classification in the perception data and the rendered Model IDs to reflect the accuracy of the mapping from perception results to rendering objects. Step 35: In the AEB / FCW trigger critical zone, calculate the critical target loss rate = number of lost frames / total number of frames in the dangerous scene. When the relative distance Dsr displayed by SR rendering is greater than the actual relative distance Dreal, and the difference exceeds the preset safety threshold, it is judged as a high-risk error.

5. The method for evaluating the rendering quality of an SR interface according to claim 1, characterized in that, Among the weighting coefficients, the security index has the highest weighting coefficient.

6. The method for evaluating the rendering quality of an SR interface according to claim 1, characterized in that, The rating system includes five levels: Excellent, Good, Fair, Satisfactory, and Unsatisfactory.

7. A system for implementing the method for evaluating the rendering quality of an SR interface as described in any one of claims 1-6, characterized in that, The system includes: The data acquisition module is used to embed lightweight tracking logic to obtain common features of different rendering engines, and to acquire perception data through the intelligent driving domain controller. The synchronization module, connected to the acquisition module, is used to align common features and perceived data in time and space. The processing module, connected to the synchronization module, calculates the core evaluation indicators based on the aligned data, sets the weight coefficient vector W=[w1,w2,...,w5] for each indicator, and calculates the comprehensive score S through normalization and weighted summation: S=∑(Metrici×Wi). Based on the comprehensive score S, the SR rendering quality is divided into several levels, and a visual evaluation report containing radar charts and analysis of the reasons for the score loss is generated. Here, Metrici is the normalized score of the core evaluation indicator, i is all evaluation dimensions, and Wi is the preset weight coefficient corresponding to the i-th indicator.

8. The system for evaluating the rendering quality of an SR interface according to claim 7, characterized in that, Among the weighting coefficients, the security index has the highest weighting coefficient.

9. The system for evaluating the rendering quality of an SR interface according to claim 7, characterized in that, The rating system includes five levels: Excellent, Good, Fair, Satisfactory, and Unsatisfactory.

10. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a computer program that, when executed by a processor, implements the steps of the method according to any one of claims 1 to 6.

Images