Quality checking and correction method and apparatus for static data of autonomous driving

Abstract

Provided in the present application are a quality checking and correction method and apparatus for static data of autonomous driving. The method comprises: acquiring pre-annotated data and data collected by a sensor; and on the basis of the data collected by the sensor, performing quality checking and correction on N quality checking dimensions of the pre-annotated data, wherein the N quality checking dimensions comprise the n-th quality checking dimension and the (n+1)-th quality checking dimension, a quality checking and correction result of the n-th quality checking dimension is an input of the (n+1)-th quality checking dimension, N>1 and N is an integer, and 1≤n<N and n is an integer. The present application helps to reduce the false alarm rate of errors when performing quality checking on annotated data, and also helps to improve the effectiveness of correcting pre-annotated data.

Description

Methods and devices for quality inspection and repair of static data in autonomous driving

[0001] This application claims priority to Chinese Patent Application No. 202411428743.9, filed on October 12, 2024, entitled "Method and Apparatus for Quality Inspection and Repair of Static Data of Autonomous Driving", the entire contents of which are incorporated herein by reference. Technical Field

[0002] This application relates to the field of intelligent vehicles, and more specifically, to a method and apparatus for quality inspection and repair of static data for autonomous driving. Background Technology

[0003] With the continuous development of autonomous driving technology, static feature annotation from a bird's-eye view (BEV) perspective has gradually become the mainstream annotation scheme for autonomous driving perception. Regardless of whether it's multi-trip reverse projection for single-trip pre-annotation, high-precision map pre-annotation, or purely manual annotation, static feature annotation data inevitably suffers from various quality issues. How to reduce the false alarm rate during quality inspection of static feature annotation data has become an urgent problem to be solved. Summary of the Invention

[0004] This application provides a method and apparatus for quality inspection and repair of static data for autonomous driving, which helps to reduce the false alarm rate when inspecting labeled data and also helps to improve the effectiveness of repairing pre-labeled data.

[0005] In a first aspect, this application provides a method for quality inspection and repair of static data for autonomous driving. The method includes: acquiring pre-labeled data and data collected by sensors; and performing quality inspection and repair on N quality inspection dimensions of the pre-labeled data based on the data collected by the sensors, wherein the N quality inspection dimensions include an nth quality inspection dimension and an (n+1)th quality inspection dimension, and the quality inspection and repair result of the nth quality inspection dimension is the input of the (n+1)th quality inspection dimension; wherein N > 1 and N is an integer, 1 ≤ n < N and n is an integer.

[0006] Because quality inspection errors across different dimensions can influence each other—for example, topology errors can affect geometric offset errors, which in turn can affect element attribute errors—non-layered quality inspection across N dimensions leads to a high false positive rate and low effectiveness in correcting these errors. Based on the aforementioned technical solution, layered quality inspection and correction (or sequential quality inspection and correction) of the nth and (n+1th)th quality inspection dimensions helps reduce the false positive rate when inspecting labeled data and also improves the effectiveness of correcting pre-labeled data.

[0007] In some possible implementations, based on the data collected by the sensor, N quality control dimensions of the pre-labeled data are inspected and repaired, including: based on the data collected by the sensor, the nth quality control dimension of the (n-1)th labeling result is inspected and repaired to obtain the nth labeling result; based on the data collected by the sensor, the (n+1)th quality control dimension of the nth labeling result is inspected and repaired to obtain the (n+1)th labeling result; wherein, when n is 1, the (n-1)th labeling result includes multiple elements in the pre-labeled data; or, when 1 < n < N and n is an integer, the (n-1)th labeling result includes the labeling results obtained after inspecting and repairing the 1st to (n-1)th quality control dimensions.

[0008] In conjunction with the first aspect, in some implementations of the first aspect, based on the data collected by the sensor, quality inspection and repair are performed on N quality inspection dimensions of the pre-labeled data, including: based on the data collected by the sensor, quality inspection and repair are performed on the nth quality inspection dimension of the (n-1)th labeling result to obtain the nth labeling result and the nth set of erroneous elements, the nth set of erroneous elements including elements that failed quality inspection for the nth quality inspection dimension and were not successfully modified; based on the data collected by the sensor, the (n+1)th quality inspection dimension of the elements in the nth labeling result other than the nth set of erroneous elements is repaired. The quality inspection and repair of the inspection dimension are performed to obtain the (n+1)th annotation result and the (n+1)th set of error elements. The (n+1)th set of error elements includes the elements in the nth annotation result other than the nth set of error elements, which are elements that failed to be repaired for the (n+1)th quality inspection dimension. When n is 1, the (n-1)th annotation result includes multiple elements in the pre-annotated data; or, when 1 < n < N and n is an integer, the (n-1)th annotation result includes the annotation results obtained after quality inspection and repair of the 1st to (n-1)th quality inspection dimensions.

[0009] The process of performing quality inspection and repair on the nth and (n+1)th quality inspection dimensions can be called performing layered quality inspection and repair on the nth and (n+1)th quality inspection dimensions, or performing serial quality inspection and repair on the nth and (n+1)th quality inspection dimensions.

[0010] Based on the above technical solution, the (n+1)th quality inspection dimension of elements other than the nth erroneous element set in the nth annotation result can be inspected and repaired. This can exclude elements that have quality inspection errors in the previous level and have not been successfully repaired, which helps to further reduce the false alarm rate of quality inspection errors when inspecting the labeled data, and also helps to improve the effectiveness of repairing pre-labeled data.

[0011] In some possible implementations, based on the data collected by the sensor, the N quality inspection dimensions of the pre-labeled data are inspected and repaired to obtain the labeling results. This includes: based on the data collected by the sensor, the N quality inspection dimensions of the pre-labeled data are inspected and repaired hierarchically to obtain the labeling results.

[0012] In conjunction with the first aspect, in some implementations of the first aspect, after quality inspection and repair of the Nth quality inspection dimension, the annotation result is obtained. The N quality inspection dimensions include the mth quality inspection dimension, 1≤m≤N and m is an integer. The method further includes: based on the data collected by the sensor, quality inspection and repair of the mth quality inspection dimension of the annotation result is performed to obtain the m'th annotation result and the m'th set of error elements. The m'th set of error elements includes the elements in the multiple elements of the annotation result that have been found to have quality inspection errors for the mth quality inspection dimension and have not been successfully repaired.

[0013] Based on the above technical solution, by performing multiple quality inspections and repair iterations on multiple quality inspection dimensions, it is possible to ensure that quality inspection errors are repaired as much as possible, so that the number of unrepairable quality inspection errors in the final repaired annotation results is minimized, which helps to reduce the cost and time of subsequent manual repair.

[0014] In some possible implementations, m is 1.

[0015] In some possible implementations, the method also includes: performing hierarchical quality inspection and repair on the N quality inspection dimensions of the annotation result based on the data collected by the sensor, to obtain another annotation result.

[0016] In conjunction with the first aspect, in some implementations of the first aspect, based on the data collected by the sensor, the nth quality inspection dimension of the (n-1)th annotation result is inspected and repaired to obtain the nth annotation result and the nth set of error elements. This includes: based on the data collected by the sensor, the nth quality inspection dimension of the (n-1)th annotation result is inspected and repaired to obtain the n”th annotation result and the n”th set of error elements, wherein the n”th set of error elements includes elements in the (n-1)th annotation result that have been inspected and repaired for the nth quality inspection dimension and have not been successfully modified; based on the data collected by the sensor, the nth quality inspection dimension of the n”th annotation result is inspected and repaired to obtain the nth annotation result and the nth set of error elements.

[0017] Based on the above technical solution, by performing multiple quality inspections and repair iterations on the same quality inspection dimension, it is possible to ensure that quality inspection errors generated within that dimension are repaired as much as possible. This minimizes the number of unrepairable quality inspection errors in the final repaired annotation results, thereby reducing the cost and time of subsequent manual repair.

[0018] In conjunction with the first aspect, in some implementations of the first aspect, the method further includes: obtaining the quality inspection order of the N quality inspection dimensions; wherein, performing quality inspection and repair on the N quality inspection dimensions of the pre-labeled data according to the data collected by the sensor includes: performing quality inspection and repair on the N quality inspection dimensions of the pre-labeled data according to the quality inspection order based on the data collected by the sensor.

[0019] Based on the above technical solution, by obtaining the quality inspection order of N quality inspection dimensions and performing quality inspection according to the quality inspection order, it is helpful to reduce the false alarm rate of quality inspection errors when inspecting labeled data, and also to improve the effectiveness of repairing pre-labeled data.

[0020] In conjunction with the first aspect, in some implementations of the first aspect, the quality inspection order is determined by the drawing order of the same element or the same layer in the pre-annotated data.

[0021] In conjunction with the first aspect, in some implementations of the first aspect, the order of the N quality inspection dimensions includes at least two of the following: topology, geometric offset, association, and element attributes.

