Perception confidence evaluation method of simulation lane and related device

Abstract

The invention discloses a perception confidence evaluation method for a simulation lane and a related device. The perception confidence evaluation method for the simulation lane comprises the steps that simulation environment perception data of the simulation lane in a simulation environment are acquired, and the simulation lane is obtained based on digital twinborn reconstruction of a real lane; calculating a simulation perception error between the simulation environment perception data and the perception truth value of the real lane, and a real perception error between real environment perception data corresponding to the real lane and the perception truth value; and evaluating the perception confidence of the simulation lane based on the simulation perception error and the real perception error. Through the method, accurate, efficient and reliable evaluation of the perception confidence of the simulation lane can be realized, whether the perception of the lane in the corresponding simulation environment meets the perception requirement in the real environment can be comprehensively judged, and the development and test of the simulation environment of the automatic driving system can be subsequently and specifically optimized.

Description

Technical Field

[0001] The present application relates to the field of autonomous driving simulation technology, and in particular to a perception confidence assessment method for a simulated lane and related devices. Background Art

[0002] In the simulation test of the autonomous driving system, a simulation environment with a higher confidence level can reflect the conditions of the real environment more realistically and accurately. Testing the autonomous driving system in a high-confidence simulation environment can ensure the safety, reliability and effectiveness of the simulation test and reduce the development cost and time of the simulation test.

[0003] In a simulation environment, the perception of simulated lanes can help autonomous vehicles accurately locate on the road, stay in the lane, and comply with relevant regulations when changing lanes and turning. Therefore, accurate assessment of the perception confidence of simulated lanes is crucial to improving the effectiveness and reliability of autonomous driving. However, there is currently no universal and mature method in the industry to evaluate the perception confidence of simulated lanes in a simulation environment, making it difficult for the simulation test of the lane perception function of the autonomous driving system to meet the actual needs of the real environment. Summary of the invention

[0004] The present application mainly provides a method and related device for evaluating the perception confidence of a simulated lane, aiming to solve the technical problem that the perception confidence of a simulated lane is difficult to evaluate.

[0005] In order to solve the above technical problems, the technical solution adopted in this application is: to provide a perception confidence evaluation method for a simulated lane. The perception confidence evaluation method for a simulated lane includes: obtaining simulated environment perception data of a simulated lane in a simulated environment, wherein the simulated lane is obtained based on the digital twin reconstruction of a real lane; calculating the simulated perception error between the simulated environment perception data and the perception true value of the real lane, and the real perception error between the real environment perception data corresponding to the real lane and the perception true value; and evaluating the perception confidence of the simulated lane based on the simulated perception error and the real perception error.

[0006] In some embodiments, the simulation environment perception data includes the lane type and lane curve of the simulation lane, and the real environment perception data includes the lane type and lane curve of the real lane; after acquiring the simulation environment perception data of the simulation lane in the simulation environment, it also includes: confirming whether the lane type of the simulation lane and the lane type of the real lane are the same; calculating the degree of fit between the lane curve of the simulation lane and the lane curve of the real lane; establishing a mapping relationship between the simulation lane and the real lane with the same lane type and the degree of fit within a preset fit range; wherein the simulation perception error and the real perception error are calculated based on the simulation lane and the real lane having a mapping relationship. In some embodiments,

[0007] In some embodiments, the calculating the degree of fit between the lane curve of the simulated lane and the lane curve of the real lane includes: calculating the frame fit degree between the lane curve of the simulated lane in each frame and the lane curve of the real lane in the corresponding frame based on a preset fit degree algorithm; performing weighted averaging of the frame fit degrees according to the appearance time of the simulated lane in the simulation environment perception data of each frame, and using the result of the weighted average as the degree of fit between the lane curve of the simulated lane and the lane curve of the real lane.

[0008] In some embodiments, the preset fitting degree algorithm includes mean square error, root mean square error, mean absolute error, determination coefficient and Hausdorff distance; the frame fitting degree is calculated between the lane curves of different lane types using the preset fitting degree algorithm corresponding to one of them.

[0009] In some embodiments, the calculation of the simulated perception error between the simulated environment perception data and the perception true value of the real lane, and the real perception error between the real environment perception data corresponding to the real lane and the perception true value, includes: in the simulated lane and the real lane of the same lane type, calculating the simulated perception error between the simulated environment perception data and the perception true value, and calculating the real perception error between the real environment perception data corresponding to the mapping relationship and the perception true value.

[0010] In some embodiments, the obtaining of simulated environment perception data of a simulated lane in a simulated environment includes: obtaining the real environment perception data generated by a physical vehicle sensing the real lane in a real environment, and recording vehicle posture data corresponding to each frame of the real environment perception data; based on the vehicle posture data, reproducing the trajectory of a virtual vehicle in a simulated environment, and using the same perception algorithm and perception points as the physical vehicle to control the virtual vehicle to sense the simulated lane in a simulated environment, so as to obtain the simulated environment perception data; wherein the physical vehicle is a vehicle equipped with a pre-installed autonomous driving system, and the virtual vehicle is a vehicle body mapping of the autonomous driving system in a simulated environment.

[0011] In some embodiments, the evaluating the perception confidence of the simulated lane based on the simulated perception error and the real perception error includes: when the simulated perception error is greater than a first preset threshold of the real perception error, determining that the perception confidence of the simulated lane is lacking in truth; when the simulated perception error is less than a second preset threshold of the real perception error, determining that the perception confidence of the simulated lane is lacking in noise.

