Driving behavior analysis method and related equipment

Abstract

The invention provides a driving behavior analysis method and related equipment. The driving behavior analysis method comprises the steps of obtaining target data of vehicle detection; inputting the target data into a decision model to obtain a target driving behavior to be executed by the vehicle and first reason information output by the decision model, the first reason information being used for describing a reason for determining the target driving behavior by the decision model; safety analysis is conducted on the target driving behavior, at least one piece of first interpretation information of the target driving behavior is obtained, and the first interpretation information is used for describing the safety of the vehicle executing the target driving behavior; and outputting the target driving behavior, the first reason information and the at least one piece of first interpretation information. According to the invention, the explanation of the decision model on the driving behavior is enriched.

Description

Technical Field

[0001] This application relates to the field of autonomous driving technology, and in particular to a driving behavior analysis method and related equipment. Background Technology

[0002] With the development of multimodal large models, decision-making models are set up in vehicles to determine the driving behavior of the vehicle.

[0003] In the exemplary technology, the decision model has the ability to explain the driving behavior it has decided on, specifically by explaining the reasons for the driving behavior in natural language, such as "stopping because of a red light".

[0004] Although the decision-making model provides causal explanations for driving behavior, causal explanations are not equivalent to the driving behavior decided by the decision-making model being safe and reliable. The model's explanation of driving behavior is relatively simplistic. Summary of the Invention

[0005] Based on the aforementioned technological status, this application provides a driving behavior analysis method and related equipment to address the problem that the model's explanation of driving behavior is relatively simplistic.

[0006] To achieve the above-mentioned technical objectives, this application proposes the following technical solution: Firstly, this application provides a method for analyzing driving behavior, including: Acquire target data for vehicle inspection; The target data is input into the decision model to obtain the target driving behavior to be performed by the vehicle and the first reason information output by the decision model. The first reason information is used to describe the reason why the decision model determines the target driving behavior. A safety analysis is performed on the target driving behavior to obtain at least one first explanatory information of the target driving behavior, wherein the first explanatory information is used to describe the safety of the vehicle performing the target driving behavior; Output the target driving behavior, the first cause information, and the at least one first explanation information.

[0007] In some implementations, the first explanatory information includes behavioral consistency explanatory information, and the safety analysis of the target driving behavior includes: The first object in the target data is processed to obtain the second data containing the second object, wherein the first object is the object mentioned in the first reason information, and the second object is the first object that is occluded, the first object after the behavior is modified, or a preset object that replaces the first object. The second data is input into the decision model to obtain the first driving behavior and the second cause information corresponding to the first driving behavior; Based on the first object, the third object mentioned in the second reason information, the first driving behavior, and the target driving behavior, determine the explanation information for behavioral consistency.

[0008] In some implementations, the first explanatory information includes explanatory information on attention alignment, and the safety analysis of the target driving behavior includes: Based on the target data, determine the image region where the first object mentioned in the first reason information is located; Based on the image region, the matching degree between the first cause information and the attention layer in the decision model is determined as explanatory information for attention alignment.

[0009] In some implementations, determining the matching degree between the first cause information and the attention layer in the decision model based on the image region includes: Determine the attention hotspots corresponding to the attention layers in the decision model; Determine the first intersection region and the union region between the attention hotspot and the image region; The matching degree between the first cause information and the attention layer is determined based on the ratio between the first intersection region and the union region.

[0010] In some implementations, determining the matching degree between the first cause information and the attention layer in the decision model based on the image region includes: Determine the heatmap corresponding to the decision layer in the decision model; Determine the second intersection region between the heatmap and the image region; The matching degree between the first cause information and the attention layer is determined based on the ratio between the second intersection region and the image region.

[0011] In some implementations, determining the matching degree between the first cause information and the attention layer in the decision model based on the image region includes: Determine the first text vector corresponding to the first cause information and the image vector corresponding to the image region; The matching degree between the first cause information and the attention layer is determined based on the first similarity between the first text vector and the image vector.

