Data evaluation method, electronic device, and storage medium

Abstract

The application discloses a data evaluation method, an electronic device and a storage medium. The method comprises the following steps: obtaining target image data and target mask data corresponding to a repositioning track; performing projection mapping by using the target image data to obtain projection image data; calculating by using the target mask data to construct a weight constraint function; calculating a feature error value between the projection image data and the target mask data based on the projection image data, the target mask data and the weight constraint function, and then judging whether the repositioning data corresponding to the repositioning track is qualified. The application improves the efficiency and reliability of the quality evaluation of the repositioning data.

Description

Technical Field

[0001] This application relates to the field of autonomous driving technology, and in particular to a data evaluation method, electronic device and storage medium. Background Technology

[0002] In fields such as robotics and autonomous driving, the quality of the pose output by a visual positioning system directly determines the reliability of navigation and mapping tasks. Therefore, automated quality inspection and evaluation of positioning results are necessary. Current mainstream positioning quality inspection methods typically project map features onto an image and use the Intersection over Union (IoU) and centroid distance to measure the degree of overlap between the projected area and the image area to determine positioning quality. However, this method relies solely on a single matching metric and does not address the identification and processing of invalid or false-matching regions in complex environments. Affected by factors such as scene occlusion, dynamic interference, and false features, projection matching calculations are easily influenced by false regions, resulting in significant random errors and a high false detection rate, making it unreliable for assessing positioning quality in complex scenarios. Summary of the Invention

[0003] This application provides at least one data evaluation method, electronic device, and storage medium.

[0004] The first aspect of this application provides a data evaluation method, comprising: acquiring target image data and target mask data corresponding to a repositioning trajectory; performing projection mapping using the target image data to obtain projected image data; constructing a weight constraint function using the target mask data; calculating the feature error value between the projected image data and the target mask data based on the projected image data, the target mask data, and the weight constraint function, thereby determining whether the repositioning data corresponding to the repositioning trajectory is qualified.

[0005] In some embodiments, the step of using the target mask data to calculate and construct a weight constraint function includes: calculating the semantic confidence of the target mask data based on camera parameters and the target mask data; and constructing a weight constraint function of the target mask data based on the semantic confidence.

[0006] In some embodiments, the target mask data includes semantic geometric surfaces; calculating the semantic confidence of the target mask data based on camera parameters and the target mask data includes: calculating the distance from the semantic geometric surface to the camera; determining the semantic confidence based on the distance and the semantic area of ​​the semantic geometric surface; wherein, when the semantic area is less than a preset area, the semantic confidence is 0; when the semantic area is greater than the preset area, the semantic confidence is the ratio of the semantic area to a preset value, and the preset value is the sum of the distance and 1.

[0007] In some embodiments, the method further includes: in response to continuous features in the semantic geometry representation map, calculating the average confidence of all pixels of the continuous features within the target mask data as the semantic confidence.

[0008] In some embodiments, constructing the weight constraint function for the target mask data based on the semantic confidence includes: in response to the semantic confidence being less than or equal to a first preset confidence, the weight constraint function is 0; in response to the semantic confidence being less than a second preset confidence and greater than the first preset confidence, the weight constraint function is the difference between the semantic confidence and the first preset confidence; in response to the semantic confidence being greater than or equal to the second preset confidence, the weight constraint function is 1.

[0009] In some embodiments, calculating the feature error value between the projected image data and the target mask data based on the projected image data, the target mask data, and the weight constraint function includes: sampling the projected image data to obtain projected feature points; determining semantic feature points adjacent to the projected feature points in the target mask data; calculating the geometric distance between the projected feature points and the semantic feature points; and using the product of the weight constraint function and the geometric distance as the feature error value.

[0010] In some embodiments, obtaining target image data corresponding to the repositioning trajectory includes: obtaining the vehicle speed and wheel speed corresponding to the trajectory point in the repositioning trajectory; in response to the vehicle speed being greater than a first threshold and the wheel speed being less than a second threshold within a preset time period, determining that the initial image data of the trajectory point at the current moment is abnormal data, wherein the current moment is located in the middle of the preset time period; and removing the initial image data of the trajectory point at the current moment to obtain the target image data.