[0012] To solve the above technical problems, another technical solution adopted in the present application is: to provide a perception confidence assessment device for a simulated lane, the perception confidence assessment device for a simulated lane comprising: an acquisition module, used to acquire simulated environment perception data of a simulated lane in a simulated environment, wherein the simulated lane is obtained based on a digital twin reconstruction of a real lane; a calculation module, used to calculate a simulated perception error between the simulated environment perception data and a perception true value of the real lane, and a real perception error between the real environment perception data corresponding to the real lane and the perception true value; an evaluation module, used to evaluate the perception confidence of the simulated lane based on the simulated perception error and the real perception error.

[0013] In order to solve the above-mentioned technical problems, another technical solution adopted in the present application is: providing a storage medium having program data stored thereon, characterized in that when the program data is executed by a processor, the steps of the perception confidence assessment method of the simulated lane as mentioned above are implemented.

[0014] In order to solve the above technical problems, another technical solution adopted in the present application is: to provide a computer device, which includes a processor and a memory connected to each other, and the memory stores a computer program. When the processor executes the computer program, the steps of the perception confidence assessment method of the simulated lane as mentioned above are implemented.

[0015] The beneficial effect of the present application is that, different from the prior art, the present application discloses a perception confidence evaluation method for a simulated lane and a related device. The present application calculates the simulated perception error and the real perception error between the corresponding perception true value and the simulated environment perception data obtained by perceiving the simulated lane and the real environment perception data obtained by perceiving the real lane, respectively, so as to achieve an accurate, efficient and reliable evaluation of the perception confidence of the simulated lane, and can comprehensively judge whether the perception of the lane in the corresponding simulated environment meets the needs of perception in the real environment, which is conducive to the subsequent targeted optimization of the development and testing of the simulation environment of the autonomous driving system. BRIEF DESCRIPTION OF THE DRAWINGS

[0016] In order to more clearly illustrate the embodiments of the present application or the technical solutions in the prior art, the drawings required for use in the embodiments or the prior art descriptions are briefly introduced below. Obviously, the drawings described below are only some embodiments of the present application. For those of ordinary skill in the art, other drawings can be obtained based on these drawings without creative work, among which:

[0017] Figure 1 It is a flow chart of an embodiment of a method for evaluating the perception confidence of a simulated lane provided by the present application;

[0018] Figure 2 yes Figure 1 A schematic flow chart of an embodiment of step 10 in the embodiment;

[0019] Figure 3 yes Figure 1 A schematic diagram of a flow chart of an embodiment after step 10 in the embodiment;

[0020] Figure 4 yes Figure 3 A schematic flow chart of an embodiment of step 14 in the embodiment;

[0021] Figure 5 yes Figure 1 A schematic flow chart of an embodiment of step 30 in the embodiment;

[0022] Figure 6 It is a structural schematic diagram of an embodiment of a perception confidence assessment method for a simulated lane provided by the present application;

[0023] Figure 7 It is a structural schematic diagram of an embodiment of a storage medium provided by the present application;

[0024] Figure 8 It is a structural diagram of an embodiment of a computer device provided by the present application. DETAILED DESCRIPTION

[0025] The following will be combined with the drawings in the embodiments of the present application to clearly and completely describe the technical solutions in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, not all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of this application.

[0026] The terms "first", "second", "third" in the embodiments of the present application are only used for descriptive purposes, and cannot be understood as indicating or implying relative importance or implicitly indicating the number of indicated technical features. Thus, the feature defined as "first", "second", "third" can expressly or implicitly include at least one of the features. In the description of the present application, the meaning of "multiple" is at least two, such as two, three, etc., unless otherwise clearly and specifically limited. In addition, the terms "including" and "having" and any of their variations are intended to cover non-exclusive inclusions. For example, a process, method, system, product or device comprising a series of steps or units is not limited to the listed steps or units, but optionally also includes steps or units that are not listed, or optionally also includes other steps or units inherent to these processes, methods, products or devices.

[0027] Reference to "embodiments" herein means that a particular feature, structure, or characteristic described in conjunction with the embodiments may be included in at least one embodiment of the present application. The appearance of the phrase in various locations in the specification does not necessarily refer to the same embodiment, nor is it an independent or alternative embodiment that is mutually exclusive with other embodiments. It is explicitly and implicitly understood by those skilled in the art that the embodiments described herein may be combined with other embodiments.

[0028] This application provides a method for evaluating the perception confidence of a simulated lane. Figure 1 , Figure 1 1 is a flow chart of an embodiment of a method for evaluating the perception confidence of a simulated lane provided by the present application. The method for evaluating the perception confidence of a simulated lane comprises:

[0029] Step 10: Acquire simulation environment perception data of the simulation lane in the simulation environment.

[0030] Among them, the simulated lane is obtained based on the digital twin reconstruction of the real lane.

[0031] Digital twin reconstruction refers to the process of digitally modeling and reconstructing physical entities using digital twin technology to create virtual copies of them. This process involves collecting various types of data on the corresponding physical objects, and converting these data into a real-time updated virtual model that is synchronized with the physical entity through modeling and simulation technology. Therefore, by reconstructing the digital twin of the real lane, the real lane can be mapped to the simulated lane in the corresponding simulation environment.