[0012] In some implementations, the first explanatory information includes explanatory information on model stability and robustness, and the safety analysis of the target driving behavior includes: Obtain target parameters, which include at least one of model stability parameters and time consistency parameters; Based on the target parameters, determine the explanatory information for model stability and robustness.

[0013] In some implementations, the target parameters include model stability parameters, and obtaining the target parameters includes: The parameters in the target data are modified by a target amplitude to obtain third data, wherein the target amplitude is less than a preset amplitude; The third data is input into the decision model to obtain the third cause information corresponding to the second driving behavior output by the decision model; Determine the second text vector corresponding to the first reason information and the third text vector corresponding to the third reason information; The model stability parameter is determined based on the second similarity between the second text vector and the third text vector.

[0014] In some implementations, the target parameter includes a time consistency parameter, and obtaining the target parameter based on the target data includes: Obtain the historical cause information corresponding to the driving behavior output by the decision model at each historical moment, wherein each historical moment is continuous in time with the current moment; Determine the historical text vector corresponding to each of the historical cause information, and determine the third similarity between the historical text vectors corresponding to adjacent historical moments; Based on each of the aforementioned third similarities, the time consistency parameters are determined.

[0015] In some implementations, the first explanatory information includes assessment information of model risk perception, and the safety analysis of the target driving behavior includes: Based on the target data, determine the fourth objects that need to be considered when driving the vehicle and the first risk factors that exist when driving the vehicle. Based on the first cause information, determine the first objects and second risk factors that the decision model focuses on; The safety recall rate is determined based on the ratio between the number of the first objects and the number of the fourth objects, and the risk omission rate is determined based on each of the first risk factors and each of the second risk factors. The risk perception assessment information of the model is determined based on the safety recall rate and the risk omission rate.

[0016] In some implementations, the first explanatory information includes descriptive information about the model's safety performance, and the safety analysis of the target driving behavior includes: Acquire multiple target information, including at least two of the following: explanatory information on behavioral consistency, explanatory information on attention alignment, explanatory information on model stability and robustness, and evaluation information on model risk perception. The first driving scenario of the vehicle is determined based on the target data, and the first safety performance parameter of the decision model in the first driving scenario is determined based on the first driving scenario and each of the target information. The decision model is obtained in the second driving scenario, and the safety performance description information of the model is determined based on the first safety performance parameter and the second safety performance parameter. The first driving scenario is different from the second driving scenario.

[0017] Secondly, this application provides a driving behavior analysis device, including a memory and a processor, wherein, The memory is connected to the processor and is used to store programs; The processor is used to implement the driving behavior analysis method as described in the first aspect or any implementation thereof by running a program in the memory.

[0018] Thirdly, this application provides a vehicle that includes a driving behavior analysis device, which implements the driving behavior analysis method as described in the first aspect or any implementation thereof.

[0019] Fourthly, this application provides a computer program product, which, when executed by a processor, implements the driving behavior analysis method as described in the first aspect or any implementation thereof.

[0020] Fifthly, this application provides a computer-readable storage medium storing a computer program, which, when executed by a processor, implements the driving behavior analysis method as described in the first aspect or any implementation thereof.

[0021] This application provides a driving behavior analysis method and related equipment. The method inputs target data from vehicle detection into a decision model, obtains the target driving behavior output by the decision model, and causal information explaining the cause of the target driving behavior. Then, a safety analysis is performed on the target driving behavior to obtain at least one explanatory information describing the safety of the vehicle performing the target driving behavior. Finally, the method outputs the explanatory information, the target driving behavior, and the causal information. In this application, after deciding on the driving behavior, the decision model performs a safety analysis on the driving behavior to obtain explanatory information describing the safety of the vehicle performing the driving behavior. That is, the explanatory information and causal information enrich the decision model's explanation of the driving behavior. Attached Figure Description

[0022] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only embodiments of this application. For those skilled in the art, other drawings can be obtained based on the provided drawings without creative effort.

[0023] Figure 1 A flowchart of a driving behavior analysis method provided in this application embodiment Figure 1 .

[0024] Figure 2 A flowchart of a driving behavior analysis method provided in this application embodiment Figure 2 .

