Data quality inspection method and system based on time-space semantic information regularization and medium

Abstract

The invention relates to a data quality inspection method and system based on time-space semantic information regularization and a medium, and the method comprises the steps: U1, carrying out the size constraint of different types of obstacles: firstly determining the approximate range of the size of the obstacles through employing a national standard query mode, then determining the drop point distribution condition of the obstacle data through a big data statistics method, and carrying out the size constraint of the obstacles; finally, the data information of the size constraint of the obstacle is determined through manual joint judgment; and U2, carrying out space-time jump constraint on the front and rear frames of the same obstacle: carrying out constraint on the front and rear frames of the same obstacle by adopting a time sequence trend mode to obtain data information of the space-time jump constraint of the front and rear frames of the obstacle. According to the method, the size of the obstacle target which does not accord with the threshold range can be quickly detected, the investment of quality inspection personnel is reduced, accurate prior information is provided for sensing, recognition, trajectory prediction and other modules, and the false detection and omission ratio is remarkably reduced.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of data quality inspection, in particular to a method and system for regularizing data quality inspection based on spatio-temporal semantic information and a medium. BACKGROUND

[0002] The quality inspection work in the labeling task refers to checking and evaluating the labeling results to ensure the accuracy, consistency and standardization of the labeled data. The quality inspection work mainly checks the labeled data to determine whether the labeling results are consistent with the actual situation and whether there are errors or deviations. The consistency check mainly checks whether the labeling results of different labelers for the same object are consistent and whether there are situations such as missing labeling, mislabeling and the like for the same object in the front and back frames. At present, the data quality inspection work mainly relies on manual sampling inspection of the labeled data to determine whether the sampling results are correctly labeled by checking the position and attributes of the labeled objects. Missing labeling, multiple labeling and mislabeling all belong to labeling errors. However, the manual quality inspection to determine the labeling condition is low in efficiency, and the quality inspection result depends on the accuracy of the data and the judgment of the quality inspector, so a large amount of manpower is required for the quality inspection, and the accuracy and time consumption of the quality inspection cannot be ensured.

[0003] In the prior art, a Chinese patent (application number: 202110509045.1, publication number: CN 113127471A) discloses a method, device and equipment for automatic data quality inspection and a storage medium, wherein the method comprises: triggering a quality inspection service, acquiring data features, acquiring a plurality of preset quality inspection rules according to the data features, and acquiring data requiring quality inspection through a preset data query method; executing each quality inspection rule, respectively checking the data, obtaining a quality inspection result and storing the quality inspection result; and displaying the quality inspection result. However, the rule configuration of this scheme depends on user input and lacks intelligent suggestions. The user can customize the quality inspection rules through the page, but it is not mentioned whether the system has a rule recommendation or intelligent prompt function. For non-technical personnel or users unfamiliar with data quality, it is easy to make mistakes when configuring complex rules (such as regular expressions and advanced logic), and the learning cost is high, which may lead to improper rule setting, affecting the accuracy of quality inspection. The system matches the quality inspection rules according to the “data features”, but the definition of “data features” is relatively vague (such as classification and label), and it is not mentioned how to standardize or manage these features. If the feature system is not designed reasonably, it may lead to rule mismatching or mismatching, affecting the coverage and accuracy of quality inspection. When multiple rules are matched at the same time, the patent does not mention the execution priority, dependency or conflict processing mechanism of the rules. Some advanced logic quality inspection may need to be executed in a specific order, otherwise it may lead to misjudgment or resource waste. SUMMARY

[0004] In view of the shortcomings of the prior art, the present invention provides a data quality inspection method, system and medium based on spatiotemporal semantic information rules. It can not only quickly detect the size of obstacles that do not meet the threshold range, reducing the input of quality inspection personnel, but also provide accurate prior information for modules such as perception, recognition and trajectory prediction, significantly reducing the false detection and false detection rates.