[0011] In some embodiments, obtaining target mask data corresponding to the relocation trajectory includes: constructing a region of interest bounding box centered on the trajectory point, wherein the length direction of the region of interest bounding box is determined based on the trajectory forward direction; and cropping the mask data corresponding to the target image data based on the region of interest bounding box to obtain the target mask data.

[0012] In some embodiments, the step of using the target image data to perform projection mapping to obtain projected image data includes: based on camera parameters, projecting and rendering vector features in the high-precision map onto the target image data, thereby obtaining the projected image data.

[0013] In some embodiments, in response to the vector feature being a lane line, a weighted average error between the projected image data and the corresponding target mask data is obtained as the feature error value; in response to the vector feature being an arrow or a stop line, the intersection-union ratio between the projected image data and the corresponding target mask data is obtained as the feature error value.

[0014] In some embodiments, the relocation data includes image-level data, frame-level data, and packet-level data. The method further includes: determining whether the image-level data is qualified based on the pass rate of semantic objects in the image-level data; determining whether the frame-level data is qualified based on the pass rate of multiple viewpoint images in the frame-level data; and determining whether the packet-level data is qualified based on the pass rate of multiple frames in the packet-level data.

[0015] A second aspect of this application provides an electronic device including a memory and a processor coupled to each other, the processor being configured to execute program instructions stored in the memory to implement the data evaluation method described above.

[0016] A third aspect of this application provides a computer-readable storage medium having program instructions stored thereon, which, when executed by a processor, implement the data evaluation method described above.

[0017] The beneficial effects of this application are as follows: It acquires the target image data and target mask data corresponding to the repositioning trajectory; performs projection mapping using the target image data to obtain projected image data; calculates and constructs a weight constraint function using the target mask data; and calculates the feature error value between the projected image data and the target mask data based on the projected image data, target mask data, and weight constraint function, thereby determining whether the repositioning data corresponding to the repositioning trajectory is qualified. Specifically, calculating the feature error value between the projected image data and the target mask data using the projected image data, target mask data, and weight constraint function improves the efficiency of quality assessment of the repositioning data and ensures the reliability of the repositioning data output.

[0018] It should be understood that the above general description and the following detailed description are exemplary and explanatory only, and are not intended to limit this application. Attached Figure Description

[0019] The accompanying drawings, which are incorporated in and constitute a part of this specification, illustrate embodiments consistent with this application and, together with the specification, serve to explain the technical solutions of this application.

[0020] Figure 1 This is a flowchart illustrating the data evaluation method according to an embodiment of this application; Figure 2 This is a schematic diagram of the structure of the data evaluation device according to an embodiment of this application; Figure 3 This is a schematic diagram of the structure of an electronic device according to an embodiment of this application; Figure 4 This is a schematic diagram of the structure of a non-volatile computer-readable storage medium according to an embodiment of this application. Detailed Implementation

[0021] In this application, the reference to "embodiment" means that a specific feature, structure, or characteristic described in connection with an embodiment may be included in at least one embodiment of this application. The appearance of this phrase in various places throughout the specification does not necessarily refer to the same embodiment, nor is it a separate or alternative embodiment mutually exclusive with other embodiments. It will be explicitly and implicitly understood by those skilled in the art that the embodiments described herein can be combined with other embodiments.

[0022] In this document, the term "and / or" is merely a description of the relationship between related objects, indicating that three relationships can exist. For example, A and / or B can represent: A existing alone, A and B existing simultaneously, or B existing alone. Additionally, the character " / " in this document generally indicates that the preceding and following related objects are in an "or" relationship. Furthermore, "many" in this document means two or more. Additionally, the term "at least one" in this document means any combination of at least two of any one or more of a plurality of objects. For example, including at least one of A, B, and C can mean including any one or more elements selected from the set consisting of A, B, and C. Furthermore, the terms "first," "second," and "third" in this application are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of indicated technical features.

[0023] To enable those skilled in the art to better understand the technical solution of this application, the technical solution of this application will be further described in detail below with reference to the accompanying drawings and specific embodiments.

[0024] Please see Figure 1 , Figure 1 This is a flowchart illustrating the data evaluation method according to an embodiment of this application. The subject executing this method can be an electronic device with computing capabilities, such as a microcomputer, a server, and mobile devices such as laptops and tablets.