[0025] Figure 3 A flowchart of a driving behavior analysis method provided in this application embodiment Figure 3 .

[0026] Figure 4 A flowchart of a driving behavior analysis method provided in this application embodiment Figure 4 .

[0027] Figure 5 A flowchart of a driving behavior analysis method provided in this application embodiment Figure 5 .

[0028] Figure 6 A flowchart of a driving behavior analysis method provided in this application embodiment Figure 6 .

[0029] Figure 7 This is a schematic diagram of the functional modules of a driving behavior analysis device provided in an embodiment of this application.

[0030] Figure 8 This is a structural schematic diagram of a vehicle provided in an embodiment of this application. Detailed Implementation

[0031] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0032] It should be noted that the user information (including but not limited to electrical equipment information, user personal information, etc.) and data (including but not limited to data used for analysis, data stored, data displayed, etc.) involved in this application are all information and data authorized by the user or fully authorized by all parties. Furthermore, the collection, use and processing of the relevant data must comply with relevant laws, regulations and standards, and corresponding operation entry points are provided for users to choose to authorize or refuse.

[0033] With the development of multimodal large models, decision-making models are set up in vehicles to determine the driving behavior of the vehicle.

[0034] In the exemplary technology, the decision model has the ability to explain the driving behavior it has decided on, specifically by explaining the reasons for the driving behavior in natural language, such as "stopping because of a red light".

[0035] Although the decision-making model provides causal explanations for driving behavior, causal explanations are not equivalent to the driving behavior decided by the decision-making model being safe and reliable. The model's explanation of driving behavior is relatively simplistic.

[0036] To address the aforementioned issues, this application provides a driving behavior analysis method. The method inputs target data from vehicle detection into a decision model, obtaining the target driving behavior output by the model and causal information explaining the reasons for the target driving behavior. Then, a safety analysis is performed on the target driving behavior to obtain at least one explanatory piece of information describing the safety of the vehicle performing the target driving behavior. Finally, the method outputs the explanatory information, the target driving behavior, and the causal information. In this application, after determining the driving behavior, the decision model performs a safety analysis to obtain explanatory information describing the safety of the vehicle performing the driving behavior. In other words, the explanatory information and causal information enrich the decision model's explanation of the driving behavior.

[0037] The driving behavior analysis method proposed in this application will be described in detail below through various embodiments.

[0038] Reference Figure 1 , Figure 1 A flowchart of a driving behavior analysis method provided in this application embodiment Figure 1 .like Figure 1 As shown, the driving behavior analysis method provided in this embodiment includes: Step S101: Obtain the target data for vehicle detection.

[0039] In this embodiment, the executing entity is a driving behavior analysis device. This device can be a terminal device or server with driving behavior analysis capabilities for model output, or it can be the vehicle itself or a component within the vehicle. For ease of description, the term "device" will be used to refer to the driving behavior analysis device below.

[0040] The device acquires data currently detected by the vehicle, which is defined as target data. Target data can be the vehicle's perception data, such as images, bird's-eye views, point cloud data, etc., collected by the vehicle.

[0041] Step S102: Input the target data into the decision model to obtain the target driving behavior to be executed by the vehicle and the first cause information output by the decision model. The first cause information is used to describe the reason why the decision model determines the target driving behavior.

[0042] The vehicle is equipped with a decision-making model that can plan the driving behavior for the next moment, such as braking, steering, using turn signals, accelerating, decelerating, and honking the horn. The decision-making model can be a deep learning model.

[0043] After obtaining the target data, the device inputs it into the decision model. The decision model uses the target data to determine the driving behavior and the causal information for that behavior. The driving behavior determined by the decision model is defined as the target driving behavior, and the causal information for the target driving behavior is defined as the primary causal information. The primary causal information describes the reason for the target driving behavior determined by the decision model. For example, the primary causal information might be "braking due to a red light ahead," where "braking" is the target driving behavior.

[0044] Step S103: Perform a safety analysis on the target driving behavior to obtain at least one first explanatory information of the target driving behavior. The first explanatory information is used to describe the safety of the vehicle performing the target driving behavior.