[0032] Among them, the real lane is the lane existing in the actual road, which is the object that the autonomous driving system needs to perceive and understand in the real environment. In the process of obtaining the real lane based on the digital twin reconstruction of the real lane, data can be collected from the road network within a certain range. The range can be determined according to the simulation needs and the computing power of the autonomous driving system, which can be 100km, 50km or 20km, etc. In the data collection process, the real environment is perceived through the perception algorithm, so that the structural data such as the road center, line type and lane line correlation corresponding to the real lane can be obtained. Based on these structural data, the road model can be programmatically established through geometric algorithms such as Ear-cut algorithm, Greiner-Hormann algorithm or Vatti algorithm, and the obstacle part related to the road and helpful for road perception can be established in the road model, so that the simulated lane corresponding to the real lane in the simulation environment can be obtained. Subsequently, the similarity of the simulation environment reconstructed by the digital twin can be evaluated through manual review and three-dimensional reconstruction, such as evaluating whether the width of the roads in the simulation environment is similar, whether the road line types and positions match, etc. by sampling, which is conducive to the subsequent optimization of the generation method of the simulation environment. This subsequent process is not within the scope of protection of this application and will not be elaborated here.

[0033] After reconstructing the digital twin of the real lane to obtain the simulated lane, the virtual vehicle corresponding to the autonomous driving system in the simulated environment can collect the corresponding simulated environment perception data, which can provide a basis for calculating the simulation perception error in the subsequent steps. At the same time, since the simulated lane is obtained based on the digital twin reconstruction of the real lane, these simulation data can usually better reflect the characteristics of the real lane, which helps to evaluate the perception confidence of the simulated lane accordingly.

[0034] Optionally, see Figure 2 , to obtain the simulation environment perception data of the simulation lane in the simulation environment, you can follow the following steps:

[0035] Step 11: Acquire real environment perception data generated by the physical vehicle sensing the real lane in the real environment, and record the vehicle posture data corresponding to each frame of the real environment perception data.

[0036] Step 12: Based on the vehicle posture data, the trajectory of the virtual vehicle in the simulation environment is reproduced, and the same perception algorithm and perception points as the physical vehicle are used to control the virtual vehicle to perceive the simulated lane in the simulation environment to obtain the simulation environment perception data.

[0037] Among them, the physical vehicle is a vehicle equipped with a pre-installed autonomous driving system, and the virtual vehicle is the vehicle body mapping of the autonomous driving system in a simulated environment.

[0038] One or more physical vehicles equipped with pre-configured autonomous driving systems deployed in the real environment can be used to truly perceive and record data in the real environment, including key information such as the center of the road, line type, and lane relevance of the real lane. These can be collected in advance, so when conducting confidence assessments, these real environment perception data can be obtained through relevant databases or interfaces in non-immediate task scenarios. At the same time, in order to accurately restore the driving status of the vehicle in the real environment in the simulation environment later, it is also necessary to record the vehicle posture data corresponding to each frame of real environment perception data, which reflects information such as the vehicle's position, orientation, and posture.

[0039] According to the vehicle posture data, the corresponding virtual vehicle can be controlled in the simulation environment to reproduce the trajectory according to the corresponding posture of the physical vehicle. Afterwards, the same perception algorithm and perception points as the sensor or perception system in the physical vehicle are used to control the virtual vehicle to perceive the simulated lane in the simulation environment, so as to obtain the required simulation environment perception data, wherein the perception algorithm is the same as the algorithm of the sensor or perception system of the physical vehicle, which can be a specific recognition algorithm, scanning algorithm, tracking algorithm and ranging algorithm, etc., and the perception point is the position, orientation, reference system and internal and external parameters of the corresponding sensor or perception system. Keeping the virtual vehicle perception algorithm and perception point consistent with the perception algorithm and perception point of the physical vehicle can ensure that the perception data in the simulation environment is as close to the real environment as possible, which is conducive to simulating the perception process of the virtual vehicle in the real environment, so as to evaluate the perception confidence of the automatic driving system to the simulated lane in the simulation environment later.

[0040] This trajectory reproduction is an open-loop operation mode. The behavior of the virtual vehicle is simulated and controlled based on the behavior of the physical vehicle, rather than being adjusted according to the feedback of the simulation environment itself, to ensure the high controllability of the virtual vehicle's behavior. All movement and perception behaviors of the virtual vehicle will be consistent with those of the physical vehicle. Only the perceived results will be regenerated, which can effectively ensure that the driving trajectory of the virtual vehicle in the simulation environment is consistent with the formal trajectory of the physical vehicle in the real environment, and ensure that the obtained simulation environment perception data and the real environment perception data maintain consistency of other objective factors on the basis of different environments, thereby improving the effectiveness and reliability of subsequent evaluations of the simulation environment.

[0041] Optionally, the simulated environment perception data includes a lane type and a lane curve of a simulated lane, and the real environment perception data includes a lane type and a lane curve of a real lane.

[0042] Lane types are divided according to their purpose and attributes, such as straight lanes, turning lanes, and overtaking lanes, etc. They can also be further specified, for example, lane lines can be distinguished into dotted lines that can be changed and solid lines that cannot be changed, and the left lane can be distinguished into lanes that can only turn left, lanes that can turn left and turn around, and lanes that can turn left or go forward, etc. In the simulation environment, the virtual vehicle can identify the type of simulated lane and adjust the corresponding driving strategy according to the lane type to ensure that the vehicle drives safely in accordance with traffic rules.