[0025] It should be noted that if substantially the same result is obtained, the method of this application is not based on... Figure 1 The sequence of processes shown is limited.

[0026] In some possible implementations, this method can be implemented by the processor calling computer-readable instructions stored in memory, such as... Figure 1 As shown, the method may include the following steps: Step S11: Obtain the target image data and target mask data corresponding to the relocation trajectory.

[0027] A relocalization trajectory is a pose sequence formed by arranging multiple relocalization trajectory points in chronological order. The target image data can be a keyframe image, and the target mask data can be a semantic segmentation mask of the keyframe image, where the semantic segmentation mask carries pixel-level confidence.

[0028] Step S12: Perform projection mapping using the target image data to obtain projected image data.

[0029] Projection mapping is performed based on the target image data to obtain projected image data. For example, using the repositioned camera pose, the ground... Figure 3 After the dimensional features are projected and mapped onto the current image coordinate system, the projected image data is obtained. This projected image data includes the repositioned data information and the actual data information projected onto the image.

[0030] Step S13: Calculate using the target mask data to construct the weight constraint function.

[0031] The confidence level of the target mask data is calculated, and a weight constraint function is constructed based on this confidence level. This weight constraint function can include an error weight function and an evaluation weight function. For example, high-confidence regions can be assigned low error weights or high evaluation weights, while low-confidence regions can be assigned high error weights or low evaluation weights. The weight constraint function is used to apply weighted constraints to the evaluation metrics of the relocation data, reducing the impact of invalid regions on the relocation quality assessment.

[0032] Step S14: Based on the projected image data, target mask data and weight constraint function, calculate the feature error value between the projected image data and the target mask data, and then determine whether the relocation data corresponding to the relocation trajectory is qualified.

[0033] By using projected image data, target mask data, and a weight constraint function, the feature error value between the projected image data and the target mask data is calculated. Based on the magnitude of the feature error value, the quality of the repositioning data corresponding to the repositioning trajectory is determined. For example, if the feature error value between the projected image data and the target mask data is less than a preset value, the repositioning data corresponding to that trajectory is deemed qualified, meaning its quality meets the standard and has high reliability. If the feature error value between the projected image data and the target mask data is greater than a preset value, the repositioning data corresponding to that trajectory is deemed unqualified and cannot be used for subsequent operations, such as map building or training autonomous driving localization algorithms.

[0034] In this embodiment, target image data and target mask data corresponding to the repositioning trajectory are acquired. Projection mapping is performed using the target image data to obtain projected image data. Weight constraint function is constructed using the target mask data. Based on the projected image data, target mask data, and weight constraint function, the feature error value between the projected image data and the target mask data is calculated. This allows for the determination of whether the repositioning data corresponding to the repositioning trajectory is qualified. The use of projected image data, target mask data, and weight constraint function to calculate the feature error value between the projected image data and the target mask data improves the efficiency of quality assessment of the repositioning data and ensures the reliability of the repositioning data output.

[0035] Understandably, by constructing a weighted constraint function based on target mask data, the system can quantify the uncertainty of visual perception, allowing high-confidence semantic regions to dominate error calculation. This reduces the interference of low-quality perceptual noise on the localization evaluation results and improves the robustness of the evaluation system in complex environments. Using the weighted constraint function to weight the feature errors between the projected image and the target mask data suppresses the contribution of invalid or false-matching regions to the final error value, reducing random errors caused by scene occlusion, dynamic interference, or pseudo-features, and lowering the false detection rate. Calculating feature error values ​​based on the constructed weighted constraint function to determine the qualification of the initial relocalization data ensures that the quality inspection results truly reflect the core performance of the localization system, improving the accuracy and interpretability of the evaluation results.

[0036] In some embodiments, obtaining target image data corresponding to the repositioning trajectory includes: obtaining the vehicle speed and wheel speed corresponding to the trajectory point in the repositioning trajectory; in response to the vehicle speed being greater than a first threshold and the wheel speed being less than a second threshold within a preset time period, determining that the initial image data of the trajectory point at the current moment is abnormal data, and the current moment is located in the middle of the preset time period; removing the initial image data of the trajectory point at the current moment to obtain the target image data.

