Method and apparatus for correcting a digital map

Abstract

The invention relates to a method and a device for correcting a digital map. The method comprises the following steps: performing a calibration phase by selecting a first map sheet of the digital map; checking the first map sheet for errors and correcting found errors in the first map sheet in a first correction level; checking the corrected first map sheet for errors and correcting found errors in the corrected first map sheet in a further correction level; repeating the further correction level until less than a predefined number of errors are found in at least two correction levels following each other, in particular no more errors are found, and in this way a first number of correction levels for the first map sheet from the calibration phase is determined; determining a second number of correction levels necessary for the remaining map sheets based on the first number of correction levels performed in the calibration phase for the first map sheet; wherein the remaining map sheets of the digital map are checked and corrected by the determined second number of correction levels.

Description

Technical Field

[0001] The present invention relates to a method for calibrating digital maps and an apparatus for performing the method. Background Technology

[0002] Existing technologies in the field of error identification and correction of digital maps, especially high-definition maps, include various schemes for checking and improving map quality.

[0003] A known method, as described in US 2023 / 0137111 A1, describes determining the frequency of inspections for a map segment based on the frequency of errors and the time until the next encounter with that segment. Here, manual inspection of the segments is planned, which improves map quality but requires a high amount of manual labor. Summary of the Invention

[0004] The objective of this invention is to provide a method and system for correcting digital maps, especially high-definition maps, which enables efficient and accurate determination of error probabilities.

[0005] This invention relates to a method for calibrating a digital map, the method comprising the steps of: performing a calibration phase by: selecting a first map segment of the digital map; in a first calibration level, checking the first map segment for errors and correcting any errors found in the first map segment; in a further calibration level, checking the calibrated first map segment for errors and correcting any errors found in the calibrated first map segment; repeating the further calibration levels until fewer than a predetermined number of errors are found in at least two successive calibration levels, in particular no more errors are found, and thereby determining a first number of calibration levels performed on the first map segment from the calibration phase; based on the first number of calibration levels performed on the first map segment in the calibration phase, determining a second number of calibration levels necessary for the remaining map segments, wherein the second number of necessary calibration levels is at least one less calibration level, preferably at least three less, than the first number of calibration levels performed in the calibration phase; wherein each of the remaining map segments of the digital map is checked and calibrated using the determined second number of calibration levels.

[0006] The first map segment refers to the selected, defined portion of a digital map for the calibration phase. A digital map is a map representation stored electronically, which may contain geographic, structural, or functional information. Errors are deviations from the defined target state or from a pre-given reference, which may affect the accuracy or functionality of the digital map. Correction stages are iterative steps used to check and eliminate errors in the map segments. The remaining map segments are the parts of the digital map that are checked and corrected after the calibration phase.

[0007] In this method, the detection rate for each correction level can be calculated using the number of errors identified and eliminated in each correction level. This detection rate describes the percentage of remaining errors that can be corrected at each level. For example, a detection rate of 90% for each correction level means that 90% of the remaining errors are identified and corrected at each level.

[0008] Additionally, this method can be used to determine the initial number of errors per map length, such as per 100 km, in the initial map. Based on these two variables—the detection rate and the initial error density—it is possible to calculate how many correction levels are needed to reach a determined maximum error frequency, which can then be used as a quality target for validation.

[0009] The calculation is illustrated by the following exemplary implementation of the method: If the initial map has an error density of "1000 errors / 100km" and the detection rate reaches 90% at each correction level, the quality target of "maximum 5 errors / 100km" can be achieved, exemplarily, through three correction levels. Therefore, after the first level, there are "100 errors / 100km" remaining; after the second level, there are "10 errors / 100km" remaining; and after the third level, the number of errors decreases to "1 error / 100km," thus exceeding the target.

[0010] This method significantly improves the efficiency of the verification process because it allows for the targeted determination of the number of correction steps needed to achieve the desired error frequency. This leads to optimized resource utilization and improvements in map verification for vehicle navigation.

[0011] The advantage of this method is that the correction levels are specifically adjusted according to the map's specific requirements, thereby improving inspection efficiency and simultaneously enhancing map quality. The inspection workload is minimized by reducing the number of correction levels used for remaining map fragments, without compromising accuracy.