[0022] The inventors of this application have discovered that the annotation quality of static features exhibits a significant hierarchical structure. Quality inspection errors arising from different quality inspection dimensions can influence each other; for example, topological structure errors can affect geometric offset errors, which in turn can affect element attribute errors. Based on the aforementioned technical solution, by performing quality inspection and repair according to at least two of the following: topological structure, geometric offset, association relationships, and element attributes, it is helpful to reduce the false alarm rate of quality inspection errors when inspecting labeled data, and also to improve the effectiveness of repairing pre-labeled data.

[0023] In conjunction with the first aspect, in some implementations of the first aspect, the quality inspection order is determined by the drawing order among multiple layers in the pre-annotated data.

[0024] In conjunction with the first aspect, in some implementations of the first aspect, the order of the N quality inspection dimensions includes at least two of the lane line layer, lane layer, and road layer.

[0025] The inventors of this application have discovered that static feature annotations typically have clearly defined drawing orders for different layers. Therefore, layered quality inspection and repair can be performed based on the annotation layers as a layering principle. Based on the above technical solution, by performing quality inspection and repair according to at least two of the lane line layer, lane layer, and road layer, it helps reduce the false alarm rate of quality inspection errors when inspecting annotation data, and also helps improve the effectiveness of repairing pre-annotated data.

[0026] In conjunction with the first aspect, in some implementations of the first aspect, the pre-labeled data is labeled data for static features.

[0027] Secondly, an apparatus for quality inspection and repair of static data for autonomous driving is provided. The apparatus includes: an acquisition unit for acquiring pre-labeled data and data collected by sensors; and a data quality inspection and repair unit for performing quality inspection and repair on N quality inspection dimensions of the pre-labeled data based on the data collected by the sensors. The N quality inspection dimensions include an nth quality inspection dimension and an (n+1)th quality inspection dimension, and the quality inspection and repair result of the nth quality inspection dimension is the input of the (n+1)th quality inspection dimension. N > 1 and N is an integer, 1 ≤ n < N and n is an integer.

[0028] In conjunction with the second aspect, in some implementations of the second aspect, the data quality inspection and repair unit is specifically used for: performing quality inspection and repair on the nth quality inspection dimension of the (n-1)th annotation result based on the data collected by the sensor, to obtain the nth annotation result and the nth set of erroneous elements, wherein the nth set of erroneous elements includes elements that have failed quality inspection for the nth quality inspection dimension and have not been successfully modified; and performing quality inspection and repair on the (n+1)th quality inspection dimension of the elements in the nth annotation result other than the nth set of erroneous elements based on the data collected by the sensor. This yields the (n+1)th annotation result and the (n+1)th set of error elements. The (n+1)th set of error elements includes elements from the nth annotation result that were not successfully corrected for the (n+1)th quality inspection dimension, excluding the elements in the nth annotation result. When n is 1, the (n-1)th annotation result includes multiple elements from the pre-annotated data. Alternatively, when 1 < n < N and n is an integer, the (n-1)th annotation result includes the annotation results obtained after quality inspection and correction for the 1st to (n-1)th quality inspection dimensions.

[0029] In conjunction with the second aspect, in some implementations of the second aspect, after quality inspection and repair of the Nth quality inspection dimension, a labeling result is obtained. The N quality inspection dimensions include the mth quality inspection dimension, 1≤m≤N and m is an integer. The data quality inspection and repair unit is further configured to: perform quality inspection and repair on the mth quality inspection dimension of the labeling result based on the data collected by the sensor, to obtain the m'th labeling result and the m'th set of error elements. The m'th set of error elements includes elements that have failed to be repaired due to quality inspection errors in the mth quality inspection dimension.

[0030] In conjunction with the second aspect, in some implementations of the second aspect, the data quality inspection and repair unit is specifically used for: performing quality inspection and repair on the nth quality inspection dimension of the (n-1)th annotation result based on the data collected by the sensor, to obtain the n”th annotation result and the n”th error element set, wherein the n”th error element set includes elements in the (n-1)th annotation result that have been found to have quality inspection errors for the nth quality inspection dimension and have not been successfully modified; and performing quality inspection and repair on the n”th quality inspection dimension of the n”th annotation result based on the data collected by the sensor, to obtain the nth annotation result and the nth error element set.

[0031] In conjunction with the second aspect, in some implementations of the second aspect, the acquisition unit is further used to acquire the quality inspection order of the N quality inspection dimensions; wherein, the data quality inspection and repair unit is specifically used to: perform quality inspection and repair on the N quality inspection dimensions of the pre-labeled data according to the quality inspection order based on the data collected by the sensor.

[0032] In conjunction with the second aspect, in some implementations of the second aspect, the quality inspection order is determined by the drawing order of the same element or the same layer in the pre-annotated data.

[0033] In conjunction with the second aspect, in some implementations of the second aspect, the order of the N quality inspection dimensions includes at least two of the following: topology, geometric offset, association, and element attributes.

[0034] In conjunction with the second aspect, in some implementations of the second aspect, the quality inspection order is determined by the drawing order among multiple layers in the pre-annotated data.

[0035] In conjunction with the second aspect, in some implementations of the second aspect, the order of the N quality inspection dimensions includes at least two of the lane line layer, lane layer, and road layer.

[0036] In conjunction with the second aspect, in some implementations of the second aspect, the pre-labeled data is labeled data for static features.

[0037] Thirdly, this application provides a quality inspection and repair device for static data of autonomous driving. The device includes a memory and a processor. The memory is used to store a computer program, and the processor is used to execute the computer program in the memory, so that the intelligent driving device can implement the methods in the first aspect and any possible implementation thereof.

[0038] Fourthly, this application provides a quality inspection and repair device for static data of autonomous driving, which includes the apparatus described in the second or third aspect above.

[0039] For example, the quality inspection and repair equipment for the static data of autonomous driving can be a server.

[0040] Fifthly, this application provides a computer program product comprising: computer program code, which, when executed on a computer, causes the computer to perform the method in any possible implementation of the first aspect.

[0041] In a sixth aspect, this application provides a computer-readable storage medium storing a computer program that, when run on a computer, causes the computer to perform the method in any possible implementation of the first aspect.

[0042] In a seventh aspect, this application provides a chip including circuitry for performing the method in any possible implementation of the first aspect described above.

[0043] Eighthly, this application provides a method for quality inspection and repair of static data for autonomous driving. The method includes: acquiring pre-labeled data and data collected by sensors; performing quality inspection on N quality inspection dimensions of the pre-labeled data based on the data collected by the sensors, wherein the N quality inspection dimensions include an nth quality inspection dimension and an (n+1)th quality inspection dimension, and the quality inspection result of the nth quality inspection dimension is the input of the (n+1)th quality inspection dimension; wherein N > 1 and N is an integer, 1 ≤ n < N and n is an integer.

[0044] In conjunction with the eighth aspect, in some implementations of the eighth aspect, quality checks are performed on the N quality inspection dimensions of the pre-labeled data based on the data collected by the sensor, including: performing quality checks on the nth quality inspection dimension of the (n-1)th labeling result based on the data collected by the sensor to obtain the nth error element set, which includes elements that have errors in the quality inspection for the nth quality inspection dimension; and performing quality checks on the (n+1)th quality inspection dimension of the elements other than the nth error element set in the (n-1)th labeling result based on the data collected by the sensor to obtain the (n+1)th error element set, which includes elements other than the nth error element set in the (n-1)th labeling result that have errors in the quality inspection for the (n+1)th quality inspection dimension; wherein, the (n-1)th labeling result is the pre-labeled data; or, the (n-1)th labeling result includes the quality inspection results obtained after performing quality checks on the 1st to (n-1)th quality inspection dimensions.

[0045] In conjunction with the eighth aspect, in some implementations of the eighth aspect, the method further includes: performing data repair based on the nth set of erroneous elements and the (n+1)th set of erroneous elements. Attached Figure Description

[0046] Figure 1 is a functional block diagram of the vehicle provided in an embodiment of this application.

[0047] Figure 2 is a schematic diagram of non-layered quality inspection and repair.

[0048] Figure 3 is a schematic diagram of the architecture of the quality inspection and repair system provided in the embodiments of this application.

[0049] Figure 4 is a schematic flowchart of the quality inspection and repair method for static data of autonomous driving provided in an embodiment of this application.

[0050] Figure 5 is a schematic diagram of multi-level quality inspection and repair provided in the embodiments of this application.

[0051] Figure 6 is a schematic diagram of the four-layer quality inspection and repair provided in the embodiment of this application.

[0052] Figure 7 is a schematic diagram of multiple quality inspections and repair iterations within the same level provided in an embodiment of this application.

[0053] Figure 8 is a schematic diagram of multiple quality inspections and repair iterations at multiple levels provided in the embodiments of this application.

[0054] Figure 9 is a schematic diagram of the quality inspection and repair architecture provided in an embodiment of this application.