[0043] The lane curve describes the shape and curvature of the lane. On real roads, the lane curve often changes in perception as the line of sight changes and speed is adjusted. The autonomous driving system needs to be able to perceive the changes in the lane curve in advance and make corresponding driving decisions. Correspondingly, in the simulation environment, by simulating lane curves of different shapes and curvatures, the autonomous driving system's ability to cope with complex road environments can be evaluated. The determined lane types and lane curves of the simulated lanes and real lanes can be used for subsequent judgments on the degree of fit between the simulated lanes and real lanes, as well as type-based classification processing.

[0044] Optionally, see Figure 3 After obtaining the simulated environment perception data of the simulated lane in the simulated environment, the perception confidence evaluation method of the simulated lane also includes:

[0045] Step 13: Confirm whether the lane type of the simulated lane is the same as the lane type of the real lane.

[0046] Step 14: Calculate the degree of fit between the lane curve of the simulated lane and the lane curve of the real lane.

[0047] Step 15: Establish a mapping relationship between the simulated lanes and the real lanes with the same lane type and a fitting degree within a preset fitting degree range.

[0048] Among them, the simulated perception error and the real perception error are calculated based on the simulated lane and the real lane with a mapping relationship.

[0049] In order to ensure the consistency of the basic properties of the simulated lane and the real lane, it is possible to preliminarily determine whether the lane types of the simulated lane and the real lane are the same. If there is a deviation in the lane type between the simulated lane and the real lane, the assessment of the perception confidence of the simulated lane will lose practical significance. Therefore, by comparing the lane types of the simulated lane and the real lane, it is possible to preliminarily determine whether the simulation environment accurately reflects the basic characteristics of the real environment.

[0050] Among them, since the lane type is usually a discrete value or a specific identifier in the corresponding data of the simulated lane and the real lane, there are usually only two results when comparing the road types: the same and the different. After comparing the simulated lanes and the real lanes one by one, one or more lanes of the same type can be determined, and subsequent evaluations are performed according to different lane types to ensure the effectiveness and reliability of the evaluation.

[0051] Calculating the degree of fit between the lane curve of the simulated lane and the lane curve of the real lane can be used to evaluate the degree of restoration of complex features such as shape and curvature of the simulated lane in the simulation environment. The degree of fit of the lane curve directly affects the autonomous driving system's ability to predict and respond to road changes.

[0052] After confirming that the lane types are the same and the degree of fit meets the preset requirements, the mapping relationship between the simulated lane and the real lane can be established. This mapping relationship is not only the basis for the subsequent calculation of the simulated perception error and the real perception error, but also the key to achieving seamless connection between the simulated environment and the real environment. Through the mapping relationship, it can be ensured that the simulated environment perception data can be accurately matched with the real environment perception data in the subsequent evaluation of the performance of the autonomous driving system, thereby improving the accuracy, effectiveness and reliability of subsequent related evaluations.

[0053] Optionally, see Figure 4 , calculate the degree of fit between the lane curve of the simulated lane and the lane curve of the real lane, which can be performed as follows:

[0054] Step 141: Calculate the frame fitting degree between the lane curve of the simulated lane of each frame and the lane curve of the real lane of the corresponding frame based on a preset fitting degree algorithm.

[0055] Step 142: weighted average the frame fitting degrees according to the appearance time of the simulated lane in the simulated environment perception data of each frame, and use the weighted average result as the fitting degree between the lane curve of the simulated lane and the lane curve of the real lane.

[0056] In the process of evaluating the perception confidence of the simulated lane, the degree of fit of the lane curve is the similarity between the simulated lane and the real lane, which can further evaluate the ability of the simulation environment to restore the complex features of the real environment and determine the accuracy of the lane type judgment. The preset degree of fit algorithm is a mathematical or statistical method used to calculate the similarity between two curves, such as error algorithm, least squares method, curve matching algorithm and distance algorithm. In this calculation process, the reliability of the calculation can also be further improved by adding preprocessing and filtering algorithms. Through the preset degree of fit algorithm, the lane curve of each frame of the simulated lane can be compared one by one with the lane curve of the real lane corresponding to the corresponding frame, and the similarity or difference between the two can be calculated, so as to obtain an indicator that quantitatively describes the similarity between the two curves.

[0057] However, since the lane curve of the simulated lane and the lane curve of the real lane are both perceived and determined by the automatic driving system, the same lane may appear in multiple frames during the perception and determination process, and since the same lane has different acquisition angles, different vehicle postures, and different surrounding environments in different frames, the degree of fit between the simulated lane and the real lane in different frames fluctuates to a certain extent. In this regard, this embodiment performs a weighted average of the frame fit degree according to the appearance time of the simulated lane in the simulated environment perception data of each frame, thereby determining the fit degree of the corresponding lane.

[0058] According to the appearance time, when the virtual vehicle or the physical vehicle is in the center of the lane, its perception result is usually relatively accurate, while the accuracy of the perception result at both ends of the lane will decrease. Therefore, according to the weighting of the appearance time, more weight is set in the middle of the lane appearance time, and less weight is gradually set towards the front and back of the time. For example, in the perception process of a certain lane, five frames of simulated environment perception data and five frames of real environment perception data are obtained, which correspond to V1-R1, V2-R2, V3-R3, V4-R4 and V5-R5 in time sequence, and the obtained frame fitting degrees are 0.91, 0.89, 0.92, 0.90, and 0.83, respectively. The weights from the middle to the two ends are 4, 2, and 1, respectively. Then the obtained fitting degree value is (0.91+0.89*2+0.92*4+0.90*2+0.83*1) / 10, and the fitting degree of the corresponding lane is calculated to be 0.90. It should be noted that this example is only one way of weighted calculation. The specific weight allocation strategy can be adjusted according to the actual application scenario and the complexity of the simulation environment. Different weighting methods and weights can be used. As long as the corresponding weighting can be adjusted according to accuracy and occurrence time, it can be understood to be within the protection scope of this solution.