[0012] Advantageously, the calibration phase can be performed separately for each map type or for specific map elements of the digital map in order to optimize the inspection load.

[0013] Map types refer to the different categories of digital maps, characterized by their specific features, such as urban areas or forest areas. Map elements are specific features or structures within a digital map.

[0014] This ensures that the inspection process matches the corresponding map type or specific map element, which improves the accuracy and efficiency of the correction.

[0015] Advantageously, the correction can be performed by a person or human inspector, who compares the first map segment with a comparison map segment in at least one feature and manually corrects the deviation between the first map segment and the comparison map segment in the first map segment, especially with the aid of a graphical user interface.

[0016] Human inspectors are personnel who check map fragments for errors and correct them. A graphical user interface (GUI) is a digital tool that allows inspectors to visually analyze and adjust map fragments.

[0017] Therefore, accurate corrections can be ensured by using human experts, especially in cases of complex map fragments or specific error types that are difficult to identify automatically.

[0018] Advantageously, the correction level can be implemented automatically by an artificial intelligence model, wherein the artificial intelligence model examines the first map segment in at least one feature and corrects errors in the first map segment.

[0019] An artificial intelligence (AI) model is an algorithmic system developed based on machine learning to automatically learn and perform defined tasks through the analysis of data. In the current context, the AI ​​model is trained to identify and classify errors in digital maps. This AI model uses the results of earlier correction levels as training data to identify patterns and typical errors in the map data. Through this learning process, the AI ​​model is able to perform checks increasingly efficiently and accurately. Here, the AI ​​model operates independently and is able to continuously improve its prediction and verification of errors, leading to the automation of the inspection process.

[0020] This results in high processing speed and the ability to efficiently analyze and correct large-area map fragments.

[0021] Advantageously, in each individual correction level, errors can be checked for in the corresponding map segments of the digital map by multiple human inspectors using a graphical user interface or by multiple artificial intelligence models, in order to improve the accuracy of the inspection, wherein the error rate is estimated by applying statistical methods to the results of different inspectors or artificial intelligence models, preferably by averaging.

[0022] Statistical methods are a mathematical technique used to analyze and interpret data in order to calculate, for example, an accurate error rate by averaging.

[0023] Advantageously, the correction stages can be performed sequentially at regular time intervals.

[0024] Advantageously, the software platform can coordinate multiple inspections and aggregate the results from individual inspectors or individual artificial intelligence models.

[0025] The software platform is a central application that organizes the tests and collects and merges the different test results.

[0026] As a result, the inspection structure is made more efficient by coordinating and analyzing different results centrally.

[0027] Advantageously, errors can be stored in a database and used for further analysis.

[0028] A database is a storage system that structurally stores collected error data so that it can be used for future analysis and optimization.

[0029] Therefore, erroneous data can be used for future verification or for analyzing error trends, which helps to continuously improve map quality.

[0030] Advantageously, the artificial intelligence model can be optimized through machine learning, where the results of each calibration level are used as training data for the artificial intelligence model, thereby improving future calibration stages.

[0031] Machine learning is a method in which artificial intelligence models process data to identify patterns and continuously improve their performance based on these patterns. Training data is information that is provided to the model to improve the accuracy of its predictions and actions.

[0032] This ensures that the AI ​​model works more accurately by learning from the experience collected at the calibration stage, which improves the efficiency and accuracy of future calibration phases.

[0033] Advantageously, map segments for the calibration phase can be selected based on their complexity, particularly map segments in urban centers that are more complex than those in forest areas or parking lots, and are therefore selected accordingly.

[0034] Complexity describes the difficulty of map elements, derived from the number and diversity of features within a map fragment. Typically, map fragments in city centers contain more detail, such as intersections, street names, and POIs (points of interest), while map fragments in forest areas are less detailed.

[0035] Therefore, the calibration phase focuses on more challenging segments, which improves the effectiveness and target accuracy of the entire method.

[0036] Advantageously, the results of the calibration phase can be displayed in a visualization of the calibrated digital map.

[0037] Visualization is a graphical representation of a calibrated map, showing its state after calibration.

[0038] This allows users to trace the corrections that have been performed and check their quality, which improves the transparency and usability of the method.