[0055] Figure 10 is a schematic diagram of the three-layer quality inspection and repair provided in the embodiment of this application.

[0056] Figure 11 is a schematic block diagram of the quality inspection and repair device for static data of autonomous driving provided in an embodiment of this application. Detailed Implementation

[0057] The technical solutions of the embodiments of this application will be described below with reference to the accompanying drawings. In the description of the embodiments of this application, unless otherwise stated, " / " means "or," for example, A / B can mean A or B; "and / or" in this document is merely a description of the association 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, and B existing alone. "At least one" refers to one or more. For example, "at least one of A and B," similar to "A and / or B," describes the association relationship between related objects, indicating that three relationships can exist. For example, at least one of A and B can represent: A existing alone, A and B existing simultaneously, and B existing alone.

[0058] The prefixes such as "first" and "second" used in this application embodiment are merely for distinguishing different descriptive objects and do not limit the position, order, priority, quantity, or content of the described objects. The use of ordinal numbers and other prefixes used to distinguish descriptive objects in this application embodiment does not constitute a limitation on the described objects. The description of the described objects is given in the claims or the context of the embodiments, and should not constitute unnecessary restrictions due to the use of such prefixes. Furthermore, in the description of this embodiment, unless otherwise stated, "multiple" means two or more.

[0059] Figure 1 is a functional block diagram of a vehicle 100 provided in an embodiment of this application. The vehicle 100 may include a perception system 110 and a computing platform 120. The perception system 110 may include one or more sensors for sensing information about the environment surrounding the vehicle 100. For example, the perception system 110 may include a positioning system, which may be a Global Positioning System (GPS), a BeiDou Navigation Satellite System, or another positioning system. As another example, the perception system 110 may include one or more of the following: an inertial measurement unit (IMU), an accelerometer, a lidar, a millimeter-wave radar, an ultrasonic radar (sonar), and a camera device.

[0060] Some or all of the functions of vehicle 100 can be controlled by computing platform 120. Computing platform 120 may include one or more processors, such as processors 121 to 12n (n being a positive integer). A processor is a circuit with signal processing capabilities. In one implementation, the processor can be a circuit with instruction read and execute capabilities, such as a central processing unit (CPU), microprocessor, graphics processing unit (GPU) (which can be understood as a type of microprocessor), or digital signal processor (DSP). In another implementation, the processor can implement certain functions through the logical relationships of hardware circuits. These logical relationships are fixed or reconfigurable. For example, the processor may be a hardware circuit implemented using an application-specific integrated circuit (ASIC) or a programmable logic device (PLD), such as a field-programmable gate array (FPGA). In reconfigurable hardware circuits, the process of the processor loading a configuration document and configuring the hardware circuit can be understood as the process of the processor loading instructions to implement some or all of the functions of the aforementioned units. Furthermore, the processor can also be a hardware circuit designed for artificial intelligence, which can be understood as an ASIC, such as a neural network processing unit (NPU), tensor processing unit (TPU), deep learning processing unit (DPU), etc. In addition, the computing platform 120 may also include a memory for storing instructions. Some or all of the processors 121 to 12n can call the instructions in the memory to implement the corresponding functions.

[0061] Optionally, the structure of the vehicle 100 described above is merely illustrative. In actual applications, various components of the vehicle 100 may be added or removed as needed.

[0062] With the continuous development of autonomous driving technology, static feature annotation in BEVs (Battery Electric Vehicles) has gradually become the mainstream annotation scheme for autonomous driving perception. Regardless of whether it's multi-trip reverse projection for single-trip pre-annotation, high-precision map pre-annotation, or purely manual annotation, static feature annotation data inevitably suffers from various quality issues:

[0063] (1) Pre-marked changes in reality: Due to road construction and other reasons, some data have changed in reality, resulting in differences in topology and other aspects between the latest single-trip data and the original multi-trip data.

[0064] For example, the latest single-trip data indicates that the vehicle is currently on a two-lane road, while the original multi-trip data indicates that the road is a three-lane road.

[0065] (2) Pre-marked geometric offset: The repositioning error of multiple and single-trip quality inspections results in the geometric position offset of the lane lines in the back projection of multiple trips and single trips.

[0066] For example, for the same lane line, the latest single-pass data indicates the position of the lane line as position 1, while the original multi-pass data indicates the position of the lane line as position 2, with a certain positional offset between position 1 and position 2.

[0067] (3) Mislabeling of element attributes: Due to defects in the pre-annotation algorithm or errors in manual annotation, the attributes of elements are mislabeled.

[0068] For example, taking a lane line as an example, due to defects in the pre-annotation algorithm or errors in manual annotation, the lane line is labeled as a solid line, when in fact it is a dashed line.

[0069] For example, taking a lane as an example, due to defects in the pre-labeling algorithm or errors in manual labeling, the lane is labeled as a regular lane, while in reality it is a bus lane.

[0070] (4) Error in association: Due to geometric errors or annotation errors, the binding relationship between elements is incorrect.

[0071] For example, taking the lane centerline as an example, due to geometric errors or labeling errors, the static feature labeling data may show the lane centerline as associated with left lane line 1 and right lane line 2, while in reality the lane centerline is associated with left lane line 3 and right lane line 4.

[0072] Therefore, in order to obtain high-quality static feature annotation data and improve the efficiency of repairing data with quality inspection errors, automated quality inspection and automatic error repair become particularly important.

[0073] The inventors of this application discovered that static element annotation usually has a clear set of annotation rules. Based on this characteristic, all quality inspection errors of static elements are re-sorted and summarized, which can be classified into four categories: (1) Topology: geometric topology errors within or between elements, such as real-world changes, topology errors between elements, missing element labels, omissions in element labels, etc.; (2) Geometric offset: horizontal or vertical geometric offset of lines and surfaces relative to the BEV annotation base map; (3) Association: various association layer errors, that is, the binding relationship between elements is wrong; (4) Element attributes: various attribute errors of elements, such as lane attribute errors, lane line color attribute errors, etc.

[0074] Through the analysis and review of quality inspection errors, the inventors of this application discovered that the annotation quality of static elements exhibits a significant hierarchical structure. Errors arising from different quality inspection dimensions can influence each other; for example, topological errors can affect geometric offset errors, which in turn can affect element attribute errors. For instance, if a full, non-hierarchical quality inspection is performed—meaning all inspection items are run independently—many invalid and correlated errors can occur: an element might detect three errors (A, B, and C), the root cause of which is error A, while errors B and C are correlated and invalid. After fixing A, errors B and C naturally disappear. For example, error A might be a change from three lanes to two lanes (due to construction), error B might be a geometric offset caused by the construction lane, and error C might be an attribute error of the construction lane. Both the geometric offset and attribute error of the construction lane are caused by topological issues. If topological structures, geometric offsets, and element attributes are not inspected and corrected hierarchically, errors A, B, and C might all appear simultaneously in the quality inspection errors. Among them, repairing the geometric offset of the construction lane and repairing the attributes of the construction lane (such as repairing the construction lane into a regular lane or bus lane) are both considered invalid repairs.

[0075] For example, automatically repairing all quality inspection errors without grading can also lead to ineffective repairs: repairing error A might affect the repair of error B or introduce additional error C. For instance, error A might be a lane attribute error, such as changing a lane from a regular lane to a reversible lane by repairing error A. This would then require additional repairs to lane line attributes (e.g., reversible lanes use double yellow dashed lines; repairing lane line attributes would change the lane lines from white lines to double yellow dashed lines). In other words, while parallel repair may fix error A, it can affect the repair of error B or introduce additional error C, resulting in a relatively high number of errors in the final modified data, or lower repair efficiency.

[0076] For example, Figure 2 illustrates a schematic diagram of non-layered quality inspection and repair. After performing a full non-layered quality inspection and repair process on the pre-labeled data based on the data collected by the sensors, some quality inspection errors still remain unresolved. Some of these problems are caused by false alarms from the quality inspection algorithm, while others are genuine errors that have not been effectively repaired.

[0077] For example, Figure 3 shows a schematic diagram of the architecture of the quality inspection and repair system 300 provided in an embodiment of this application.

[0078] Multi-source input alignment module 310: In this embodiment, multi-source results are used as input, including static feature vector elements in pre-annotated data or manually annotated data, perspective view (PV) images, BEV semantic base maps, etc. The multi-source input alignment module 310 performs temporal and spatial alignment on the above-mentioned multi-source inputs to ensure that the multi-source inputs are self-consistent in time and space.

[0079] The multi-level iterative quality inspection and repair module 320 is used to perform hierarchical quality inspection and repair on the aligned multi-source inputs.

[0080] Optionally, the multi-level iterative quality inspection and repair module 320 is also used to perform iterative quality inspection and repair within each level of quality inspection dimension and between multiple levels of quality inspection dimensions, and finally output quality inspection error information and the annotation results after repair.