[0059] Through this weighted processing, the fitting degree, stability and effectiveness of the lane curve in the entire simulation process can be better reflected, so that the obtained fitting degree can more accurately and reliably reflect the fitting situation of the lane curve in the entire simulation process, and the evaluation of the fitting degree is more in line with the dynamic change characteristics in the actual driving process. It can more comprehensively reflect the simulation environment's ability to restore the complex characteristics of the real environment, providing a more accurate and reliable basis for subsequent evaluation.

[0060] Optionally, the preset fitting degree algorithms include mean square error, root mean square error, mean absolute error, determination coefficient and Hausdorff distance; and the frame fitting degree is calculated between lane curves of different lane types using a preset fitting degree algorithm corresponding to one of them.

[0061] Mean-Square Error (MSE) is an algorithm that calculates the average of the sum of the squares of the point-to-point distances between two curves. The smaller the mean square error, the better the fit. Its formula is: Where n is the number of corresponding points between the lane curve of the simulated lane and the lane curve of the real lane, y i and They are the corresponding points of the lane curve of the simulated lane and the lane curve of the real lane respectively.

[0062] The root mean square error (RMSE) is the square root of the mean square error, which can more intuitively reflect the magnitude of the error. Its formula is: Its parameter definition is the same as that of the mean square error and will not be repeated here.

[0063] Mean Absolute Error (MAE) is a way to calculate the average of the absolute values ​​of the point-to-point distances between two curves. Compared with the mean square error, it is less sensitive to outliers. Its formula is: Its parameter definition is the same as that of the mean square error and will not be repeated here.

[0064] The coefficient of determination is symbolized by R 2 It reflects the explanatory power of the simulation environment. 2 The closer it is to 1, the better the fitting effect. Its formula is in, is the mean of the observed values. The definitions of other parameters are the same as those of the mean square error and will not be repeated here.

[0065] Hausdorff distance is an algorithm that can take into account the global characteristics of the curve and measure the maximum distance between two curves. Its formula is Among them, A and B represent the point sets of the lane curve of the simulated lane and the lane curve of the real lane respectively, ∥ba∥ represents the Euclidean distance between each two points in the two point sets, sup is the supremum, and inf is the infimum.

[0066] The above different algorithms have different focuses. For example, mean square error and root mean square error focus more on the size of the quantified error, mean absolute error provides a robust error index, the coefficient of determination evaluates the overall fitting quality of the simulation environment, and Hausdorff distance focuses on the evaluation of shape similarity. Therefore, by combining these algorithms according to the specific lane type and evaluation requirements, one or more of the above preset fitting degree algorithms can be selected to calculate the degree of fit of the lane curve between the simulated lane and the real lane, which can more comprehensively evaluate the ability of the simulated lane to restore the real environment in terms of complex features such as shape and curvature.

[0067] Step 20: Calculate the simulated perception error between the simulated environment perception data and the perception truth value of the real lane, and the real perception error between the real environment perception data and the perception truth value corresponding to the real lane.

[0068] The perception truth value is obtained by annotating the real environment perception data with real lanes. For example, the point cloud data in the real environment perception data can be superimposed in two dimensions into a two-dimensional plane to reconstruct the road condition information, and then the lane center, line type, road edge and lane line correlation data are annotated and corrected manually or by high-precision machines. The annotation result is used as the perception truth value through reasonable and accurate annotation. Since the perception truth value is the result of the truth value annotation, it usually has extremely high accuracy and can reflect the actual and accurate situation of the real environment.

[0069] By calculating the difference between the simulated environment perception data and the real environment perception data and the true value respectively, the difference between the perception and the correct situation can be determined by the corresponding simulated perception error and the real perception error, reflecting the accuracy and reliability of the perception. The calculation of the error can also be based on the aforementioned mean square error, root mean square error and mean absolute error, etc., or other specific comparison models or algorithms can be used. At the same time, before calculating the simulated perception error and the real perception error, the simulated environment perception data and the real environment perception data can be aligned with the perception true value according to the comparison length to ensure the accuracy and comparability of the error calculation. The alignment process can be based on number matching, timestamp matching or spatial position calibration. In addition, it should be noted that the simulated perception error and the real perception error can further optimize the error calculation by introducing positive and negative directionality. For example, when the perceived lane is on the left side of the true value lane or on the right side of the true value lane, the error can be set to positive or negative for distinction, so as to better process the corresponding error in the future and determine the direction of error processing.

[0070] Optionally, the simulated perception error between the simulated environment perception data and the perception true value of the real lane, and the real perception error between the real environment perception data corresponding to the real lane and the perception true value are calculated, including: in the simulated lane and the real lane of the same lane type, the simulated perception error between the simulated environment perception data and the perception true value is calculated, and the real perception error between the real environment perception data corresponding to the mapping relationship and the perception true value is calculated.

[0071] In the process of calculating the simulated perception error and the real perception error, the simulated environment perception data and the real environment perception data used for calculation can be determined based on the lane type and the mapping relationship, and the dimension of error evaluation can be further refined. Among them, subsequent evaluations are carried out according to different lane types, so that the corresponding evaluations are more consistent and accurate. Subsequently, the lane line perception of a certain lane type can be optimized, or the corresponding simulation environment can be optimized. Based on the mapping relationship, it can be guaranteed that the subsequent perception confidence evaluation of the simulated lane can be based on the real perception error of the corresponding real lane, ensuring the effectiveness and reliability of the evaluation.