[0039] Advantageously, the digital map can have at least two different region types, wherein the calibration phase is performed on two regions of different types, and different numbers of necessary correction levels obtained during the calibration phase are applied to the corresponding region types of the digital map.

[0040] A region refers to a geographically or functionally subdivided part of a digital map, such as an urban area, a rural area, or an industrial zone. Different types of regions have characteristic features that must be taken into account during the labeling process.

[0041] This ensures that the correction is optimally matched to the specific requirements of the corresponding region, which improves the accuracy and efficiency of the method.

[0042] Advantageously, an estimate or calculation of the remaining error rate or error probability can be obtained after each correction stage, so as to determine whether a predetermined value of the error rate has been reached.

[0043] Error rate indicates "how many errors remain in a map segment", while error probability is an estimate of the probability that "errors will continue to occur".

[0044] This allows for targeted control and optimization of the correction stage, avoiding unnecessary correction stages and enabling the method to operate more efficiently.

[0045] Advantageously, all map features and / or feature groups can be evaluated and corrected in a single correction method, or the correction method, including the calibration phase, can be repeated separately for each individual map feature and / or for each individual feature group.

[0046] Map features are specific elements of a map, such as streets, buildings, or bodies of water, while feature groups are combinations of similar map features.

[0047] This allows for flexible adjustments to the correction method, enabling both global and targeted corrections, thus improving the method's adaptability.

[0048] Advantageously, an estimate or calculation of the remaining error rate or error probability can be obtained after each correction stage, so as to determine whether a predetermined error rate value, preferably up to 5%, has been reached.

[0049] The estimate of the remaining error rate is based on the results of each correction level and indicates "how many errors still exist".

[0050] This allows for early determination of whether the target error rate has been achieved, avoiding unnecessary further checks.

[0051] The AI ​​model is used to perform correction levels and automatically identify errors in digital map fragments. The AI ​​model learns from the correction levels so far in order to more accurately identify future errors.

[0052] Using multiple individually trained AI models allows for the targeted identification of different error categories in digital maps. Each AI model is trained for a specific error category, such as location errors, missing objects, or incorrect attribute assignments. Through this targeted specificity, these models can reliably identify and correct different error types independently of each other.

[0053] Compared to a single AI model trained for multiple error categories, this approach significantly improves the accuracy of error identification and correction. Individually trained AI models, through their specialized architecture, are particularly adept at identifying specific features corresponding to their respective error categories. The independent models work in parallel and complement each other, resulting in more comprehensive and accurate map verification.

[0054] Another aspect of the present invention is an apparatus for performing the aforementioned method for correcting digital maps, the apparatus comprising a calibration unit, an analysis and evaluation unit, and a correction unit. These components work together to efficiently analyze, calibrate, and correct map fragments.

[0055] The calibration unit is a component of the device used to perform the calibration phase. During this phase, a first map segment is selected, and a determined number of correction levels are performed on this first map segment. Each correction level is an iterative step that identifies and eliminates errors in the map segment.

[0056] The analysis and evaluation unit analyzes the results of the calibration phase, particularly the number of correction levels performed in the first map segment. Based on these results, the analysis and evaluation unit calculates the number of correction levels required for the remaining map segments.

[0057] The calibration unit applies the calculated number of calibration levels to the remaining map segments. Here, the remaining map segments are all parts of the digital map that were not calibrated during the calibration phase.

[0058] The advantage of this device lies in its ability to optimize the number of necessary correction stages based on the results of the calibration phase. This makes the correction of the remaining map fragments more efficient and resource-saving. Attached Figure Description

[0059] This invention will be explained with reference to the following figures: Figure 1 A schematic diagram of a method for calibrating digital maps; Figure 2 A diagram illustrating the subdivision of a digital map into segments for calibration and segments for routine inspection; Figure 3 A graph showing the number of map errors found during the calibration phase of randomly selected map features; Figure 4 A schematic diagram of one embodiment of an apparatus for performing the method is shown. Detailed Implementation

[0060] Figure 1 This invention describes a method for correcting digital maps, particularly for individual map features or groups of map features. The method includes multiple steps that systematically demonstrate error checking and error reduction.