[0081] The quality inspection error information display module 330 is used to display quality inspection errors on the front-end interface or in the form of a formatted document. It will display the quality inspection errors at each level and a list of the superior quality inspection error elements (which cannot be repaired) filtered out at that level.

[0082] The repair result update module 340 is used to synchronously update the annotation results obtained by the multi-level iterative quality inspection and repair module 320 to the latest annotation results in both the front end and the back end.

[0083] The above-mentioned error information display module 330 and repair result update module 340 are optional modules in the quality inspection and repair system 300.

[0084] The front end can refer to the annotation interface of the quality inspection and repair system 300; the back end can refer to the annotation database, which can be used to store annotation data.

[0085] Based on the above analysis, the inventors of this application have found that layered quality inspection and repair are particularly important. Each level of quality inspection focuses on detecting similar problems and automatically repairs them. The output of each level can include two parts: (1) elements that are faulty and cannot be repaired; (2) the annotation results after repair. The two outputs of the current level can be used as inputs for the next level. For example, when each level performs quality inspection, it can automatically filter out elements that are faulty in the previous level (i.e., the level will no longer perform quality inspection on these elements) based on the list of faulty elements in the previous level (including elements that are faulty and cannot be repaired in the previous level) to avoid other invalid association errors caused by one error; at the same time, when each level performs repair, it can repair based on the latest annotation results after automatic repair in the previous level to improve the effectiveness of quality inspection and repair.

[0086] Figure 4 shows a schematic flowchart of a quality inspection and repair method 400 for static data of autonomous driving provided in an embodiment of this application. This method 400 can be executed by the aforementioned quality inspection and repair system 300; alternatively, it can be executed by a data quality inspection and repair device (e.g., a server). The method 400 includes:

[0087] S410 acquires pre-labeled data and data collected by sensors.

[0088] For example, the pre-labeled data can be the original multi-trip data or manually labeled data mentioned above, and the data collected by the sensor can be the latest single-trip data mentioned above.

[0089] S420, Based on the data collected by the sensor, perform quality inspection and repair on N quality inspection dimensions of the pre-labeled data. The N quality inspection dimensions include the nth quality inspection dimension and the (n+1)th quality inspection dimension. The quality inspection and repair result of the nth quality inspection dimension is the input of the (n+1)th quality inspection dimension. Where N > 1 and N is an integer, 1 ≤ n < N and n is an integer.

[0090] The quality inspection and repair results of the nth quality inspection dimension are the input of the (n+1)th quality inspection dimension. This can also be understood as the quality inspection of the nth and (n+1)th quality inspection dimensions being serial quality inspection and repair, or it can be understood as the quality inspection of the nth and (n+1)th quality inspection dimensions being hierarchical quality inspection and repair.

[0091] Optionally, based on the data collected by the sensor, quality inspection and repair are performed on N quality inspection dimensions of the pre-labeled data, including: based on the data collected by the sensor, quality inspection and repair are performed on the nth quality inspection dimension of the (n-1)th labeling result to obtain the nth labeling result; based on the data collected by the sensor, quality inspection and repair are performed on the (n+1)th quality inspection dimension of the nth labeling result to obtain the (n+1)th labeling result; wherein, when n is 1, the (n-1)th labeling result includes multiple elements in the pre-labeled data; or, when 1 < n < N, the (n-1)th labeling result includes the labeling results obtained after quality inspection and repair of the 1st to (n-1)th quality inspection dimensions.

[0092] The nth annotation result above can be the quality inspection and repair result of the nth quality inspection dimension in S420 above.

[0093] For example, when n is 1, the first quality inspection dimension of the pre-labeled data is inspected and repaired based on the data collected by the sensor to obtain the first labeling result; the second quality inspection dimension of the first labeling result is inspected and repaired based on the data collected by the sensor to obtain the second labeling result.

[0094] For example, the pre-labeled data includes 10 elements. First, quality control can be performed on the first quality control dimension of these 10 elements, resulting in elements 1-5 with quality control errors. Repairing the first quality control dimension of elements 1-5 yields the first labeling result (e.g., elements 1-3 were not successfully repaired in the first labeling result, while elements 4-5 were successfully repaired). Next, quality control can be performed on the second quality control dimension of the first labeling result, resulting in elements 1-3 with quality control errors. Repairing the second quality control dimension of elements 1-3 yields the second labeling result (e.g., elements 1-2 were not successfully repaired in the second labeling result, while element 3 was successfully repaired).

[0095] Optionally, based on the data collected by the sensor, quality inspection and repair are performed on N quality inspection dimensions of the pre-labeled data, including: based on the data collected by the sensor, quality inspection and repair are performed on the nth quality inspection dimension of the (n-1)th labeling result to obtain the nth labeling result and the nth set of erroneous elements, which includes elements that failed quality inspection for the nth quality inspection dimension and were not successfully modified; based on the data collected by the sensor, quality inspection and repair are performed on the (n+1)th quality inspection dimension of the elements in the nth labeling result other than the nth set of erroneous elements. The process involves inspection and repair, resulting in the (n+1)th annotation result and the (n+1)th set of erroneous elements. The (n+1)th set of erroneous elements includes elements from the nth annotation result that were erroneous in the quality inspection for the (n+1)th quality inspection dimension and were not successfully repaired. When n is 1, the (n-1)th annotation result includes multiple elements from the pre-annotated data. Alternatively, when 1 < n < N, the (n-1)th annotation result includes the annotation results obtained after quality inspection and repair for the 1st to (n-1)th quality inspection dimensions.

[0096] The nth annotation result and the nth set of error elements above can be the quality inspection and repair results of the nth quality inspection dimension in S420 above.

[0097] For example, taking N as 4, n can be 1. A hierarchical quality inspection and remediation approach can be used for the first and second quality inspection dimensions. Optionally, a hierarchical quality inspection and remediation approach can be used for the third and fourth quality inspection dimensions, or alternatively, a non-hierarchical approach can be used for the third and fourth quality inspection dimensions (e.g., a parallel quality inspection and remediation approach).

[0098] Optionally, based on the data collected by the sensor, the nth quality control dimension of the (n-1)th annotation result is inspected and repaired to obtain the nth annotation result and the nth set of error elements, including: based on the data collected by the sensor, the Nth quality control dimension of the (n-1)th annotation result is inspected to obtain multiple error elements; the multiple error elements are repaired to obtain the nth annotation result and the nth set of error elements, wherein the nth set of error elements includes the elements among the multiple error elements that were not successfully modified.

[0099] Optionally, based on the data collected by the sensor, at least two of the N quality inspection dimensions of the pre-labeled data are subjected to hierarchical quality inspection and repair to obtain the labeling result, including: based on the data collected by the sensor, all of the N quality inspection dimensions of the pre-labeled data are subjected to hierarchical quality inspection and repair to obtain the labeling result.

[0100] Optionally, based on the data collected by the sensor, hierarchical quality inspection and repair are performed on all quality inspection dimensions among the N quality inspection dimensions of the pre-labeled data, including:

[0101] (1) Based on the data collected by the sensor, the first quality inspection dimension of the pre-labeled data is inspected and repaired to obtain the first labeling result and the first set of error elements. The first set of error elements includes the elements in the pre-labeled data that have been inspected for errors in the first quality inspection dimension and have not been successfully modified.

[0102] …

[0103] (2) Based on the data collected by the sensor, the nth quality inspection dimension of the elements in the (n-1)th annotation result, excluding the first set of erroneous elements to the (n-1)th set of erroneous elements, is inspected and repaired to obtain the nth annotation result and the nth set of erroneous elements. The nth set of erroneous elements includes the elements in the (n-1)th annotation result, excluding the first set of erroneous elements to the (n-1)th set of erroneous elements. The elements that are inspected for errors in the nth quality inspection dimension and have not been successfully modified are included. 1 < n < N and n is an integer.

[0104] …

[0105] (3) Based on the data collected by the sensor, the Nth quality inspection dimension of the elements in the (N-1)th annotation result, excluding the first set of erroneous elements to the (N-1)th set of erroneous elements, is inspected and repaired to obtain the Nth annotation result and the Nth set of erroneous elements. The Nth set of erroneous elements includes the elements in the (N-1)th annotation result, excluding the first set of erroneous elements to the (N-1)th set of erroneous elements, that have been inspected for errors in the Nth quality inspection dimension and have not been successfully modified.

[0106] For example, N is 3, and the pre-labeled data includes 15 elements, such as 5 lanes and 10 lane lines. Based on the data collected by the sensors, the first quality inspection dimension of the 15 elements in the pre-labeled data is inspected, and elements 1, 2, and 3 with quality inspection errors are obtained. Elements 1, 2, and 3 are repaired, resulting in the first labeling result and the first set of erroneous elements. The first set of erroneous elements includes elements 1 and 2, which are the elements among the 15 elements that failed the quality inspection for the first quality inspection dimension and were not successfully repaired. Element 3 was successfully repaired.