[0072] Step 30: Evaluate the perception confidence of the simulated lane based on the simulated perception error and the real perception error.

[0073] Perception confidence is an indicator used to measure the effectiveness and reliability of the simulation environment and the simulated lanes used for perception, as well as the degree of fit between the simulation environment and the real environment. A simulation environment with high perception confidence can be better used to test the perception performance of the autonomous driving system to ensure that the autonomous driving system can be better applied to the real environment. However, a simulation environment with low perception confidence usually has a large deviation from the real environment due to lack of noise or the degree of restoration of the real environment. It cannot well reflect the situations encountered by the autonomous driving system in the real environment and the correct autonomous driving processing, which is not conducive to the testing and functional optimization of the corresponding autonomous driving system.

[0074] After obtaining the simulated perception error and the real perception error as mentioned above, the two errors can be compared to realize the evaluation of the perception confidence of the simulated lane. For example, when evaluating the perception confidence of the simulated lane, the simulated perception error and the real perception error can be compared and analyzed. If the simulated perception error and the real perception error are close in value, or the change trends of the two are similar, then it can be considered that the perception confidence of the simulated lane is high; the evaluation criteria can also be further refined, and a hierarchical evaluation method can be adopted to conduct a more detailed evaluation by comparing the distribution of errors, comparing by time series, and evaluating by influencing factors. In the process of evaluation, different evaluation methods can also be adopted for simulation environments of different scenarios, and different algorithms, models and parameter settings can also be used to make the evaluation more intuitive and effective, so as to accurately reflect the real performance of the simulation environment.

[0075] Through the above-mentioned evaluation of the perception confidence of the simulated lane based on the simulated perception error and the real perception error, it is possible to achieve an accurate, efficient and reliable evaluation of the perception confidence of the simulated lane, and comprehensively judge whether the perception of the lane in the corresponding simulation environment meets the perception needs in the real environment, which is conducive to the subsequent targeted optimization of the development and testing of the simulation environment of the autonomous driving system.

[0076] Optionally, see Figure 5 , based on the simulated perception error and the real perception error, the perception confidence of the simulated lane can be evaluated as follows:

[0077] Step 31: When the simulated perception error is greater than a first preset threshold of the real perception error, it is determined that the perception confidence of the simulated lane is false.

[0078] Step 32: When the simulated perception error is less than a second preset threshold of the real perception error, it is determined that the perception confidence of the simulated lane lacks noise.

[0079] Based on the obtained simulated perception error and real perception error, there will be two types of results: false-negative and false-positive. Among them, in the case of false-negative results, the simulated perception error is greater than the real perception error, and there is a greater difference between the simulated environment perception data and the perception true value, indicating that the perception effect of the simulated lane is worse than that of the real lane. This situation is often due to omissions or defects in the simulation environment when simulating the real environment, resulting in certain features or situations of the simulated lane not being fully or accurately reproduced, resulting in a lack of truth. In the case of false-positive results, the simulated perception error is less than the real perception error, and the simulated perception result is closer to the perception true value, indicating that the perception effect of the simulated lane is better than that of the actual lane. This situation is often due to the lack of noise, interference or other unforeseen factors in the real world in the simulation environment, resulting in abnormally accurate perception data in the simulation environment and a lack of noise.

[0080] On this basis, the first preset threshold and the second preset threshold can be set accordingly. The setting of the first preset threshold and the second preset threshold can be based on actual test data and empirical analysis to ensure the accuracy and reliability of the evaluation. In practical applications, these thresholds can also be set to values ​​that can be dynamically adjusted according to the specific simulation environment and the needs of the autonomous driving system.

[0081] When the simulated perception error is greater than the first preset threshold of the real perception error, it means that the simulated perception has a larger error than the real perception and deviates seriously from the real perception, and it is considered that the lack of truth is more serious. When the simulated perception error is less than the second preset threshold of the real perception error, it means that the simulated perception has a smaller error than the real perception but deviates seriously from the real perception, and it is considered that the lack of noise is more serious.

[0082] The lack of truth and lack of noise results of this method can effectively evaluate the inherent defects of the simulation environment and reveal the two defect situations. Through the above analysis and comparison of the lack of truth and lack of noise based on the simulated perception error and the real perception error, it is possible to accurately evaluate the perception confidence of the simulated lane. This evaluation result not only helps to understand the real performance and reliability of the simulation environment, and achieve accurate, efficient and reliable evaluation of the perception confidence of the simulated lane, but also can provide strong support for the subsequent optimization of the simulation environment and the test optimization of the autonomous driving system.

[0083] See also Figure 6 , Figure 6 It is a structural schematic diagram of an embodiment of a perception confidence assessment device for a simulated lane provided in the present application.

[0084] The perception confidence assessment device 40 of the simulated lane includes: an acquisition module 41, used to obtain simulated environment perception data of the simulated lane in a simulated environment, wherein the simulated lane is obtained based on the digital twin reconstruction of the real lane; a calculation module 42, used to calculate the simulated perception error between the simulated environment perception data and the perception true value of the real lane, and the real perception error between the real environment perception data corresponding to the real lane and the perception true value; an evaluation module 43, used to evaluate the perception confidence of the simulated lane based on the simulated perception error and the real perception error.