[0061] The first step is map creation, which involves creating a digital map. In this step 1, the raw data is merged into digital map fragments. This map contains geographical features relevant to the application, such as for autonomous driving. This step forms the basis for all subsequent checks and error corrections.

[0062] Subsequently, in another step 2, a first error estimate is performed for each map feature, or for each group of map features. This first estimate allows for a preliminary assessment of the probability of errors in the digital map. The error rate indicates "how many errors are presumed" and serves as the basis for further correction steps.

[0063] In calibration phase 3 (calibrating the necessary number of correction levels), the necessary number of correction levels for the selected first map segment of the map is determined. Here, multiple checks are performed at the first correction level 4, the second correction level 5, the third correction level 6, and so on up to the penultimate correction level 7 and the final correction level 8, where the error rates 9, 10, 11, 12, and 13 are calculated for each correction level, until no further errors 12 or 13 are found in two consecutive correction levels 7 and 8. This step ensures that the final number of correction levels is sufficient to achieve the determined error rate. This phase determines how many correction levels are needed to minimize the error rate as much as possible.

[0064] Following calibration, in another step 14, the average error rate is calculated using statistical methods. Subsequently, in another step 15, the remaining necessary correction levels are estimated, in which the number of correction levels necessary for the remaining map segments of the digital map is determined. Based on the calibration, the number of correction levels for these remaining segments is reduced such that this number is at least one level less, preferably three levels less, than the number of correction levels performed in the first map segment, in order to minimize the inspection load without exceeding the determined error rate.

[0065] In the next step 16, correction levels 17, 18, and 19 are performed for each of the remaining map fragments, with the required number obtained. Here, either a human inspector or an artificial intelligence model trained on the training data can be used to check the map for errors and achieve the determined error rate.

[0066] Finally, in another step 20, error correction is performed, in which the identified errors are corrected either manually by an inspector through a graphical user interface or automatically by another artificial intelligence model.

[0067] Figure 2 The method for subdividing a digital map into segments for calibration and segments for routine inspection is further explained. This method enables efficient calibration of the map by selectively dividing it into distinct map segments that can be inspected and calibrated individually.

[0068] First, the digital map 30, containing road directions 31 and intersections 32, is subdivided into multiple map segments, each representing a specific area of ​​the map. This subdivision allows for targeted application of calibration and inspection processes to each segment of the map. The size and cropping style (Zuschnitt) of the map segments can vary depending on their purpose and geographic relevance.

[0069] First, first map segments 33 are selected for calibration; these segments are used to determine the calibration correction levels. During the calibration phase, these segments are examined to determine the necessary number of correction levels. These calibration segments 33 serve as representative portions of the entire map 30, and it is ensured that the error rate in these segments is reduced to such a degree that this error rate can also be applied to the remaining second segments 34. During calibration, it is determined how many correction levels are needed to achieve the determined error rate, preferably a maximum of 5%.

[0070] After calibration, the remaining regular segments 34 of the map are checked, but with a reduced number of correction levels compared to the calibrated segments. The number of correction levels in the regular segments 34 is determined based on the results of the calibration phase. Here, the number of correction levels used for the regular segments is minimized to reduce the workload of the inspection while retaining the determined error rate.

[0071] Figure 2 It is also shown that map segmentation can be done not only spatially, but also in terms of the corresponding map content. Depending on the importance and geographical relevance of the segments, these segments can be treated in different ways. For example, the examination of more important first segments 33, such as urban areas, can be more detailed than the examination of less important segments 34, such as segments in rural or remote areas.

[0072] Maps are divided into segments based on different criteria, such as the geographical relevance of the segments, the frequency of use of map segments, or security relevance.

[0073] Therefore, during the calibration phase, the first segment 33 is subjected to a more rigorous check in order to determine how many correction levels are necessary to achieve the determined error rate, preferably, the remaining error is at most 5% compared to the error of the first correction level.

[0074] Typically, the number of correction levels for the second segment 34 is determined such that it is at least one level less, preferably three levels less, than the number of levels performed in the first calibration segment 33. This reduces the workload of verification without affecting the accuracy of the correction.

[0075] Here, the calibration allocation follows a 5% to 95% ratio: approximately 5% of the total map area, shown by the first calibration segment 33, is used for rigorous calibration, while the remaining 95% of the map, shown by the second regular segment 34, is validated through a reduced level of calibration. This ensures efficient checking of the entire map, as the calibration results from the first segment are representative of the entire map.