[0107] Based on the data collected by the sensor, the second quality inspection dimension of the elements other than elements 1 and 2 in the first annotation result is inspected, and elements 4, 5, and 6 with quality inspection errors are obtained. Elements 4, 5, and 6 are repaired, and the second annotation result and the second set of error elements are obtained. The second set of error elements includes element 6. Element 6 is the element in the first annotation result other than elements 1 and 2 that has failed the quality inspection for the second quality inspection dimension and has not been successfully repaired. Elements 4 and 5 have been successfully repaired.

[0108] Based on the data collected by the sensor, a quality inspection is performed on the third quality inspection dimension of the elements other than elements 1, 2, and 6 in the second annotation result. Elements 7, 8, and 9 with quality inspection errors are obtained. Elements 7, 8, and 9 are repaired to obtain the third annotation result and the third set of error elements. The third set of error elements includes elements 7 and 8. Elements 7 and 8 are the elements in the second annotation result other than elements 1, 2, and 6 that have quality inspection errors in the third quality inspection dimension and have not been successfully repaired. Element 9 has been successfully repaired.

[0109] This completes the hierarchical quality inspection and repair of the pre-labeled data across three quality control dimensions, ultimately yielding the third labeling result: the first set of erroneous elements, the second set of erroneous elements, and the third set of erroneous elements. For example, the first, second, and third sets of erroneous elements can be further repaired manually.

[0110] For example, Figure 5 illustrates a schematic diagram of multi-level quality inspection and repair provided in an embodiment of this application. As shown in Figure 5, after multi-level hierarchical quality inspection and repair of the pre-labeled data based on the data collected by the sensors, quality inspection errors can be automatically repaired.

[0111] Taking pre-labeled data consisting of static features as an example, based on the characteristics of static feature quality issues, such as topological errors typically affecting the quality inspection and repair of subsequent geometric offsets, element attributes, and relationships, and geometric offset errors typically affecting the quality inspection and repair of element attributes and relationships, the following layering principle can be adopted to obtain a systematic and comprehensive quality inspection and repair plan: topological errors can be prioritized over other quality inspection dimensions, and geometric offset errors can be prioritized over element attributes and relationships.

[0112] Based on this principle, quality inspection and repair can be carried out in multiple levels according to the actual situation. For example, quality inspection and repair can be carried out in two levels: quality inspection and repair in the order of topology structure and geometric offset; or quality inspection and repair in the order of topology structure and element attributes; or quality inspection and repair in the order of topology structure and association relationship.

[0113] For example, quality inspection and repair can also be carried out in three layers: quality inspection and repair in the order of topology, geometric offset, and element attributes; or, quality inspection and repair in the order of topology, geometric offset, and association; or, quality inspection and repair in the order of geometric offset, element attributes, and association; or, quality inspection and repair in the order of geometric offset, association, and element attributes.

[0114] For example, quality inspection and repair can also be performed in four layers: quality inspection and repair can be performed in the order of topology structure, geometric offset, association relationship, and element attributes; or, quality inspection and repair can be performed in the order of topology structure, geometric offset, element attributes, and association relationship. For example, Figure 6 shows a schematic diagram of the four-layer quality inspection and repair provided in the embodiment of this application.

[0115] For example, we can first perform hierarchical quality inspection and repair according to the quality inspection order of topology and geometric offset, and then perform quality inspection and repair on element attributes and relationships without hierarchical order (or perform quality inspection and repair on element attributes and relationships in parallel).

[0116] For example, based on the characteristics of the annotation task, quality inspection and repair can be carried out in multiple levels, satisfying the principle that topology class precedes other quality inspection dimensions, and / or geometric offset class precedes element attribute class and association class.

[0117] Optionally, after performing quality inspection and repair on the N quality inspection dimensions of the pre-labeled data, the labeling result can be obtained. The N quality inspection dimensions include the m-th quality inspection dimension, 1≤m≤N and m is an integer. The method 400 further includes: performing quality inspection and repair on the m-th quality inspection dimension of the labeling result based on the data collected by the sensor, to obtain the m'-th labeling result and the m'-th error element set. The m'-th error element set includes elements that have failed to be repaired due to quality inspection errors on the m-th quality inspection dimension.

[0118] For example, after performing quality checks and repairs on the N quality control dimensions of the pre-labeled data, the labeling result can be obtained (e.g., the Nth labeling result mentioned above). The N quality control dimensions of the labeling result can then be checked and repaired again.

[0119] Optionally, the method 400 further includes: performing quality inspection and repair on the N quality inspection dimensions of the annotation result again based on the data collected by the sensor, to obtain another annotation result.

[0120] Optionally, the N quality control dimensions of the annotation result are inspected and repaired again, including: performing hierarchical quality control and repair on the N quality control dimensions of the annotation result again to obtain another annotation result.

[0121] For example, the process of performing layered quality inspection and repair again is as follows:

[0122] (1) Based on the data collected by the sensor, the first quality inspection dimension of the annotation result is inspected and repaired to obtain the first annotation result and the first error element set. The first error element set includes the elements in the annotation result that have been inspected for errors in the first quality inspection dimension and have not been successfully modified.

[0123] …

[0124] (2) Based on the data collected by the sensor, the nth quality inspection dimension of the elements in the (n-1)th annotation result, excluding the 1'th to (n-1)th error element sets, is inspected and repaired to obtain the n'th annotation result and the n'th error element set. The n'th error element set includes the elements in the (n-1)th annotation result, excluding the 1'th to (n-1)th error element sets, that are inspected for errors in the nth quality inspection dimension and have not been successfully modified, where 1 < n < N.

[0125] …

[0126] (3) Based on the data collected by the sensor, the Nth quality inspection dimension of the elements in the N-1'th annotation result, excluding the 1'th to N-1'th error element sets, is inspected and repaired to obtain the N'th annotation result and the N'th error element set. The N'th error element set includes the elements in the N-1'th annotation result, excluding the 1'th to N-1'th error element sets, that have been inspected for errors in the Nth quality inspection dimension and have not been successfully modified.

[0127] Optionally, based on the data collected by the sensor, the nth quality inspection dimension of the (n-1)th annotation result is inspected and repaired to obtain the nth annotation result and the nth set of error elements. This includes: based on the data collected by the sensor, the nth quality inspection dimension of the (n-1)th annotation result is inspected and repaired to obtain the n”th annotation result and the n”th set of error elements, where the n”th set of error elements includes elements in the (n-1)th annotation result that have been inspected and repaired for the nth quality inspection dimension and have not been successfully modified; based on the data collected by the sensor, the nth quality inspection dimension of the n”th annotation result is inspected and repaired to obtain the nth annotation result and the nth set of error elements.

[0128] For example, based on the data collected by the sensor, the second quality inspection dimension of the elements other than elements 1 and 2 in the first annotation result is inspected, and elements 4, 5 and 6 with quality inspection errors are obtained. Elements 4, 5 and 6 are repaired, and the second "" annotation result and the second "" set of error elements are obtained. The second "" set of error elements includes element 6. Element 6 is the element in the first annotation result other than elements 1 and 2 that has failed the quality inspection for the second quality inspection dimension and has not been successfully repaired. Elements 4 and 5 are successfully repaired.

[0129] Based on the data collected by the sensor, the second quality control dimension of the elements other than elements 1 and 2 in the second ""th annotation result is checked again, resulting in elements 4, 5, and 6 with quality control errors. Elements 4, 5, and 6 are then repaired, yielding the second annotation result and the second set of error elements. The second set of error elements includes element 6, which is the element in the first annotation result other than elements 1 and 2 that failed to be repaired in the second quality control dimension. Elements 4 and 5 were successfully repaired. Since the number and ID of error elements in the second ""th set of error elements and the second set of error elements are the same in multiple iterations of quality control and repair for the same quality control dimension, it can be considered that the quality control and repair results for the second quality control dimension have converged. At this point, the second annotation result and the second set of error elements can be output to the third quality control dimension.

[0130] When performing repairs on each quality control dimension, the result of each repair may introduce additional quality control errors or affect the repair of other quality control dimensions. For example, if the first quality control inspection reveals error A and automatically repairs error A, but the repair of error A introduces an additional error B, then a second quality control inspection is needed to find error B and automatically repair it.

[0131] For example, error A and error B could be errors targeting the same element. If construction causes a change in reality, lane lines may need geometric alignment repair. After the geometric alignment repair is complete, if an element attribute quality check reveals an error in the lane line's color attribute, the lane line's color attribute can be repaired again.

[0132] For example, errors A and B could also be errors targeting different elements. For instance, if a lane attribute is incorrectly labeled, a regular lane could be labeled as a tidal flow lane; after fixing this, the lane lines could be re-inspected, and the lane lines should show double yellow dashed lines. The lane line attributes could then be corrected again.