[0085] Optionally, the acquisition module 41 is also specifically used to: obtain real environment perception data generated by the physical vehicle sensing the real lane in the real environment, and record the vehicle posture data corresponding to each frame of the real environment perception data; based on the vehicle posture data, reproduce the trajectory of the virtual vehicle in the simulation environment, and use the same perception algorithm and perception points as the physical vehicle to control the virtual vehicle to perceive the simulated lane in the simulation environment, so as to obtain the simulation environment perception data, wherein the physical vehicle is a vehicle equipped with a pre-installed autonomous driving system, and the virtual vehicle is the body mapping of the autonomous driving system in the simulation environment.

[0086] Optionally, the simulated environment perception data includes a lane type and a lane curve of a simulated lane, and the real environment perception data includes a lane type and a lane curve of a real lane.

[0087] Optionally, after the acquisition module 41 acquires the simulation environment perception data of the simulated lane in the simulation environment, the calculation module 42 is specifically used to: confirm whether the lane type of the simulated lane and the lane type of the real lane are the same; calculate the degree of fit between the lane curve of the simulated lane and the lane curve of the real lane; establish a mapping relationship between the simulated lane and the real lane with the same lane type and the degree of fit within a preset fitting range, wherein the simulation perception error and the real perception error are calculated based on the simulated lane and the real lane having a mapping relationship.

[0088] Optionally, the calculation module 42 is further used to calculate the frame fitting degree between the lane curve of the simulated lane of each frame and the lane curve of the real lane of the corresponding frame based on a preset fitting degree algorithm. The frame fitting degrees are weighted averaged according to the appearance time of the simulated lane in the simulated environment perception data of each frame, and the weighted average result is used as the fitting degree between the lane curve of the simulated lane and the lane curve of the real lane.

[0089] Optionally, the preset fitting degree algorithms include mean square error, root mean square error, mean absolute error, determination coefficient and Hausdorff distance; and the frame fitting degree is calculated between lane curves of different lane types using a preset fitting degree algorithm corresponding to one of them.

[0090] Optionally, the calculation module 42 is also specifically used to calculate the simulated perception error between the simulated environment perception data and the perception true value, and to calculate the real perception error between the real environment perception data corresponding to the mapping relationship and the perception true value in the simulated lane and the real lane of the same lane type.

[0091] Optionally, the evaluation module 43 is also specifically used to: determine that the perception confidence of the simulated lane lacks truth when the simulated perception error is greater than a first preset threshold of the real perception error; and determine that the perception confidence of the simulated lane lacks noise when the simulated perception error is less than a second preset threshold of the real perception error.

[0092] Since the embodiments of the device part correspond to the embodiments of the above-mentioned method, the introduction of the simulated lane perception confidence assessment device 40 provided in the embodiment of the present invention and the corresponding beneficial effects can refer to the above-mentioned simulated lane perception confidence assessment method embodiment, and this embodiment will not be repeated here.

[0093] See also Figure 7 , Figure 7 It is a structural diagram of an embodiment of the storage medium provided by the present application.

[0094] The storage medium 50 stores program data 51. When the program data 51 is executed by the processor, the following is achieved: Figures 1 to 5 The described perception confidence assessment method for simulated lanes.

[0095] The program data 51 is stored in a storage medium 50, and includes a number of instructions for enabling a network device (such as a router, a personal computer, a server, or other network device) or a processor to execute all or part of the steps of the method described in each embodiment of the present application.

[0096] Optionally, the storage medium 50 may be a U disk, a mobile hard disk, a read-only memory (ROM), a random access memory (RAM), a magnetic disk, an optical disk, or other medium that can store the program data 51 .

[0097] See also Figure 8 , Figure 8 It is a structural schematic diagram of an embodiment of a computer device provided by the present application.

[0098] The computer device 60 includes a processor 62 and a memory 61 connected to each other. The memory 61 stores a computer program. When the processor 62 executes the computer program, the following is achieved: Figures 1 to 5 The described perception confidence assessment method of the simulated lane. The memory 61 may include the storage medium 50, or may be other independently developed memory.

[0099] Different from the prior art, the present application discloses a perception confidence evaluation method for a simulated lane and a related device. By respectively calculating the simulated environment perception data obtained by perceiving the simulated lane and the real environment perception data obtained by perceiving the real lane, the corresponding simulated perception error and real perception error between the perception true value are obtained, thereby being able to achieve an accurate, efficient and reliable evaluation of the perception confidence of the simulated lane, and being able to comprehensively judge whether the perception of the lane in the corresponding simulated environment meets the perception needs in the real environment, which is conducive to the subsequent targeted optimization of the development and testing of the simulation environment of the autonomous driving system.

[0100] Each embodiment in this specification is described in a progressive manner, and the same or similar parts between the embodiments can be referred to each other, and each embodiment focuses on the differences from other embodiments. In particular, for the device embodiment, storage medium embodiment and computer equipment embodiment, since they are basically similar to the method embodiment, the description is relatively simple, and the relevant parts can be referred to the partial description of the method embodiment.

[0101] The present application can be used in many general or special computing system environments or configurations, such as personal computers, server computers, handheld or portable devices, tablet devices, multiprocessor systems, microprocessor-based systems, set-top boxes, programmable consumer electronic devices, network PCs, minicomputers, mainframe computers, distributed computing environments including any of the above systems or devices, etc.