[0005] To achieve the above and other related objectives, the present invention provides the following technical solution: A data quality inspection method based on spatiotemporal semantic information rules, the method comprising: U1. Size constraints for different types of obstacles: First, determine the approximate range of obstacle size by using national standard query, then determine the distribution of obstacle data landing points by big data statistics, and finally determine the size constraint data information of obstacles by joint manual judgment. U2. Spatiotemporal jump constraints on the same obstacle in consecutive frames: By using time series trends, constraints are applied to the same obstacle in consecutive frames to obtain data information on the spatiotemporal jump constraints of the obstacle in consecutive frames; U3. Based on the data information of the obstacle's size constraints and the data information of the spatiotemporal jump constraints of the obstacle in the preceding and following frames, a joint constraint mechanism is constructed to label the abnormal jump data and perform quality inspection to obtain the data information of the data quality inspection result.

[0006] Furthermore, in step U1, the dimensional constraints on different types of obstacles include: U11. Based on the standard size range initial screening, common static and dynamic obstacles are classified and typical set parameter ranges are extracted. Through standard queries, the theoretical size boundaries of various obstacles in length, width and height dimensions are established as the initial constraints for subsequent model data input. U12. Size distribution modeling based on real driving data, manual annotation and quality inspection based on historical data of autonomous driving fleets on public roads, and statistical analysis of massive amounts of actual observation data to obtain the actual distribution characteristics of various obstacle sizes; U13. Joint review and solidification of size constraints: Based on the dual support of standards and big data statistics, manual joint judgment is carried out to refine the initially confirmed size range.

[0007] Furthermore, in step U12, obtaining the actual distribution characteristics of the sizes of various obstacles includes: U121. The mean and mode are used to reflect the central tendency of dimensions; U122. The standard deviation and confidence interval are used to reflect the dispersion of the distribution; U123. Extreme values ​​and long-tail distribution; U124. Size variation patterns in different regions or time periods.

[0008] Furthermore, the manual joint judgment includes: whether the abnormal data value is a real problem, whether it is a problem of incorrect detection or calibration parameters of the acquisition sensor, or whether certain size combinations are physically unreasonable, and whether a conservative boundary is set according to the risk level.

[0009] Furthermore, in step U2, the spatiotemporal jump constraint on the same obstacle in consecutive frames includes: U21. For each tracked obstacle instance, the state vector sequence over several historical frames: S t =[x t ,y t ,z t ,v t ,l t ,w t ,h t ,θ t ], Among them, (x t ,y t ,z t Let v be the coordinates of the center point of the obstacle at time t. t Let l be the estimated velocity of the obstacle at time t. t ,w t ,h t Let θ be the length, width, and height of the obstacle at time t. t Let t be the heading angle of the obstacle. U22. Extract time series trend features; U23. Set the spatiotemporal jump constraint threshold. Based on the trend prediction results, define reasonable spatiotemporal consistency constraints to determine whether the detection result of the current frame has an abnormal jump.

[0010] Furthermore, in step U22, the extraction of time series trend features includes: U221. A kinematic model is established using Kalman filtering to recursively estimate and predict the state of obstacles; U222. Moving average and smoothing: Using a sliding window to model the trend of historical state sequences, identify the evolution of motion and geometric properties, and perform exponential weighted averaging on volatile parameters such as size and heading to suppress abnormal jumps.

[0011] Furthermore, in step U23, setting the spatiotemporal jump constraint threshold includes: U231. Position jump constraint: If the Euclidean distance between the detected position and the predicted position in the current frame exceeds the threshold, it is determined to be a position jump, which may be a mismatch or a new target; U232. Velocity jump constraint: If the difference between the estimated velocity and the trend predicted velocity in the current frame exceeds the threshold, it is considered an unreasonable acceleration and needs to be smoothed or removed. U233. Size jump constraint: If the rate of change of the length, width and height of the obstacle exceeds 18-20% in a single frame, it is caused by occlusion or partial occlusion. The trend value should be retained or the historical average value should be used instead. U234. Dynamic Correction and Trajectory Maintenance: When the detection result exceeds the above-mentioned jump constraint, the system does not immediately create a new trajectory or terminate the old trajectory, but uses state correction and trajectory interpolation to correct it; U235. Adaptive Mechanism: To adapt to different scenarios, a scene awareness module is introduced to dynamically adjust the jump threshold.