[0133] Therefore, multiple rounds of "quality inspection + repair" can be performed until the quality inspection errors are completely converged, meaning that the number of quality inspection errors no longer decreases or increases. Through iteration, repairable problems are completely fixed, while problems that are completely unrepairable are filtered out.

[0134] For example, Figure 7 illustrates a schematic diagram of multiple quality inspection and repair iterations within the same level provided in an embodiment of this application. As shown in Figure 7, multiple "quality inspection + repair" iterations are performed within the same level.

[0135] For example, based on the data collected by the sensor, the nth quality inspection dimension of the elements in the (n-1)th annotation result, excluding the first set of erroneous elements to the (n-1)th set of erroneous elements, is inspected and repaired to obtain the nth annotation result and the nth set of erroneous elements. The nth set of erroneous elements includes the elements in the (n-1)th annotation result that have been inspected for errors in the nth quality inspection dimension and have not been successfully modified.

[0136] After obtaining the nth annotation result, a second quality check can be performed on the nth quality check dimension of all elements in the nth annotation result excluding the first to (n-1)th error sets. The convergence of the quality check errors is determined based on these errors. If convergence occurs, the nth annotation result and the nth error set are output. Alternatively, if convergence does not occur, the quality check errors are corrected, and a second quality check is performed on the nth quality check dimension of all elements in the corrected annotation result excluding the first to (n-1)th error sets. The convergence is then determined based on these errors. This process is repeated until the quality check errors converge.

[0137] For example, Figure 8 illustrates a schematic diagram of multiple quality inspection and repair iterations across multiple levels provided in the embodiments of this application. As shown in Figure 8, multiple "quality inspection + repair" iterations can also be performed between multiple levels. For example, after quality inspection and repair of the topology of the pre-labeled data, labeling result 1 and error element set 1 can be obtained. Error element set 1 includes elements that failed the quality inspection of the topology and were not successfully repaired. After quality inspection and repair of the geometric offsets of the other elements in labeling result 1 (excluding error element set 1), labeling result 2 and error element set 2 can be obtained. Error element set 2 includes elements in labeling result 1 (excluding error element set 1) that failed the quality inspection of the geometric offset and were not successfully repaired. After quality inspection and repair of the relationships between the other elements in labeling result 2 (excluding error element set 1 and error element set 2), labeling result 3 and error element set 3 can be obtained. Error element set 3 includes elements in labeling result 2 (excluding error element set 1 and error element set 2) that failed the quality inspection of the relationships and were not successfully repaired. After performing quality checks and repairs on the element attributes of all elements in annotation result 3 except for error element sets 1-3, annotation result 4 and error element set 4 can be obtained. Error element set 4 includes elements in annotation result 3 that have errors in element attribute quality checks and have not been successfully repaired. Based on error element sets 1, 2, 3, and 4, it can be determined whether the quality check errors have converged. If they have converged, the repaired annotation result 4 is output; otherwise, the topology, geometric offset, association relationship, and element attribute quality checks and repairs are performed again on annotation result 4.

[0138] For example, methods for determining whether a quality inspection error has "converged" include, but are not limited to, one or more of the following:

[0139] (1) No quality inspection errors occurred, that is, the quality inspection errors were completely repaired during the quality inspection repair process;

[0140] (2) The number of quality inspection errors in the last two inspections remains unchanged, and the IDs of the elements that have quality inspection errors remain unchanged;

[0141] (3) The number of quality inspection errors in recent inspections has shown a regular distribution, for example, the number of quality inspections shows a change of "ABABABA";

[0142] (4) Set a fixed number of iterations, such as m times. After this number of iterations, the quality inspection error has been converged by default.

[0143] At the same time, different iterative combinations can exist within the same level and between multiple levels. For example, it can iterate once within the same level and iterate multiple times between multiple levels; or it can iterate multiple times within the same level and iterate once between multiple levels; or it can iterate multiple times within the same level and iterate multiple times between multiple levels; or it can iterate once within the same level and iterate once between multiple levels, etc.

[0144] For example, Figure 9 shows a schematic diagram of the quality inspection and repair architecture provided in an embodiment of this application.

[0145] (1) Based on the characteristics of static feature annotation, this embodiment of the application performs hierarchical quality inspection and repair on static feature annotation. Each level may include a quality inspection and repair process. The output of each level includes two parts: elements with quality inspection errors and the annotation results after repair. The two output information of each level also serve as the input information of the next level. Specifically, each level of quality inspection is based on the quality inspection error list of the previous level, automatically filtering out elements with quality inspection errors in the previous level (and these elements cannot be repaired) to avoid other invalid association errors caused by one error. At the same time, each level performs quality inspection and repair based on the latest annotation results after automatic repair of the previous level to improve the effectiveness of quality inspection and repair, and also helps to reduce the quality inspection error detection rate.

[0146] (2) Since fixing a single quality inspection error can introduce additional related quality inspection errors or affect the fixing of other quality inspection errors, multiple iterations and fixing iterations can be performed for each quality inspection dimension, or multiple quality inspection dimensions can be subjected to multiple quality inspection and fixing iterations. Through multiple iterations, all fixable quality inspection errors are fixed, while all unfixable quality inspection errors are filtered out.

[0147] Optionally, the method 400 further includes: obtaining the quality inspection order of the N quality inspection dimensions; wherein, performing quality inspection and repair on the N quality inspection dimensions of the pre-labeled data according to the data collected by the sensor includes: performing quality inspection and repair on the N quality inspection dimensions of the pre-labeled data according to the quality inspection order based on the data collected by the sensor.

[0148] Optionally, the quality inspection order is determined by the drawing order of the same element or the same layer in the pre-annotated data.

[0149] Optionally, the order of the N quality inspection dimensions includes at least two of the following: topology, geometric offset, association, and element attributes.

[0150] For example, as shown in Figure 6, the order of the N quality inspection dimensions can be topology, geometric offset, association, and element attributes.

[0151] Optionally, the quality inspection order is determined by the drawing order among multiple layers in the pre-annotated data.

[0152] Optionally, the order of the N quality inspection dimensions includes at least two of the lane line layer, lane layer, and road layer.

[0153] The inventors of this application have discovered that static feature annotations typically have clearly defined drawing orders for different layers. Therefore, multi-level iterative quality inspection and repair can be performed based on the annotation layers as a layering principle. If the drawing order requires drawing the boundary layer first, then the lane layer, and finally the road layer, then the corresponding layer-level quality inspection and repair order can be: quality inspection and repair in the order of lane layer, lane layer, and road layer. For example, Figure 10 shows a schematic diagram of the three-layer quality inspection and repair provided in an embodiment of this application.

[0154] For example, based on the characteristics of the annotation task and while adhering to the principle of a "specific annotation order," quality inspection and repair can be flexibly carried out in multiple levels. For instance, these multiple layers may also include a curb layer and a road surface annotation layer, which can be inspected and repaired in the following order: curb layer, lane line layer, lane layer, road layer; or, lane line layer, lane layer, road marking layer, road layer.

[0155] Optionally, the pre-labeled data is the labeling data for static features.

[0156] This application provides a method for quality inspection and repair of static data for autonomous driving. The method includes: acquiring pre-labeled data and data collected by sensors; performing quality inspection on N quality inspection dimensions of the pre-labeled data based on the data collected by the sensors, wherein the N quality inspection dimensions include an nth quality inspection dimension and an (n+1)th quality inspection dimension, and the quality inspection result of the nth quality inspection dimension is the input of the (n+1)th quality inspection dimension; wherein N > 1 and N is an integer, 1 ≤ n < N and n is an integer.

[0157] Optionally, based on the data collected by the sensor, quality checks are performed on N quality control dimensions of the pre-labeled data, including: based on the data collected by the sensor, quality checks are performed on the nth quality control dimension of the (n-1)th labeling result to obtain the nth error element set, which includes elements that have errors in quality control for the nth quality control dimension; based on the data collected by the sensor, quality checks are performed on the (n+1)th quality control dimension of the elements in the (n-1)th labeling result other than the nth error element set to obtain the (n+1)th error element set, which includes elements in the (n-1)th labeling result other than the nth error element set that have errors in quality control for the (n+1)th quality control dimension; wherein, the (n-1)th labeling result is the pre-labeled data; or, the (n-1)th labeling result includes the results obtained after quality checks on the 1st to (n-1)th quality control dimensions.

[0158] Optionally, the method further includes: performing data repair based on the nth set of erroneous elements and the (n+1)th set of erroneous elements.

[0159] For example, when performing quality inspection on N quality inspection dimensions, a hierarchical quality inspection approach can be adopted. For instance, the N quality inspection dimensions include quality inspection dimension a, quality inspection dimension b, and quality inspection dimension c.

[0160] (1) Based on the data collected by the sensor, the quality inspection dimension a of the pre-labeled data is inspected to obtain the set of error elements a. The set of error elements a includes the elements that are inspected for errors in the quality inspection dimension a among the multiple elements of the pre-labeled data.