[0045] After determining the target driving behavior, a safety analysis is performed on the target driving behavior to obtain at least one explanatory information of the target driving behavior. The explanatory information is defined as the first explanatory information, which is used to describe the safety of the vehicle performing the target driving behavior. The first explanatory information may also be an assessment of the reliability, credibility, and availability of the vehicle performing the target driving behavior.

[0046] At least one of the following primary explanatory information includes: explanatory information on behavioral consistency, explanatory information on attention alignment, explanatory information on model stability and robustness, assessment information on model risk perception, and descriptive information on model safety performance.

[0047] Explanatory information for behavioral consistency refers to whether the primary cause information is logically consistent with the driving behavior decided by the decision model. For example, if the decision model explains "braking because there is a vehicle ahead" (primary cause information), but the actual situation is "slowing down because there are other vehicles ahead," then there is a logical inconsistency. This can be addressed by comparing the primary cause information with the actual information obtained from the target data, and then logically comparing the actual information with the primary cause information to obtain the comparison result, which serves as the explanatory information for behavioral consistency.

[0048] The explanatory information for attention alignment refers to the degree of matching between the primary cause information and the attention layer (the algorithmic layer) in the decision model. For example, if the decision model explains "braking because there is a pedestrian crossing ahead," then the attention alignment is high if the hotspot of the visual attention in the decision model falls on "pedestrian ahead," and the similarity between the area of ​​the pedestrian in the image and the relevant text at the text level is high. To address this, the hotspot corresponding to the attention layer in the decision model and the image region where the target object is located are extracted. The target object is a pedestrian or vehicle affecting the vehicle's movement. Based on the target data, the image region is the area where the target object is located in the target data image. The explanatory information for attention alignment is determined based on the hotspot and image region, and the image region and the description of the target object in the primary cause information. For example, if the hotspot falls on the "target object" and the similarity between the area where the behavior is located on the image and the relevant text at the text level is greater than a threshold, then "high attention alignment" is used as the explanation information for attention alignment; if the hotspot does not fall on the "target object" and / or the similarity between the area where the behavior is located on the image and the relevant text at the text level is less than or equal to a threshold, then "low attention alignment" is used as the explanation information for attention alignment.

[0049] The explanatory information for model stability and robustness refers to whether the decision model can maintain a stable output when encountering input perturbations (such as illumination, angle, or minor occlusion), and whether the decision model's interpretation is consistent and without jumps across consecutive frames captured by the vehicle. To this end, the perturbed target data is input into the decision model to obtain driving behavior 'a'. The decision model then determines multiple driving behaviors 'b' based on data from each image in consecutive frames. Driving behavior 'a' is compared with the target driving behavior to obtain a first comparison result, and each driving behavior 'b' is compared to obtain a second comparison result. The explanatory information for model stability and robustness is then determined based on the first and second comparison results. For example, if the first comparison result is that driving behavior a is the same as the target driving behavior, and the second comparison result is that all driving behaviors b are the same, then "the model has high stability and high robustness" is used as the explanatory information for model stability and robustness. If the first comparison result is that driving behavior a is different from the target driving behavior, and the second comparison result is that all driving behaviors b are the same, then "the model has low stability and high robustness" is used as the explanatory information for model stability and robustness. If the first comparison result is that driving behavior a is different from the target driving behavior, and the second comparison result is that all driving behaviors b are different, then "the model has low stability and low robustness" is used as the explanatory information for model stability and robustness. If the first comparison result is that driving behavior a is the same as the target driving behavior, and the second comparison result is that all driving behaviors b are different, then "the model has high stability and low robustness" is used as the explanatory information for model stability and robustness.