[0012] Furthermore, the correction method using state correction and trajectory interpolation includes: State correction: Based primarily on trend forecasts, combined with current observations, a smooth and reasonable estimate of the current state is generated; Confidence reduction: Decrease the confidence of the detection result in this frame and prolong the trajectory survival counter; Trajectory interpolation: In missed frames, the trajectory is completed based on trend extrapolation to maintain target continuity.

[0013] To achieve the above and other related objectives, the present invention also provides a data quality inspection system based on spatiotemporal semantic information rules, including a computer device that is programmed or configured to perform the steps of any of the data quality inspection methods based on spatiotemporal semantic information rules described above.

[0014] To achieve the above and other related objectives, the present invention also provides a computer-readable storage medium storing a computer program programmed or configured to perform any of the spatiotemporal semantic information-based regularized data quality inspection methods described in the present invention.

[0015] The present invention has the following positive effects: 1. This invention constructs a scientific and reliable obstacle size constraint model by integrating standards, real-world driving data, and expert experience. It ensures compliance through initial screening based on national standards, utilizes large-scale real-vehicle data statistical analysis to obtain actual distribution characteristics, effectively reflecting the diversity and long-tail scenarios of real traffic environments, and combines multi-domain expert review to manually determine outliers and risk boundaries, improving the model's rationality and safety. The resulting structured size constraint knowledge base (JSON format) can quickly identify obstacle target sizes that do not meet threshold ranges, reducing the workload of quality inspectors. It provides accurate prior information for modules such as perception, recognition, and trajectory prediction, significantly reducing false positives and false negatives, enhancing the system's robustness and decision reliability in complex urban scenarios, and improving the overall safety and engineering feasibility of autonomous driving.

[0016] 2. By introducing time series trend analysis, this invention effectively establishes a reasonable expectation of obstacle state evolution, realizes automatic identification and suppression of abnormal jumps, and significantly improves the stability and robustness of multi-target tracking. Especially in complex urban scenarios, facing frequent occlusion, dense traffic, and sensor noise, this constraint mechanism can quickly detect the labeled data of abnormal jumps, thereby reducing the input of quality inspectors and increasing the efficiency of quality inspection. Attached Figure Description

[0017] Figure 1 This is a schematic diagram of the method flow of the present invention; Figure 2 This is a schematic diagram illustrating the process of applying size constraints to different types of obstacles according to the present invention; Figure 3 This is a schematic diagram of the process of constraining the spatiotemporal jump of the same obstacle in consecutive frames according to the present invention. Detailed Implementation

[0018] The exemplary embodiments of this disclosure are described below with reference to the accompanying drawings, including various details of the embodiments to aid understanding, and should be considered merely exemplary. Therefore, those skilled in the art will recognize that various changes and modifications can be made to the embodiments described herein without departing from the scope and spirit of this disclosure. Similarly, for clarity and brevity, descriptions of well-known functions and structures are omitted in the following description.

[0019] Example 1: As Figure 1 As shown, a data quality inspection method based on spatiotemporal semantic information rules is proposed, the method comprising: U1. Size constraints for different types of obstacles: First, determine the approximate range of obstacle size by using national standard query, then determine the distribution of obstacle data landing points by big data statistics, and finally determine the size constraint data information of obstacles by joint manual judgment. U2. Spatiotemporal jump constraints on the same obstacle in consecutive frames: By using time series trends, constraints are applied to the same obstacle in consecutive frames to obtain data information on the spatiotemporal jump constraints of the obstacle in consecutive frames; U3. Based on the data information of the obstacle's size constraints and the data information of the spatiotemporal jump constraints of the obstacle in the preceding and following frames, a joint constraint mechanism is constructed to label the abnormal jump data and perform quality inspection to obtain the data information of the data quality inspection result.

[0020] In this embodiment, as Figure 2 As shown, in step U1, the dimensional constraints on different types of obstacles include: U11. Based on the standard size range initial screening, common static and dynamic obstacles are classified and typical set parameter ranges are extracted. Through standard queries, the theoretical size boundaries of various obstacles in length, width and height dimensions are established as the initial constraints for subsequent model data input. U12. Size distribution modeling based on real driving data, manual annotation and quality inspection based on historical data of autonomous driving fleets on public roads, and statistical analysis of massive amounts of actual observation data to obtain the actual distribution characteristics of various obstacle sizes; U13. Joint review and solidification of size constraints: Based on the dual support of standards and big data statistics, manual joint judgment is carried out to refine the initially confirmed size range.