[0161] (2) Based on the data collected by the sensor, the quality inspection dimension b of the elements in the pre-labeled data other than the error element set a is inspected to obtain the error element set b. The error element set b includes the elements in the pre-labeled data other than the error element set a that have been inspected for the error in the quality inspection dimension b.

[0162] (3) Based on the data collected by the sensor, the quality inspection dimension c of the elements other than the error element set a and the error element set b in the pre-labeled data is inspected to obtain the error element set c. The error element set c includes the elements in the pre-labeled data other than the error element set a and the error element set b that have been inspected for errors in the quality inspection dimension c.

[0163] After obtaining the sets of erroneous elements a, b, and c, repairs can be performed to obtain the repaired annotation results. In this way, through multiple levels of quality inspection, the false alarm rate of quality inspection errors can be reduced.

[0164] Figure 11 shows a schematic block diagram of an autonomous driving static data quality inspection and repair device 1100 provided in an embodiment of this application. The device 1100 includes: an acquisition unit 1110, used to acquire pre-labeled data and data collected by sensors; and a data quality inspection and repair unit 1120, used to perform quality inspection and repair on N quality inspection dimensions of the pre-labeled data based on the data collected by the sensors, wherein the N quality inspection dimensions include an nth quality inspection dimension and an (n+1)th quality inspection dimension, and the quality inspection and repair result of the nth quality inspection dimension is the input of the (n+1)th quality inspection dimension; where N > 1 and N is an integer, 1 ≤ n < N and n is an integer.

[0165] Optionally, the data quality inspection and repair unit 1120 is specifically used for: performing quality inspection and repair on the nth quality inspection dimension of the (n-1)th annotation result based on the data collected by the sensor, to obtain the nth annotation result and the nth set of erroneous elements, wherein the nth set of erroneous elements includes elements that failed to be modified for the nth quality inspection dimension; and performing quality inspection and repair on the (n+1)th quality inspection dimension of the elements in the nth annotation result other than the nth set of erroneous elements based on the data collected by the sensor, to obtain the (n+1)th... There are two annotation results and a set of (n+1)th error elements. The (n+1)th error element set includes elements from the nth annotation result other than the nth error element set that failed to be corrected for the (n+1)th quality inspection dimension. When n is 1, the (n-1)th annotation result includes multiple elements from the pre-annotated data. Alternatively, when 1 < n < N and n is an integer, the (n-1)th annotation result includes the annotation results obtained after quality inspection and correction for the 1st to (n-1)th quality inspection dimensions.

[0166] Optionally, after performing quality inspection and repair on the N quality inspection dimensions of the pre-labeled data, the labeling result can be obtained. The N quality inspection dimensions include the m-th quality inspection dimension, 1≤m≤N and m is an integer. The data quality inspection and repair unit 1120 is further used to: perform quality inspection and repair on the m-th quality inspection dimension of the labeling result based on the data collected by the sensor, to obtain the m'-th labeling result and the m'-th error element set. The m'-th error element set includes elements that have failed to be repaired due to quality inspection errors on the m-th quality inspection dimension.

[0167] Optionally, the data quality inspection and repair unit 1120 is specifically used for: performing quality inspection and repair on the nth quality inspection dimension of the (n-1)th labeling result based on the data collected by the sensor, to obtain the n”th labeling result and the n”th error element set, wherein the n”th error element set includes elements in the (n-1)th labeling result that have quality inspection errors for the nth quality inspection dimension and have not been successfully modified; and performing quality inspection and repair on the n”th quality inspection dimension of the n”th labeling result based on the data collected by the sensor, to obtain the nth labeling result and the nth error element set.

[0168] Optionally, the acquisition unit 1110 is further configured to acquire the quality inspection order of the N quality inspection dimensions; wherein, the data quality inspection and repair unit 1120 is specifically configured to: perform quality inspection and repair on the N quality inspection dimensions of the pre-labeled data according to the quality inspection order based on the data collected by the sensor.

[0169] Optionally, the quality inspection order is determined by the drawing order of the same element or the same layer in the pre-annotated data.

[0170] Optionally, the order of the N quality inspection dimensions includes at least two of the following: topology, geometric offset, association, and element attributes.

[0171] Optionally, the quality inspection order is determined by the drawing order among multiple layers in the pre-annotated data.

[0172] Optionally, the order of the N quality inspection dimensions includes at least two of the lane line layer, lane layer, and road layer.

[0173] Optionally, the pre-labeled data is the labeling data for static features.

[0174] Those skilled in the art will recognize that the units and algorithm steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of this application.

[0175] It should be understood that the division of units in the above device is only a logical functional division. In actual implementation, they can be fully or partially integrated into a single physical entity, or they can be physically separated. Furthermore, the units in the device can be implemented by a processor calling software; for example, the device includes a processor connected to memory, which stores instructions. The processor calls the instructions stored in memory to implement any of the above methods or to implement the functions of each unit in the device. The processor can be, for example, a general-purpose processor, such as a CPU or microprocessor, and the memory can be internal or external to the device. Alternatively, the units in the device can be implemented as hardware circuits. The functions of some or all units can be implemented through the design of the hardware circuits, which can be understood as one or more processors. For example, in one implementation, the hardware circuit is an ASIC, and the functions of some or all units are implemented through the design of the logical relationships between the components within the circuit. In another implementation, the hardware circuit can be implemented using a PLD, such as an FPGA, which can include a large number of logic gates. The connection relationships between the logic gates are configured through configuration files, thereby implementing the functions of some or all units. All units of the above device can be implemented entirely through processor calling software, or entirely through hardware circuits, or partially through processor calling software and the remaining parts through hardware circuits.

[0176] In this application embodiment, a processor is a circuit with signal processing capabilities. In one implementation, the processor can be a circuit with instruction reading and execution capabilities, such as a CPU, microprocessor, GPU, or DSP. In another implementation, the processor can implement certain functions through the logical relationships of hardware circuits. These logical relationships are fixed or reconfigurable. For example, the processor may be a hardware circuit implemented as an ASIC or PLD, such as an FPGA. In a reconfigurable hardware circuit, the process of the processor loading a configuration document and configuring the hardware circuit can be understood as the processor loading instructions to implement the functions of some or all of the above units. Furthermore, it can also be a hardware circuit designed for artificial intelligence, which can be understood as an ASIC, such as an NPU, TPU, or DPU.

[0177] As can be seen, each unit in the above device can be one or more processors (or processing circuits) configured to implement the above methods, such as: CPU, GPU, NPU, TPU, DPU, microprocessor, DSP, ASIC, FPGA, or a combination of at least two of these processor forms.

[0178] Furthermore, the units in the above devices can be integrated in whole or in part, or they can be implemented independently. In one implementation, these units are integrated together as a System-on-a-Chip (SoC). The SoC may include at least one processor for implementing any of the above methods or implementing the functions of the units in the device. The at least one processor may be of different types, such as CPU and FPGA, CPU and AI processor, CPU and GPU, etc.

[0179] This application also provides a data quality inspection and repair device, which includes a processing unit and a storage unit. The storage unit is used to store instructions, and the processing unit executes the instructions stored in the storage unit to enable the device to perform the methods or steps described in the above embodiments.

[0180] This application also provides a data quality inspection and repair device, which includes the above-mentioned quality inspection and repair apparatus 1100.

[0181] For example, the data quality inspection and repair equipment can be a server, or other equipment with data quality inspection and repair capabilities (e.g., computer equipment).

[0182] This application also provides a computer program product, which includes computer program code that, when run on a computer, causes the computer to perform the methods described in the above embodiments.

[0183] This application also provides a computer-readable medium storing program code that, when run on a computer, causes the computer to perform the methods described in the above embodiments.

[0184] This application also provides a chip, which includes a circuit for performing the methods described in the above embodiments.

[0185] In implementation, each step of the above method can be completed by integrated logic circuits in the processor's hardware or by instructions in software. The method disclosed in the embodiments of this application can be directly implemented by a hardware processor, or by a combination of hardware and software modules within the processor. The software modules can reside in random access memory, flash memory, read-only memory, programmable read-only memory, power-on erasable programmable memory, registers, or other mature storage media in the art. This storage medium is located in memory, and the processor reads information from the memory and, in conjunction with its hardware, completes the steps of the above method. To avoid repetition, detailed descriptions are omitted here.

[0186] It should be understood that in the embodiments of this application, the memory may include read-only memory and random access memory, and provides instructions and data to the processor.

[0187] It should also be understood that, in the various embodiments of this application, the order of the above-mentioned processes does not imply the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of this application.

[0188] Those skilled in the art will recognize that the units and algorithm steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of this application.

[0189] Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the specific working processes of the systems, devices, and units described above can be referred to the corresponding processes in the foregoing method embodiments, and will not be repeated here.