[0102] In the several embodiments provided in this application, it should be understood that the disclosed methods, devices, storage media, and computer equipment can be implemented in other ways. For example, the device implementation described above is only illustrative, for example, the division of the modules or units is only a logical function division, and there may be other division methods in actual implementation, such as multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed.

[0103] The units described as separate components may or may not be physically separated, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed on multiple network units. Some or all of the units may be selected according to actual needs to achieve the purpose of the present embodiment.

[0104] In addition, each functional unit in each embodiment of the present application may be integrated into one processing unit, or each unit may exist physically separately, or two or more units may be integrated into one unit. The above-mentioned integrated unit may be implemented in the form of hardware or in the form of software functional units.

[0105] The above descriptions are merely embodiments of the present application and are not intended to limit the patent scope of the present application. Any equivalent structure or equivalent process transformation made using the contents of the present application specification and drawings, or directly or indirectly applied in other related technical fields, are also included in the patent protection scope of the present application.

Claims

1. A method for evaluating perception confidence of a simulated lane, characterized in that: include: Acquire simulation environment perception data of a simulation lane in a simulation environment, wherein the simulation lane is obtained based on a digital twin reconstruction of a real lane; Calculating a simulated perception error between the simulated environment perception data and a perception true value of the real lane, and a real perception error between the real environment perception data corresponding to the real lane and the perception true value; The perception confidence of the simulated lane is evaluated based on the simulated perception error and the real perception error.

2. The method for evaluating the perception confidence of a simulated lane according to claim 1, characterized in that: The simulation environment perception data includes the lane type and lane curve of the simulation lane, and the real environment perception data includes the lane type and lane curve of the real lane; After acquiring the simulation environment perception data of the simulation lane in the simulation environment, the method further includes: confirming whether the lane type of the simulated lane is the same as the lane type of the real lane; Calculating the degree of fit between the lane curve of the simulation lane and the lane curve of the real lane; Establishing a mapping relationship between the simulated lane and the real lane whose lane type is the same and whose fitting degree is within a preset fitting degree range; The simulated perception error and the real perception error are calculated based on the simulated lane and the real lane having a mapping relationship.

3. The method for evaluating the perception confidence of a simulated lane according to claim 2, characterized in that: The calculating the degree of fit between the lane curve of the simulation lane and the lane curve of the real lane includes: Calculating the frame fitting degree of the lane curve of the simulated lane in each frame and the lane curve of the real lane in the corresponding frame based on a preset fitting degree algorithm; The frame fitting degree is weighted averaged according to the appearance time of the simulated lane in the simulated environment perception data of each frame, and the result of the weighted average is used as the fitting degree between the lane curve of the simulated lane and the lane curve of the real lane.

4. The method for evaluating the perception confidence of a simulated lane according to claim 3, characterized in that: The preset fitting degree algorithms include mean square error, root mean square error, mean absolute error, determination coefficient and Hausdorff distance; the frame fitting degree is calculated between the lane curves of different lane types using the preset fitting degree algorithm corresponding to one of them.

5. The method for evaluating the perception confidence of a simulated lane according to claim 2, characterized in that: The calculating of the simulated perception error between the simulated environment perception data and the perception true value of the real lane, and the real perception error between the real environment perception data corresponding to the real lane and the perception true value, includes: In the simulated lane and the real lane of the same lane type, a simulated perception error between the simulated environment perception data and the perception true value is calculated, and a real perception error between the real environment perception data corresponding to the mapping relationship and the perception true value is calculated.

6. The method for evaluating the perception confidence of a simulated lane according to claim 1, characterized in that: The obtaining of simulation environment perception data of a simulation lane in a simulation environment includes: Acquire the real environment perception data generated by the physical vehicle sensing the real lane in the real environment, and record the vehicle posture data corresponding to each frame of the real environment perception data; Based on the vehicle posture data, the trajectory of the virtual vehicle in the simulation environment is reproduced, and the same perception algorithm and perception points as those of the physical vehicle are used to control the virtual vehicle to perceive the simulated lane in the simulation environment, so as to obtain the simulation environment perception data; The physical vehicle is a vehicle equipped with a pre-installed autonomous driving system, and the virtual vehicle is a vehicle body mapping of the autonomous driving system in a simulation environment.

7. The method for evaluating the perception confidence of a simulated lane according to claim 1, characterized in that: The evaluating the perception confidence of the simulated lane based on the simulated perception error and the real perception error includes: When the simulated perception error is greater than a first preset threshold of the real perception error, determining that the perception confidence of the simulated lane is false; When the simulated perception error is less than a second preset threshold of the real perception error, it is determined that the perception confidence of the simulated lane lacks noise.

8. A perception confidence assessment device for a simulated lane, characterized in that: include: An acquisition module, used to acquire simulation environment perception data of a simulation lane in a simulation environment, wherein the simulation lane is obtained based on a digital twin reconstruction of a real lane; a calculation module, configured to calculate a simulated perception error between the simulated environment perception data and a perception true value of the real lane, and a real perception error between the real environment perception data corresponding to the real lane and the perception true value; An evaluation module is used to evaluate the perception confidence of the simulated lane based on the simulated perception error and the real perception error.

9. A storage medium having program data stored thereon, characterized in that: When the program data is executed by a processor, the steps of the method for evaluating the perception confidence of a simulated lane as described in any one of claims 1 to 7 are implemented.

10. A computer device, characterized in that: It comprises a processor and a memory connected to each other, the memory stores a computer program, and when the processor executes the computer program, the steps of the perception confidence assessment method of a simulated lane as described in any one of claims 1 to 7 are implemented.

Images