[0021] In this embodiment, in step U12, obtaining the actual distribution characteristics of the sizes of various obstacles includes: U121. The mean and mode are used to reflect the central tendency of dimensions; U122. The standard deviation and confidence interval are used to reflect the dispersion of the distribution; U123. Extreme values ​​and long-tail distribution; U124. Size variation patterns in different regions or time periods.

[0022] In this embodiment, the manual joint judgment includes: whether the abnormal data value is a real problem, whether it is a problem of incorrect detection or calibration parameters of the acquisition sensor, or whether certain size combinations are physically unreasonable, and whether a conservative boundary is set according to the risk level.

[0023] In this embodiment, as Figure 3 As shown, in step U2, the spatiotemporal jump constraint on the same obstacle in consecutive frames includes: U21. For each tracked obstacle instance, the state vector sequence over several historical frames: S t =[xt ,y t ,z t ,v t ,l t ,w t ,h t ,θ t ], Among them, (x t ,y t ,z t Let v be the coordinates of the center point of the obstacle at time t. t Let l be the estimated velocity of the obstacle at time t. t ,w t ,h t Let θ be the length, width, and height of the obstacle at time t. t Let t be the heading angle of the obstacle. U22. Extract time series trend features; U23. Set the spatiotemporal jump constraint threshold. Based on the trend prediction results, define reasonable spatiotemporal consistency constraints to determine whether the detection result of the current frame has an abnormal jump.

[0024] In this embodiment, step U22, the extraction of time series trend features includes: U221. A kinematic model is established using Kalman filtering to recursively estimate and predict the state of obstacles; U222. Moving average and smoothing: Using a sliding window to model the trend of historical state sequences, identify the evolution of motion and geometric properties, and perform exponential weighted averaging on volatile parameters such as size and heading to suppress abnormal jumps.

[0025] In this embodiment, in step U23, setting the spatiotemporal jump constraint threshold includes: U231. Position jump constraint: If the Euclidean distance between the detected position and the predicted position in the current frame exceeds the threshold, it is determined to be a position jump, which may be a mismatch or a new target; U232. Velocity jump constraint: If the difference between the estimated velocity and the trend predicted velocity in the current frame exceeds the threshold, it is considered an unreasonable acceleration and needs to be smoothed or removed. U233. Size jump constraint: If the rate of change of the length, width and height of the obstacle exceeds 18-20% in a single frame, it is caused by occlusion or partial occlusion. The trend value should be retained or the historical average value should be used instead. U234. Dynamic Correction and Trajectory Maintenance: When the detection result exceeds the above-mentioned jump constraint, the system does not immediately create a new trajectory or terminate the old trajectory, but uses state correction and trajectory interpolation to correct it; U235. Adaptive Mechanism: To adapt to different scenarios, a scene awareness module is introduced to dynamically adjust the jump threshold.

[0026] In this embodiment, the correction method using state correction and trajectory interpolation includes: State correction: Based primarily on trend forecasts, combined with current observations, a smooth and reasonable estimate of the current state is generated; Confidence reduction: Decrease the confidence of the detection result in this frame and prolong the trajectory survival counter; Trajectory interpolation: In missed frames, the trajectory is completed based on trend extrapolation to maintain target continuity.

[0027] Example 2: Based on the spatiotemporal semantic information-based data quality inspection method in Example 1, the present invention will be further explained and described below.

[0028] like Figure 1 As shown, the first step is to impose size constraints on different types of obstacles. This involves first determining the approximate size range of obstacles using national standards, then using big data statistical methods to determine the distribution of a large amount of obstacle data, and finally using a collaborative manual assessment to determine the size constraints of the obstacles.