[0037] In one implementation, the vehicle speed and wheel speed corresponding to trajectory point P are obtained. In response to the fact that the vehicle speed of trajectory point P is greater than a first threshold and the wheel speed of trajectory point P is less than a second threshold within a preset time period T0-T1, the initial image data of trajectory point P at the current time Ti is determined to be abnormal data. The current time Ti is located in the middle of the preset time period T0-T1. Further, the initial image data of trajectory point P at the current time is removed from the initial image data corresponding to the repositioning trajectory to obtain the target image data.

[0038] For example, wheel speed data is collected at high frequency (sampling frequency ≥ 100Hz), and the wheel speed data is precisely matched with the repositioning trajectory points according to the timestamp. For any trajectory point P to be evaluated, the wheel speed sequence is extracted within a 5-second time window before and after the trajectory point P. The vehicle speed is the absolute longitudinal velocity V of the vehicle, and the wheel speed is the wheel speed VW.

[0039] In some examples, the first threshold is 0.3 km / h and the second threshold is -0.5 km / h. When V > 0.3 km / h and t(VW < -0.5 km / h) ≥ 500 ms, the vehicle is determined to be reversing or turning around in place. The corresponding data frame is marked as a motion abnormal frame. This frame data is only recorded and stored and is not used in the accuracy error calculation, thus effectively eliminating the negative impact of abnormal driving conditions on the overall quality inspection pass rate.

[0040] Understandably, acquiring the target image data corresponding to the relocalization trajectory includes removing initial image data that does not meet preset conditions. For example, a kinematic filtering mechanism combining wheel speed data can be used to eliminate pose jumps caused by non-localization factors (such as mechanical vibration or changes in operating conditions). Furthermore, in challenging long-tail scenarios such as parking, reversing, and U-turns, the system can still provide stable and reliable localization evaluations, significantly reducing the misjudgment rate caused by abnormal operating conditions and improving the system's generalization ability.

[0041] The initial image data can be obtained by acquiring a relocation data packet, which may include keyframe images, semantic segmentation mask (including pixel-level confidence C), relocation trajectory pose, camera intrinsic and extrinsic parameters and vehicle wheel speed data, high-precision map, etc.

[0042] In some embodiments, obtaining target mask data corresponding to the relocation trajectory includes: constructing a region of interest bounding box centered on the trajectory points, wherein the length direction of the region of interest bounding box is determined based on the trajectory forward direction; and cropping the mask data corresponding to the target image data based on the region of interest bounding box to obtain target mask data.

[0043] In one implementation, a Region of Interest (ROI) bounding box is constructed with trajectory point P as the center. The length direction of the ROI bounding box is determined based on the trajectory forward direction. Based on the ROI bounding box, the mask data corresponding to the target image data is dynamically clipped to obtain the target mask data.

[0044] For example, after obtaining the target image data, a dynamic Region of Interest (ROI) is constructed by extending a length L1 (calculated based on the application scenario or vehicle speed) along the forward direction, with a horizontal width covering L3 meters to the left and right of the trajectory centerline, centered on the trajectory point P. Based on this dynamic ROI, the semantic segmentation mask corresponding to the target image data is spatially cropped. By applying a dynamic cropping strategy, computational redundancy in invalid areas is significantly reduced, ensuring that the quality inspection and evaluation focus is highly targeted and specific to the actual vehicle passage area.

[0045] In some embodiments, projection mapping is performed using target image data to obtain projected image data, including: based on camera parameters, vector features in the high-precision map are projected and rendered onto the target image data to obtain projected image data.

[0046] The camera parameters include repositioning pose (containing translation vectors and rotation matrices) and standard camera intrinsic and extrinsic calibration parameters. Using these camera parameters, vector features from the high-definition map are projected and rendered onto the target image data, resulting in projected image data. Vector features in the high-definition map include lane lines, guide arrows, stop lines, etc.

[0047] For example, based on the repositioning pose (including translation vectors and rotation matrices) and precise camera intrinsic and extrinsic calibration parameters, a mapping relationship from the world coordinate system to the camera imaging plane is established. Furthermore, vector elements in the high-precision map, such as lane lines, guide arrows, and stop lines, are rendered onto the corresponding keyframe image plane through coordinate transformation and perspective projection calculations using their world coordinate point sets.