[0190] In the several embodiments provided in this application, it should be understood that the disclosed systems, apparatuses, and methods can be implemented in other ways. For example, the apparatus embodiments described above are merely illustrative; for instance, the division of units is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the coupling or direct coupling or communication connection shown or discussed may be through some interfaces; the indirect coupling or communication connection between apparatuses or units may be electrical, mechanical, or other forms.

[0191] The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of this embodiment according to actual needs.

[0192] In addition, 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.

[0193] If the aforementioned functions are implemented as software functional units and sold or used as independent products, they 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 a portion 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.) to execute all or part of the steps of the methods described in the 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.

[0194] The above description is merely a specific embodiment of this application, but the scope of protection of this application is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the technical scope disclosed in this application should be covered. Therefore, the scope of protection of this application should be determined by the scope of the claims.

Claims

1. A method for quality inspection and repair of static data in autonomous driving, characterized in that, include: Acquire pre-labeled data and data collected by sensors; Based on the data collected by the sensor, the N quality inspection dimensions of the pre-labeled data are inspected and repaired. The N quality inspection dimensions include the nth quality inspection dimension and the (n+1)th quality inspection dimension. The quality inspection and repair result of the nth quality inspection dimension is the input of the (n+1)th quality inspection dimension. N > 1 and N is an integer, 1 ≤ n < N and n is an integer.

2. The method according to claim 1, characterized in that, The step of performing quality inspection and repair on N quality inspection dimensions of the pre-labeled data based on the data collected by the sensor includes: Based on the data collected by the sensor, the nth quality inspection dimension of the (n-1)th annotation result is inspected and repaired to obtain the nth annotation result and the nth error element set. The nth error element set includes elements that have failed to be modified successfully in the quality inspection of the nth quality inspection dimension. Based on the data collected by the sensor, the (n+1)th quality inspection dimension of the elements other than the nth erroneous element set in the nth annotation result is inspected and repaired to obtain the (n+1)th annotation result and the (n+1)th erroneous element set. The (n+1)th erroneous element set includes the elements in the nth annotation result other than the nth erroneous element set that have been inspected for errors in the (n+1)th quality inspection dimension and have not been successfully repaired. Wherein, when n is 1, the (n-1)th annotation result includes multiple elements from the pre-annotated data; or, When 1 < n < N, the (n-1)th annotation result includes the annotation results obtained after quality inspection and repair for the 1st to (n-1)th quality inspection dimensions.

3. The method according to claim 2, characterized in that, After performing quality inspection and repair on the Nth quality inspection dimension, the annotation result is obtained. The N quality inspection dimensions include the mth quality inspection dimension, 1≤m≤N and m is an integer. The method further includes: Based on the data collected by the sensor, the m-th quality inspection dimension of the annotation result is inspected and repaired to obtain the m'-th annotation result and the m'-th error element set. The m'-th error element set includes elements that failed to be repaired for the m-th quality inspection dimension.

4. The method according to claim 2 or 3, characterized in that, The step of performing quality inspection and repair on the nth quality inspection dimension of the (n-1)th labeled result based on the data collected by the sensor includes: Based on the data collected by the sensor, the nth quality inspection dimension of the (n-1)th annotation result is inspected and repaired to obtain the n”th annotation result and the n”th error element set. The n”th error element set includes the elements in the (n-1)th annotation result that have been inspected for the nth quality inspection dimension and have not been successfully modified. Based on the data collected by the sensor, the nth quality inspection dimension of the nth labeled result is inspected and repaired to obtain the nth labeled result and the set of nth erroneous elements.

5. The method according to any one of claims 1 to 4, characterized in that, The method further includes: Obtain the quality inspection order of the N quality inspection dimensions; The step of performing quality inspection and repair on N quality inspection dimensions of the pre-labeled data based on the data collected by the sensor includes: Based on the data collected by the sensor, and in accordance with the quality inspection sequence, the N quality inspection dimensions of the pre-labeled data are inspected and repaired.

6. The method according to claim 5, characterized in that, The quality inspection order is determined by the drawing order of the same element or the same layer in the pre-labeled data.

7. The method according to claim 6, characterized in that, The order of the N quality inspection dimensions includes at least two of the following: topology, geometric offset, association, and element attributes.

8. The method according to claim 5, characterized in that, The quality inspection order is determined by the drawing order among multiple layers in the pre-labeled data.

9. The method according to claim 8, characterized in that, The order of the N quality inspection dimensions includes at least two of the lane line layer, lane layer, and road layer.

10. The method according to any one of claims 1 to 9, characterized in that, The pre-labeled data is the labeling data for static features.

11. A device for quality inspection and repair of static data for autonomous driving, characterized in that, include: The acquisition unit is used to acquire pre-labeled data and data collected by the sensor; The data quality inspection and repair unit is used to inspect and repair N quality inspection dimensions of the pre-labeled data based on the data collected by the sensor. The N quality inspection dimensions include the nth quality inspection dimension and the (n+1)th quality inspection dimension. The quality inspection and repair result of the nth quality inspection dimension is the input of the (n+1)th quality inspection dimension. N > 1 and N is an integer, 1 ≤ n < N and n is an integer.

12. The apparatus according to claim 11, characterized in that, The data quality inspection and repair unit is specifically used for: Based on the data collected by the sensor, the nth quality inspection dimension of the (n-1)th annotation result is inspected and repaired to obtain the nth annotation result and the nth error element set. The nth error element set includes elements that have failed to be modified successfully in the quality inspection of the nth quality inspection dimension. Based on the data collected by the sensor, the (n+1)th quality inspection dimension of the elements other than the nth erroneous element set in the nth annotation result is inspected and repaired to obtain the (n+1)th annotation result and the (n+1)th erroneous element set. The (n+1)th erroneous element set includes the elements in the nth annotation result other than the nth erroneous element set that have been inspected for errors in the (n+1)th quality inspection dimension and have not been successfully repaired. Wherein, when n is 1, the (n-1)th annotation result includes multiple elements from the pre-annotated data; or, When 1 < n < N, the (n-1)th annotation result includes the annotation results obtained after quality inspection and repair for the 1st to (n-1)th quality inspection dimensions.

13. The apparatus according to claim 12, characterized in that, After performing quality inspection and repair on the Nth quality inspection dimension, the annotation result is obtained. The N quality inspection dimensions include the mth quality inspection dimension, 1≤m≤N and m is an integer. The data quality inspection and repair unit is further configured to: perform quality inspection and repair on the m-th quality inspection dimension of the annotation result based on the data collected by the sensor, to obtain the m'-th annotation result and the m'-th error element set, wherein the m'-th error element set includes elements that have failed to be repaired for the m-th quality inspection dimension.

14. The apparatus according to claim 12 or 13, characterized in that, The data quality inspection and repair unit is specifically used for: Based on the data collected by the sensor, the nth quality inspection dimension of the (n-1)th annotation result is inspected and repaired to obtain the n”th annotation result and the n”th error element set. The n”th error element set includes the elements in the (n-1)th annotation result that have been inspected for the nth quality inspection dimension and have not been successfully modified. Based on the data collected by the sensor, the nth quality inspection dimension of the nth labeled result is inspected and repaired to obtain the nth labeled result and the set of nth erroneous elements.

15. The apparatus according to any one of claims 11 to 14, characterized in that, The acquisition unit is also used to acquire the quality inspection order of the N quality inspection dimensions; Specifically, the data quality inspection and repair unit is used for: Based on the data collected by the sensor, and in accordance with the quality inspection sequence, the N quality inspection dimensions of the pre-labeled data are inspected and repaired.

16. The apparatus according to claim 15, characterized in that, The quality inspection order is determined by the drawing order of the same element or the same layer in the pre-labeled data.

17. The apparatus according to claim 16, characterized in that, The order of the N quality inspection dimensions includes at least two of the following: topology, geometric offset, association, and element attributes.

18. The apparatus according to claim 15, characterized in that, The quality inspection order is determined by the drawing order among multiple layers in the pre-labeled data.

19. The apparatus according to claim 18, characterized in that, The order of the N quality inspection dimensions includes at least two of the lane line layer, lane layer, and road layer.

20. The apparatus according to any one of claims 11 to 19, characterized in that, The pre-labeled data is the labeling data for static features.

21. A device for quality inspection and repair of static data for autonomous driving, characterized in that, include: A processor for executing a computer program stored in memory to cause the apparatus to perform the method as described in any one of claims 1 to 10.

22. The apparatus according to claim 21, characterized in that, The device also includes the memory.

23. A computer-readable storage medium, characterized in that, It stores instructions that, when executed by a processor, cause the processor to implement the method as described in any one of claims 1 to 10.

24. A computer program product, characterized in that, The computer program product includes computer program code that, when run on a computer, causes the computer to perform the method as described in any one of claims 1 to 10.

25. A chip, characterized in that, The chip includes circuitry for performing the method as described in any one of claims 1 to 10.

Images