[0029] Step 1: Initial screening based on size ranges according to national standards First, based on Chinese national standards (GB) and relevant industry specifications (such as GB5768 "Road Traffic Signs and Markings", CJJ37 "Urban Road Design Code", and GB 7258 "Technical Conditions for Safe Operation of Motor Vehicles"), common static and dynamic obstacles are classified, and their typical set parameter ranges are extracted. Through standard lookup, theoretical dimensional boundaries of various obstacles in length, width, and height are established as initial constraints for subsequent model data input, ensuring that the model conforms to regulations and engineering realities.

[0030] Step 2: Size distribution modeling based on real driving data Data collected over a long period by the internal autonomous driving fleet in various driving scenarios, including urban public roads and highways, was manually labeled and quality-checked. Through statistical analysis of massive amounts of actual observation data, the actual distribution characteristics of various obstacle sizes were obtained, including: 1) Using the mean and mode to reflect the central tendency of size. 2) The standard deviation and confidence interval are used to reflect the dispersion of the distribution. 3) Extreme values ​​and long-tail distribution (e.g., oversized trucks) 4) The dimensional variation patterns in different areas or time periods (e.g., an increase in temporary obstacles in the construction area). Step 3, multi-expert joint review and size constraint solidification Based on the dual support of national standards and big data statistics, expert joint judgment is conducted to refine the initially confirmed size range. The focus is on whether outliers are genuine issues, whether they are due to sensor malfunctions or calibration problems, whether certain size combinations are physically unreasonable, and whether conservative boundaries have been set according to risk levels (e.g., setting a minimum height threshold for low obstacles to prevent missed detections).

[0031] Ultimately, considering safety, practicality, and algorithm compatibility, recommended size ranges and safe buffer zones for each type of obstacle are determined and integrated into the obstacle knowledge base of the autonomous driving system in JSON format, serving modules such as target recognition, instance segmentation, occupancy grid generation, and trajectory planning.

[0032] 2. Apply spatiotemporal jump constraints to the same obstacle in consecutive frames. Constraints on the same obstacle in consecutive frames are applied using a time series trend method.

[0033] During obstacle detection, factors such as sensor noise, occlusion, false detections, missed detections, or sudden disappearance of the target can cause abnormal spatiotemporal jumps in the detection results of the same obstacle between consecutive frames. These jumps may include abrupt changes in position, drastic fluctuations in velocity, or abnormal scaling in size. Such jumps not only affect tracking stability but may also lead to trajectory breaks or incorrect associations, thereby impacting the safety of decision-making and planning. To address this, a spatiotemporal jump constraint mechanism based on time series trend modeling is proposed to improve the continuity and rationality of obstacle state estimation.

[0034] Step 1: Construct the time series of obstacle states For each tracked obstacle instance, its state vector sequence over several historical frames: S t =[x t ,y t ,z t ,v t ,l t ,w t ,h t ,θ t ], Among them, (x t ,y t ,z t Let v be the coordinates of the center point of the obstacle at time t. t Let l be the estimated velocity of the obstacle at time t. t ,w t ,h t Let θ be the length, width, and height of the obstacle at time t. t Let t be the heading angle of the obstacle. Step 2: Extract time series trend features 1. By using Kalman filtering to establish a kinematic model, the state of obstacles can be recursively estimated and predicted, which naturally has the ability to suppress noise and extrapolate trends.

[0035] 2. Moving Average and Smoothing: A sliding window is used to model the trend of historical state sequences and identify the evolution patterns of their motion and geometric properties. An exponentially weighted average (EMA) is applied to volatile parameters such as size and heading to suppress anomalous jumps.

[0036] Step 3: Set the spatiotemporal transition constraint threshold Based on the trend prediction results, reasonable spatiotemporal consistency constraints are defined to determine whether the detection result of the current frame has an abnormal jump.

[0037] 1) Position jump constraint: If the Euclidean distance between the current frame detection position and the predicted position exceeds the threshold (the theoretical displacement limit calculated based on the maximum acceleration and frame interval), it is determined to be a position jump, which may be a mismatch or a new target.

[0038] 2) Velocity jump constraint: If the difference between the estimated velocity of the current frame and the trend prediction velocity exceeds the threshold, it is considered an unreasonable acceleration and needs to be smoothed or removed.