[0048] In some embodiments, the calculation of the target mask data to construct a weight constraint function includes: calculating the semantic confidence of the target mask data based on camera parameters and target mask data; and constructing the weight constraint function of the target mask data based on the semantic confidence.

[0049] Camera parameters include the coordinates of the camera's optical center. Based on the camera's optical center and the target mask data, the semantic confidence of the target mask data is calculated. Using semantic confidence Construct a weight constraint function W for the target mask data, for example, W=F( The weight constraint function is used to characterize the substantial contribution (relevance) of the target mask data to the final data evaluation criteria.

[0050] In some embodiments, the target mask data includes a semantic geometric surface; based on camera parameters and the target mask data, the semantic confidence of the target mask data is calculated, including: calculating the distance from the semantic geometric surface to the camera; determining the semantic confidence based on the distance and the semantic area of ​​the semantic geometric surface; wherein, when the semantic area is less than a preset area, the semantic confidence is 0; when the semantic area is greater than the preset area, the semantic confidence is the ratio of the semantic area to a preset value, where the preset value is the sum of the distance and 1.

[0051] For example, semantic confidence is The semantic area of ​​the semantic geometry surface is The coordinates of the center point of the semantic geometry surface are (x1, y1), the coordinates of the camera's optical center are (x2, y2), the distance from the semantic geometry surface to the camera is d, and the preset area is... ,in:

[0052]

[0053] Preset area It can be determined based on the actual application scenario, and .

[0054] When the semantic area is greater than the preset area, the semantic confidence is the ratio of the semantic area to the preset value. The preset value is the sum of the distance and 1. Adding 1 to the denominator can prevent the confidence from diverging due to the distance being too close.

[0055] In some embodiments, in response to continuous features in a semantic geometry surface representing a map, the average confidence of the continuous features across all pixels within the target mask data is calculated as the semantic confidence.

[0056] Continuous features in a map can refer to long strip features such as lane lines. For such areal continuous features, the average confidence of all pixels of the continuous feature within the target mask data can be calculated as the semantic confidence of any trajectory point in the corresponding semantic geometry surface.

[0057] For example, for elongated features such as lane lines, the system calculates the average confidence score of all pixels within the entire lane line mask and uses this as the unified weight for all matching points in that instance. This global calculation method can more realistically reflect the perceptual integrity of elongated features and avoid misjudging the whole from a local perspective.

[0058] In other words, for long, continuous features, using an instance-level weighting strategy can prevent high confidence in certain areas from masking low quality issues caused by occlusion along the entire line.

[0059] In some embodiments, constructing a weight constraint function for the target mask data based on semantic confidence includes: in response to semantic confidence being less than or equal to a first preset confidence, the weight constraint function is 0; in response to semantic confidence being less than a second preset confidence and greater than the first preset confidence, the weight constraint function is the difference between semantic confidence and the first preset confidence; in response to semantic confidence being greater than or equal to the second preset confidence, the weight constraint function is 1.

[0060] For example, semantic confidence is The first pre-set reliability is The second pre-set reliability is Weight constraint function as follows:

[0061] Among them, when the confidence level is lower than the low threshold, A value of 0 indicates that the low-confidence region is not included in the evaluation. When the confidence level is in the middle range, the weight value will increase linearly with the confidence level. When the confidence level is higher than the high threshold, A value of 1 indicates that high-confidence regions are assigned full weight. Through piecewise function design, data from regions where perceived quality meets a certain standard only contribute substantially to the final error evaluation.

[0062] In some embodiments, the feature error value between the projected image data and the target mask data is calculated based on the projected image data, the target mask data, and the weight constraint function, including: sampling the projected image data to obtain projected feature points; determining semantic feature points adjacent to the projected feature points in the target mask data; calculating the geometric distance between the projected feature points and the semantic feature points; and using the product of the weight constraint function and the geometric distance as the feature error value.

[0063] In one implementation, after constructing the weight constraint function for the target mask data, the projected image data is sampled to obtain projected feature points P1. Semantic feature points M1 adjacent to the projected feature points are determined in the target mask data. The geometric distance E between the projected feature points P1 and M1 is calculated. The weight constraint function is then further... The product of the geometric distance E and the geometric distance E is used as the feature error value E'.