[0050] The assessment information representation for model risk perception evaluates the decision-making model's ability to explain abnormal scenarios and its ability to comprehensively address key safety events without overlooking any risks. To this end, the device determines the number of objects (a) requiring attention during vehicle operation based on target data, and the number of objects (b) requiring attention by the decision-making model based on first-cause information. The difference between these two numbers determines a parameter indicating the decision-making model's ability to explain abnormal scenarios. For example, if the difference is less than or equal to a threshold, the decision-making model has a high ability to explain abnormal scenarios; if the difference is greater than the threshold, the decision-making model has a low ability to explain abnormal scenarios. Furthermore, the device determines risk factors (a) requiring attention during vehicle operation based on target data, and risk factors (b) requiring attention by the decision-making model based on first-cause information. Based on these two risk factors, a parameter is determined indicating the decision-making model's ability to comprehensively address key safety events without overlooking any risks. For example, if each risk factor (b) overlaps with each risk factor (a), the decision-making model has a high ability to comprehensively address key safety events without overlooking any risks; if each risk factor (b) does not overlap with each risk factor (a), the decision-making model has a low ability to comprehensively address key safety events without overlooking any risks. The two capability parameters mentioned above determine the assessment information for the model's risk perception.

[0051] The descriptive information representing the overall safety performance of the decision-making model can be obtained from the four explanatory information items mentioned above under different scenarios. For example, in scenario a, the weights of each explanatory information item are different, and the weighted sum of each explanatory information item and its corresponding weight yields the first value; in scenario b, the weights of each explanatory information item are different, and the weighted sum of each explanatory information item and its corresponding weight yields the second value; the weighted sum of the first and second values ​​gives the descriptive information of the model's safety performance. The weight corresponding to the first value is determined based on scenario a, and the weight corresponding to the second value is determined based on scenario b. Scenario a refers to the vehicle's current driving scenario, and scenario b refers to the vehicle's past driving scenarios. There can be multiple scenarios b. Different scenarios can be divided into regular scenarios and long-tail scenarios. Long-tail scenarios refer to traffic scenarios with extremely low probability of occurrence, extremely rare occurrences, but numerous types and difficult to predict, such as the "ghost peek" scenario; scenarios other than long-tail scenarios are regular scenarios.

[0052] Step S104: Output the target driving behavior, the first cause information, and at least one first explanation information.

[0053] After determining the first explanation information, the device outputs the target driving behavior, the first cause information, and the first explanation information, so that the driver can determine whether to control the vehicle to drive the target driving behavior based on the first cause information and the first explanation information.

[0054] In addition, the device can output anomaly alerts and failure tracing suggestions. For example, if the first explanatory information mentioned above contains a description of poor performance of the decision model, an alert is output. For instance, inconsistencies in behavior, low attention alignment, poor model stability, poor model robustness, poor model explanation ability for abnormal scenarios, poor model risk perception ability, and / or low overall safety performance of the model would all result in an anomaly alert. The failure tracing suggestions can be the specific calculation process corresponding to the explanatory information of the anomaly alert. This calculation process can determine the cause of the poor model performance, thereby enabling targeted optimization of the decision model based on the cause.

[0055] It should be noted that the first cause information and the first explanation information output by the device can be output in various types. For example, the first cause information and the first explanation information can be output using natural language description, attention graph, or causal path. The format of the first cause information and the first explanation information output by the device is determined by the structure of the decision model. If the decision model includes natural language functionality, the first cause information and the first explanation information are output using natural language description; if the decision model includes an attention layer, the first cause information and the first explanation information are output using attention graph; if the decision model has causal inference capabilities, the first cause information and the first explanation information are output using causal path.

[0056] In this embodiment, the target data of vehicle detection is input into the decision model to obtain the target driving behavior and the causal information explaining the cause of the target driving behavior. Then, a safety analysis is performed on the target driving behavior to obtain at least one explanatory information describing the safety of the vehicle performing the target driving behavior. Finally, the explanatory information, the target driving behavior, and the causal information are output. In this embodiment, after deciding on the driving behavior, the decision model performs a safety analysis on the driving behavior to obtain explanatory information describing the safety of the vehicle performing the driving behavior. That is, the explanatory information and the causal information enrich the decision model's explanation of the driving behavior.