[0039] 3) Size jump constraint: If the rate of change of the length, width and height of the obstacle exceeds 20% in a single frame, it is considered that it may be caused by detection failure or partial occlusion. The trend value should be retained or the historical average value should be used instead.

[0040] 4) Dynamic correction and trajectory maintenance When the detection result exceeds the above transition constraints, the system does not immediately create a new trajectory or terminate the old trajectory, but instead adopts the following strategy: 1. State Correction: Based on trend predictions, current observations are fused (weighted fusion) to generate a smoother and more reasonable estimate of the current state; 2. Confidence reduction: Reduce the confidence of the detection result of this frame, prolong the trajectory survival counter, and avoid false deletion due to temporary occlusion; 3. Trajectory interpolation: In the missed frames, the trajectory is completed based on trend extrapolation to maintain the continuity of the target.

[0041] 5) Adaptive mechanism To adapt to different scenarios, such as highways and urban intersections, a scene perception module is introduced to dynamically adjust the jump threshold. For example, a larger range of positional offset and acceleration is allowed in high-speed scenarios, while stricter size and heading constraints are adopted in low-speed congestion scenarios to prevent frequent jitter.

[0042] In this embodiment, the present invention provides a data quality inspection system based on spatiotemporal semantic information rules, including a computer device that is programmed or configured to perform the steps of any of the data quality inspection methods based on spatiotemporal semantic information rules described above.

[0043] In this embodiment, the present invention provides a computer-readable storage medium storing a computer program programmed or configured to perform any of the spatiotemporal semantic information-based regularized data quality inspection methods described above.

[0044] Any references to memory, storage, database, or other media used in the embodiments provided in this application may include non-volatile and / or volatile memory. Non-volatile memory may include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM), or flash memory. Volatile memory may include random access memory (RAM) or external cache memory. By way of illustration and not limitation, RAM is available in a variety of forms, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), dual data rate SDRAM (DDRSDRAM), enhanced SDRAM (ESDRAM), synchronous link DRAM (SLDRAM), RAMbus direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and RAMbus dynamic RAM (RDRAM), etc.

[0045] In summary, this invention can not only quickly identify the size of obstacles that do not meet the threshold range, reducing the workload of quality inspectors, but also provide accurate prior information for modules such as perception, recognition, and trajectory prediction, significantly reducing the false detection and false negative rates.

[0046] The specific embodiments described above do not constitute a limitation on the scope of protection of this disclosure. Those skilled in the art should understand that various modifications, combinations, sub-combinations, and substitutions can be made according to design requirements and other factors. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of this disclosure should be included within the scope of protection of this disclosure.

Claims

1. A data quality inspection method based on spatiotemporal semantic information rules, characterized in that, The method includes: U1. Size constraints for different types of obstacles: First, determine the approximate range of obstacle size by using national standard query, then determine the distribution of obstacle data landing points by big data statistics, and finally determine the size constraint data information of obstacles by joint manual judgment. U2. Spatiotemporal jump constraints on the same obstacle in consecutive frames: By using time series trends, constraints are applied to the same obstacle in consecutive frames to obtain data information on the spatiotemporal jump constraints of the obstacle in consecutive frames; U3. Based on the data information of the obstacle's size constraints and the data information of the spatiotemporal jump constraints of the obstacle in the preceding and following frames, a joint constraint mechanism is constructed to label the abnormal jump data and perform quality inspection to obtain the data information of the data quality inspection result.

2. The data quality inspection method based on spatiotemporal semantic information rules according to claim 1, characterized in that, In step U1, the dimensional constraints on different types of obstacles include: U11. Based on the standard size range initial screening, common static and dynamic obstacles are classified and typical set parameter ranges are extracted. Through standard queries, the theoretical size boundaries of various obstacles in length, width and height dimensions are established as the initial constraints for subsequent model data input. U12. Size distribution modeling based on real driving data, manual annotation and quality inspection based on historical data of autonomous driving fleets on public roads, and statistical analysis of massive amounts of actual observation data to obtain the actual distribution characteristics of various obstacle sizes; U13. Joint review and solidification of size constraints: Based on the dual support of standards and big data statistics, manual joint judgment is carried out to refine the initially confirmed size range.