[0064] For example, after spatial alignment and weight establishment, feature point matching and error calculation are performed. Map projection lines are uniformly sampled within the projection scale range [0.05, 0.95] to obtain map point P1 (i.e., the projected feature point), and the semantic feature point M1 closest to map point P1 is searched within the semantic segmentation Mask region. The geometric error between the two is calculated as E = ||P1||. M1||, and then the weighted error E'= ·E, to obtain a feature error value E' between the projected image data and the target mask data.

[0065] In some embodiments, in response to the vector feature being a lane line, a weighted average error between the projected image data and the target mask data is obtained as the feature error value; in response to the vector feature being an arrow or a stop line, an intersection over union between the projected image data and the target mask data is obtained as the feature error value.

[0066] For example, if the projected map vector feature is a lane line, the weighted average error between the lane line pixels in the projected image data and the mask data is calculated for final judgment to evaluate whether the data is qualified. If the projected map vector feature is an arrow / stop line, the intersection over union between the arrow / stop line projection in the projected image data and the mask data is calculated, and the final judgment is made in combination with the centroid distance to evaluate whether the data is qualified, so as to implement differentiated judgment logic for different types of high-precision map elements.

[0067] In some examples, for special scenarios, three cases of "map exists but no semantics", "semantics exists but no map", and "both exist but do not match" can be further distinguished, and labels of "qualified", "map expired" or "unqualified" are directly assigned respectively, which also implements differentiated judgment logic for different types of high-precision map elements.

[0068] In some embodiments, the repositioning data includes image-level data, frame-level data, and packet-level data, and the method further comprises: judging whether the image-level data is qualified based on the qualification rate of semantic objects in the image-level data; judging whether the frame-level data is qualified based on the qualification rate of multiple perspective images in the frame-level data; judging whether the packet-level data is qualified based on the qualification rate of multiple frames in the packet-level data.

[0069] For the image-level data, for example, for a single key frame image, the qualification rate of its internal semantic objects is counted to judge whether the data is qualified. For the frame-level data, for example, for the vehicle state at a certain moment, when all the single-frame six-perspective images corresponding to the moment are determined to be image-level qualified, the frame is determined to be frame qualified. For the packet-level data, for example, for the entire repositioning collection data packet, the proportion of qualified frames is counted. When frames accounting for 80% or more in the data packet are determined to be qualified, the repositioning collection data packet can be determined to pass the quality inspection.

[0070] as Figure 2 shown, Figure 2 is a schematic structural diagram of a data evaluation apparatus according to an embodiment of the present application. The data evaluation apparatus 20 includes a data preprocessing module 21, a projection mapping module 22, a constraint construction module 23, and an error calculation module 24.

[0071] The data preprocessing module 21 is configured to acquire the target image data and target mask data corresponding to the relocation trajectory. The projection mapping module 22 is configured to perform projection mapping using the target image data to obtain projected image data. The constraint construction module 23 is configured to use the target mask data to calculate and construct a weight constraint function. The error calculation module 24 is configured to calculate the feature error value between the projected image data and the target mask data based on the projected image data, the target mask data, and the weight constraint function, and then determine whether the relocation data corresponding to the relocation trajectory is qualified.

[0072] The data preprocessing module 21 can also be configured to acquire relocation data packets, including keyframe images, semantic segmentation masks (including pixel-level confidence C), relocation trajectory poses, camera intrinsic and extrinsic parameters and vehicle wheel speed data, and high-precision maps, and to align and group the data according to timestamps.

[0073] In one implementation, the semantic segmentation result of the current frame image is acquired by the visual perception module (not shown in the figure) within the data evaluation device 20, and the confidence score of each semantic element is extracted. Subsequently, these confidence scores are converted into corresponding error weight factors to construct a weighted error model. In the weighted error model, high-confidence regions (such as clear lane lines and stable curbs) are assigned lower error weights or higher evaluation weights, while low-confidence regions (such as blurred shadows and occluded traffic signs) are assigned higher error tolerance or lower evaluation weights. This allows high-confidence regions to dominate the final positioning evaluation result, thereby achieving the quantification and adaptive filtering of perceptual uncertainty and improving the robustness of the evaluation system.