[0076] In summary, the method for subdividing digital maps into segments enables targeted and efficient calibration and routine checks of map quality. Here, the calibrated segments serve as the basis for determining the number of correction levels to be transferred to the remaining routine segments. This optimizes the inspection workload without compromising the map's accuracy and reliability.

[0077] Figure 3 Figure 40 shows the number of map errors found during the calibration phase of randomly selected map features. This figure illustrates how the number of errors decreases as the number of correction levels increases during the calibration phase, and it is used to optimize the calibration process for a specific map feature.

[0078] The number of correction levels is plotted on the X-axis (41) of the graph. Each correction level represents a complete inspection process in which errors are systematically checked for on map features. As the number of correction levels increases, the expected number of remaining errors decreases until no further errors are found.

[0079] Y-axis 42 shows the number of map errors found. This includes all errors found at each calibration level. Typically, a larger number of errors are found early in the calibration phase because the map has not yet undergone sufficient inspection. With each further calibration level, the number of newly found errors decreases.

[0080] During the calibration phase, randomly selected map features are examined to determine how many correction levels are necessary to achieve the desired error rate. The graph shows that a relatively large number of errors are found in the first few correction levels. However, as the number of checks increases, the number of errors found steadily decreases.

[0081] The graph also illustrates that no further errors were found in the sixth correction level 43 and the seventh correction level 44, which correspond to the last two successive correction levels. This point marks the end of the calibration phase, as the necessary number of correction levels has been reached. From this point onward, the calibration results can be applied to other map segments, allowing for a reduced number of checks in these segments.

[0082] Therefore, under the current circumstances, Figure 3Note: Approximately 140 errors were found in the first correction level 45, and approximately 50 errors were found in the second correction level 46. This indicates that a relatively large number of errors exist in the early stages of the calibration phase because the map features have not yet been sufficiently validated. The first check serves as a starting point for determining the overall extent of errors in the map features.

[0083] Following further correction stages, the number of errors continued to decrease. In the fifth correction stage 47, only two errors were found. This indicates that the check was successful, as the number of newly discovered errors decreased significantly compared to previous stages.

[0084] The first line 48 in this graph represents the sum of all errors found so far. This sum shows the cumulative number of errors found across all correction levels. The sum grows rapidly in the first few correction levels because many errors are still being found initially. However, starting from level five, the line flattens out because only a small number of errors have been found.

[0085] The second line, 49, represents the remaining errors on the map that persist after each correction level. In the earlier levels, the number of remaining errors is high because many errors are discovered progressively with each correction level. With each further correction level, the number of remaining errors decreases, which can be seen from the descending slope of the second line. The third line, 50, represents the errors found at each correction level.

[0086] Figure 4 A schematic diagram of one embodiment of an apparatus is shown for performing the method for correcting a digital map. The apparatus consists of three functional units, shown as rectangles and connected to each other by arrows to illustrate the data flow between these units.

[0087] The calibration unit 51 is responsible for performing the calibration phase. During the calibration phase, a first map segment of the digital map is selected, and a determined number of correction levels are applied to this first map segment. The correction levels analyze the map segment for errors, eliminate these errors, and determine the number of correction levels applied. The output of the calibration unit provides the results of the calibration phase, particularly the number of correction levels determined, and forwards these results to the analysis and evaluation unit.

[0088] The analysis and evaluation unit 52 processes the results from the calibration phase and calculates the number of correction levels necessary for the remaining map segments based on these results. Here, the analysis and evaluation unit 52 considers the deviations and error rates from the calibration phase and adjusts the number of correction levels accordingly. The calculated number of correction levels is forwarded to the correction unit as output data from the analysis and evaluation unit 52.

[0089] The correction unit 53 is responsible for checking and correcting the remaining map segments based on the number of correction levels calculated by the analysis and evaluation unit 52. Here, the correction unit 53 analyzes errors in the remaining map segments and eliminates these errors according to the specifications of the analysis and evaluation unit. The output of the correction unit is a fully corrected digital map.