3. The data quality inspection method based on spatiotemporal semantic information rules according to claim 2, characterized in that, In step U12, obtaining the actual distribution characteristics of the sizes of various obstacles includes: U121. The mean and mode are used to reflect the central tendency of dimensions; U122. The standard deviation and confidence interval are used to reflect the dispersion of the distribution; U123. Extreme values ​​and long-tail distribution; U124. Size variation patterns in different regions or time periods.

4. The data quality inspection method based on spatiotemporal semantic information rules according to claim 2, characterized in that, The manual joint judgment includes: whether the abnormal data value is a real problem, whether it is a problem of incorrect detection or calibration parameters of the acquisition sensor, or whether certain size combinations are physically unreasonable, and whether conservative boundaries are set according to the risk level.

5. The data quality inspection method based on spatiotemporal semantic information rules according to claim 1, characterized in that, In step U2, the spatiotemporal jump constraint on the same obstacle in consecutive frames includes: U21. For each tracked obstacle instance, the state vector sequence over several historical frames: S t =[x t ,y t ,z t ,v t ,l t ,w t ,h t ,θ t ], Among them, (x t ,y t ,z t Let v be the coordinates of the center point of the obstacle at time t. t Let l be the estimated velocity of the obstacle at time t. t ,w t ,h t Let θ be the length, width, and height of the obstacle at time t. t Let t be the heading angle of the obstacle. U22. Extract time series trend features; U23. Set the spatiotemporal jump constraint threshold. Based on the trend prediction results, define reasonable spatiotemporal consistency constraints to determine whether the detection result of the current frame has an abnormal jump.

6. The data quality inspection method based on spatiotemporal semantic information rules according to claim 5, characterized in that, In step U22, the extraction of time series trend features includes: U221. A kinematic model is established using Kalman filtering to recursively estimate and predict the state of obstacles; U222. Moving average and smoothing: Using a sliding window to model the trend of historical state sequences, identify the evolution of motion and geometric properties, and perform exponential weighted averaging on volatile parameters such as size and heading to suppress abnormal jumps.

7. The data quality inspection method based on spatiotemporal semantic information rules according to claim 5, characterized in that, In step U23, setting the spatiotemporal jump constraint threshold includes: U231. Position jump constraint: If the Euclidean distance between the detected position and the predicted position in the current frame exceeds the threshold, it is determined to be a position jump, which may be a mismatch or a new target; U232. Velocity jump constraint: If the difference between the estimated velocity and the trend predicted velocity in the current frame exceeds the threshold, it is considered an unreasonable acceleration and needs to be smoothed or removed. U233. Size jump constraint: If the rate of change of the length, width and height of the obstacle exceeds 18-20% in a single frame, it is caused by occlusion or partial occlusion. The trend value should be retained or the historical average value should be used instead. U234. Dynamic Correction and Trajectory Maintenance: When the detection result exceeds the above-mentioned jump constraint, the system does not immediately create a new trajectory or terminate the old trajectory, but uses state correction and trajectory interpolation to correct it; U235. Adaptive Mechanism: To adapt to different scenarios, a scene awareness module is introduced to dynamically adjust the jump threshold.

8. The data quality inspection method based on spatiotemporal semantic information rules according to claim 7, characterized in that, The correction method using state correction and trajectory interpolation includes: State correction: Based primarily on trend forecasts, combined with current observations, a smooth and reasonable estimate of the current state is generated; Confidence reduction: Decrease the confidence of the detection result in this frame and prolong the trajectory survival counter; Trajectory interpolation: In missed frames, the trajectory is completed based on trend extrapolation to maintain target continuity.

9. A data quality inspection system based on spatiotemporal semantic information rules, comprising computer equipment, characterized in that, The computer device is programmed or configured to perform the steps of the spatiotemporal semantic information-based data quality inspection method as described in any one of claims 1 to 8.

10. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a computer program that is programmed or configured to execute the spatiotemporal semantic information-based data quality inspection method according to any one of claims 1 to 8.

Images