[0074] In one implementation, a kinematic anomaly filtering mechanism is constructed using wheel speed data from the vehicle chassis to address pose change issues under complex operating conditions. Wheel speed signals are monitored in real time to identify whether the vehicle is reversing or making a U-turn. Upon detecting such non-standard driving states, the corresponding image frames are automatically marked as anomalous frames, and their negative impact is eliminated during the repositioning projection detection and quality inspection process. This effectively eliminates pose changes caused by non-positioning factors such as violent vehicle movement or momentary sensor loss, resolving the unreliability of positioning results in long-tail scenarios such as parking, reversing, and U-turns, and ensuring the continuity and authenticity of the quality inspection data sequence.

[0075] In one implementation, a dynamic Region of Interest (ROI) bounding box is constructed within the corresponding image frame based on the camera's extrinsic position and the vehicle's historical trajectory elevation information. This dynamic ROI bounding box can adaptively adjust its coverage area in the image according to the vehicle's real-time position and attitude, prioritizing the retention of high-quality, high-reference pixels in the near end while shielding low-quality, low-reference pixels in the far end (such as distant sky or blurred horizon interference). Through dynamic ROI bounding box constraints, the system can focus on the effective perception area that contributes most to localization, reducing computational redundancy of invalid data and noise interference.

[0076] Those skilled in the art will understand that, in the above-described method of the specific implementation, the order in which each step is written does not imply a strict execution order, nor does it impose any limitation on the implementation process. The specific execution order of each step should be determined by its function and possible internal logic.

[0077] Please see Figure 3 , Figure 3 This is a schematic diagram of the structure of an electronic device according to an embodiment of this application. The electronic device 30 includes a memory 31 and a processor 32 coupled to each other. The processor 32 is used to execute program instructions stored in the memory 31 to implement the steps in any of the above-described data evaluation method embodiments. In a specific implementation scenario, the electronic device 30 may include, but is not limited to, a microcomputer or a server. Furthermore, the electronic device 30 may also include mobile devices such as laptops and tablets, without limitation.

[0078] Specifically, processor 32 controls itself and memory 31 to implement the steps in any of the above-described data evaluation method embodiments. Processor 32 can also be referred to as a CPU (Central Processing Unit). Processor 32 may be an integrated circuit chip with signal processing capabilities. Processor 32 can also be a general-purpose processor, digital signal processor (DSP), application-specific integrated circuit (ASIC), field-programmable gate array (FPGA), or other programmable logic devices, discrete gate or transistor logic devices, or discrete hardware components. A general-purpose processor can be a microprocessor or any conventional processor. Furthermore, processor 32 can be implemented using integrated circuit chips.

[0079] Please see Figure 4 , Figure 4This is a schematic diagram of the structure of a non-volatile computer-readable storage medium according to an embodiment of this application. The non-volatile computer-readable storage medium 40 stores program instructions 401 that can be executed by a processor. The program instructions 401 are used to implement the steps in the above-described data evaluation method embodiment.

[0080] In some embodiments, the functions or modules of the apparatus provided in this disclosure can be used to perform the methods described in the above method embodiments. The specific implementation can be referred to the description of the above method embodiments, and for the sake of brevity, it will not be repeated here.

[0081] The description of the various embodiments above tends to emphasize the differences between the various embodiments. The similarities or similarities between them can be referred to, and for the sake of brevity, they will not be repeated here.

[0082] In the several embodiments provided in this application, it should be understood that the disclosed methods and apparatus can be implemented in other ways. For example, the apparatus implementations described above are merely illustrative. For instance, the division of modules or units is only a logical functional division, and in actual implementation, there may be other division methods. For example, units or components may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the mutual coupling or direct coupling or communication connection shown or discussed may be through some interfaces; the indirect coupling or communication connection of devices or units may be electrical, mechanical, or other forms.

[0083] Furthermore, the functional units in the various embodiments of this application can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The integrated unit can be implemented in hardware or as a software functional unit.

[0084] If the integrated unit is implemented as a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, or all or part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) or processor to execute all or part of the steps of the methods of various embodiments of this application. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0085] Those skilled in the art will readily recognize that numerous modifications and variations can be made to the apparatus and method while maintaining the teachings of this application. Therefore, the above disclosure should be considered limited only by the scope of the appended claims.