Claims

1. A method for calibrating a digital map (30), the method comprising the following steps: Perform the calibration phase (3) as follows: Select the first map segment (33) of the digital map (30); In the first correction level (4), the first map segment (33) is checked for errors (9, 10, 11, 12, 13) and the errors found in the first map segment (33) are corrected (50). In further correction stages (5, 6, 7, 8), errors are checked (9, 10, 11, 12, 13) in the corrected first map segment (33) and errors (50) found in the corrected first map segment (33) are corrected. Repeat the further correction levels (7, 8, 43, 44) until fewer than a predetermined number of errors (12, 13) are found in at least two successive correction levels (7, 8, 43, 44), especially when no more errors (50) are found, and in this way determine the first number of correction levels (4, 5, 6, 7, 8) performed on the first map segment (33) from the calibration stage (3); Based on the first number of correction levels (4, 5, 6, 7, 8) performed for the first map segment (33) in the calibration phase (3), a second number of correction levels (17, 18, 19) necessary for the remaining map segments (34) is determined, wherein, The second number of necessary correction levels is at least one fewer correction level than the first number of correction levels performed in the calibration phase, preferably at least three fewer correction levels; The remaining map segments (34) of the digital map (30) are checked and corrected by the second number of correction levels (17, 18, 19) obtained.

2. The method according to claim 1, wherein, The calibration phase (3) is performed separately for each map type or for a specific map element of the digital map (30) in order to optimize the inspection load.

3. The method according to claim 1 or 2, wherein, The correction is performed by a person or human inspector, who compares the first map segment (33) with a comparison map segment in at least one feature and manually corrects the deviation between the first map segment (33) and the comparison map segment in the first map segment, especially by means of a graphical user interface.

4. The method according to claim 1 or 2, wherein, The correction level is implemented automatically by an artificial intelligence model, wherein the artificial intelligence model checks the first map segment (33) in at least one feature and corrects errors (50) in the first map segment (33).

5. The method according to claim 3 or 4, wherein, In each individual correction level, multiple human inspectors check the corresponding map segments (33) of the digital map (30) for errors (9, 10, 11, 12, 13) using a graphical user interface or an artificial intelligence model to improve the accuracy of the inspection, wherein the error rate is estimated by applying statistical methods, preferably by averaging, the results of different inspectors or artificial intelligence models.

6. The method according to claim 4 or 5, wherein, The artificial intelligence model is optimized by machine learning, wherein the results of each calibration level (4, 5, 6, 7, 8, 17, 18, 19) are used as training data for the artificial intelligence model to improve future calibration stages (3).

7. The method according to any one of the preceding claims, wherein, The errors (9, 10, 11, 12, 13) are stored in the database and used for further analysis.

8. The method according to any one of the preceding claims, wherein, Map segments for the calibration phase (3) are selected based on the complexity of the map segments (33). In particular, map segments in the city center are more complex than map segments in forest areas or parking lots and are therefore selected.

9. The method according to any one of the preceding claims, wherein, The results of the calibration phase (3) are displayed in a visualization of the calibrated digital map (30).

10. The method according to any one of the preceding claims, wherein, The digital map (30) has at least two different region types, wherein the calibration phase (3) is performed on the two regions of different types, and different numbers of necessary correction levels obtained during the calibration phase (3) are applied to the corresponding region types of the digital map (30).

11. The method according to any one of the preceding claims, wherein, After each correction level (4, 5, 6, 7, 8, 17, 18, 19), an estimate or calculation of the remaining error rate or error probability is obtained to determine whether the predetermined value of the error rate has been reached.

12. The method according to any one of the preceding claims, characterized in that, All map features and / or feature groups are evaluated and corrected in a single correction method, or the correction method, including the calibration phase (3), is repeated separately for each individual map feature and / or for each individual feature group.

13. An apparatus for performing the method according to any one of claims 1 to 12, the apparatus comprising: A calibration unit (51) is used to perform a calibration phase (3) with a first number of correction levels (4, 5, 6, 7, 8) in the selected first map segment (33). Analysis and evaluation unit (52) is used to calculate the number of correction levels (17, 18, 19) for the remaining map segments (34); The correction unit (53) is used to correct the remaining map fragments (34) based on the second quantity obtained from the correction levels (17, 18, 19).

Images