Claims

1. A data evaluation method, characterized in that, include: Obtain the target image data and target mask data corresponding to the relocation trajectory; Projection mapping is performed using the target image data to obtain projected image data; The target mask data is used to perform calculations to construct a weight constraint function; Based on the projected image data, the target mask data, and the weight constraint function, the feature error value between the projected image data and the target mask data is calculated, and then it is determined whether the repositioning data corresponding to the repositioning trajectory is qualified.

2. The method according to claim 1, characterized in that, The step of using the target mask data to calculate and construct a weight constraint function includes: Based on the camera parameters and the target mask data, calculate the semantic confidence of the target mask data; The weight constraint function for the target mask data is constructed based on the semantic confidence.

3. The method according to claim 2, characterized in that, The target mask data includes semantic geometry; The step of calculating the semantic confidence of the target mask data based on camera parameters and the target mask data includes: Calculate the distance from the semantic geometry surface to the camera; The semantic confidence is determined based on the distance and the semantic area of ​​the semantic geometry. Wherein, when the semantic area is less than a preset area, the semantic confidence is 0; When the semantic area is greater than the preset area, the semantic confidence is the ratio of the semantic area to the preset value, where the preset value is the sum of the distance and 1.

4. The method according to claim 3, characterized in that, Further includes: In response to continuous features in the semantic geometry surface representation map, the average confidence of all pixels of the continuous features within the target mask data is calculated as the semantic confidence.

5. The method according to claim 2, characterized in that, The weight constraint function for constructing the target mask data based on the semantic confidence includes: In response to the semantic confidence being less than or equal to a first preset confidence, the weight constraint function is 0; In response to the semantic confidence being less than a second preset confidence and greater than a first preset confidence, the weight constraint function is the difference between the semantic confidence and the first preset confidence; In response to the semantic confidence being greater than or equal to the second preset confidence, the weight constraint function is 1.

6. The method according to claim 1, characterized in that, The step of calculating the feature error value between the projected image data and the target mask data based on the projected image data, the target mask data, and the weight constraint function includes: The projected image data is sampled to obtain projected feature points; In the target mask data, determine the semantic feature points adjacent to the projected feature points; Calculate the geometric distance between the projected feature points and the semantic feature points; The product of the weight constraint function and the geometric distance is used as the feature error value.

7. The method according to claim 1, characterized in that, Obtaining the target image data corresponding to the repositioning trajectory includes: Obtain the vehicle speed and wheel speed corresponding to the trajectory points in the repositioning trajectory; In response to the vehicle speed being greater than a first threshold and the wheel speed being less than a second threshold within a preset time period, the initial image data of the trajectory point at the current moment is determined to be abnormal data, wherein the current moment is located in the middle of the preset time period; The initial image data of the trajectory points at the current moment is removed to obtain the target image data.

8. The method according to claim 7, characterized in that, Obtaining the target mask data corresponding to the relocation trajectory includes: A region of interest bounding box is constructed with the trajectory point as the center, wherein the length direction of the region of interest bounding box is determined based on the trajectory forward direction; The target image data is obtained by cropping the mask data corresponding to the target image data based on the region of interest bounding box.

9. The method according to claim 1, characterized in that, The step of using the target image data to perform projection mapping to obtain projected image data includes: Based on camera parameters, vector elements in the high-precision map are projected and rendered onto the target image data to obtain the projected image data.

10. The method according to claim 9, characterized in that, Also includes: In response to the vector feature being a lane line, the weighted average error between the projected image data and the corresponding target mask data is obtained as the feature error value; In response to the vector element being an arrow or a stop line, the intersection-union ratio between the projected image data and the corresponding target mask data is obtained as the feature error value.

11. The method according to claim 1, characterized in that, The relocation data includes image-level data, frame-level data, and packet-level data, and the method further includes: Based on the pass rate of semantic objects in the image-level data, determine whether the image-level data is qualified; Based on the pass rate of multiple viewpoint images in the frame-level data, determine whether the frame-level data is qualified; Based on the pass rate of multiple frames in the packet-level data, it is determined whether the packet-level data is qualified.

12. An electronic device, characterized in that, It includes a memory and a processor coupled to each other, the processor being used to execute program instructions stored in the memory to implement the data evaluation method according to any one of claims 1 to 11.

13. A computer-readable storage medium having program instructions stored thereon, characterized in that, When the program instructions are executed by the processor, they implement the data evaluation method according to any one of claims 1 to 11.

Images