[0057] Reference Figure 2 , Figure 2 A flowchart of a driving behavior analysis method provided in this application embodiment Figure 2 ,based on Figure 1 In the embodiment shown, step S103 includes: Step S201: Process the first object in the target data to obtain second data containing the second object, wherein the first object is the object mentioned in the first reason information, and the second object is the first object that is occluded, the first object after the behavior is modified, or a preset object that replaces the first object.

[0058] In this embodiment, the first explanatory information is behavioral consistency explanatory information. The device first determines the first object from the first cause information. The first object is the object mentioned in the first cause information, such as a person or an object. For example, if the first cause information is "braking because a pedestrian is crossing ahead", then the pedestrian is the first object.

[0059] After identifying the first object, the device processes the first object in the target data to obtain a second object. The second object can be the first object that has been occluded, the first object after a modification, or a preset object that replaces the first object. For example, the first object can be occluded by a set area, which is a preset area, to determine whether the first object is related to the target driving behavior. In addition, if the first object is a pedestrian, "driving" can be changed to "vehicle"; or "pedestrian crossing" can be changed to "pedestrian moving away".

[0060] Step S202: Input the second data into the decision model to obtain the first driving behavior.

[0061] After obtaining the second data, the second data is input into the decision model. The decision model then outputs the driving behavior to be performed by the vehicle and the causal information explaining the cause of the driving behavior based on the second data. The driving behavior is defined as the first driving behavior, and the causal information is defined as the second causal information.

[0062] Step S203: Determine the explanation information for behavioral consistency based on the first object, the third object mentioned in the second cause information, the first driving behavior, and the target driving behavior.

[0063] After determining the first driving behavior, a third object is extracted from the second cause information; the third object is the object mentioned in the second cause information. The device determines explanatory information for behavioral consistency based on the first object, the third object, the first driving behavior, and the target driving behavior. For example, the first object and the third object are compared; if they are the same, the first score is 1; if they are different, the first score is 0. The first driving behavior and the target driving behavior are compared; if they are the same, the second score is 1; if they are different, the second score is 0. The product of the first score and the second score is used as the explanatory information for behavioral consistency.

[0064] It should be noted that there may be multiple primary objects, such as traffic lights, pedestrians, or vehicles. Therefore, the explanatory information for behavioral consistency is represented by the following formula: ; .

[0065] in, , These represent the scores for the actions of occlusion, replacement, and modification, respectively, and N represents the number of entities that affect the behavior (such as traffic lights, pedestrians, vehicles, etc.). For the first object For the third object, This refers to actual behavior (the target driving behavior can be considered as actual behavior). As the first line, A value of 1 indicates the same result, otherwise 0. If multiple security analyses were performed, the result is determined by... The mean value represents the behavioral consistency score.

[0066] In this embodiment, the objects mentioned in the cause information are processed to accurately determine the explanatory information for behavioral consistency.

[0067] Figure 3 A flowchart of a driving behavior analysis method provided in this application embodiment Figure 3 ,based on Figure 1 or Figure 2 In the embodiment shown, step S103 includes: Step S301: Based on the target data, determine the image region where the first object mentioned in the first cause information is located.

[0068] In this embodiment, the first explanatory information is attention alignment explanatory information. The device first determines the image region where the first object mentioned in the first cause information is located based on the target data.

[0069] For example, the device now extracts a first object from the first cause information, which is the object mentioned in the text of the first cause information. For instance, if the first cause information is "braking because a pedestrian is crossing ahead," the pedestrian is the first object. The target data includes an image, and the region in the image that contains the first object is the image region.

[0070] Step S302: Based on the image region, determine the matching degree between the first cause information and the attention layer in the decision model, as the explanatory information for attention alignment.

[0071] The decision model includes an algorithm layer, such as an attention layer. The device determines the degree of matching between the primary cause information and the attention layer in the decision model based on the image region, serving as explanatory information for attention alignment.

[0072] In one example, at least two of the following parameters are obtained: the overlap parameter of the attention hotspot, the overlap parameter of the decision hotspot, and the multimodal alignment strength parameter. The average value of these parameters can be used as explanatory information for attention alignment.

[0073] In another example, the attention hotspot corresponding to the attention layer in the decision model is determined, and the first intersection region and the union region between the attention hotspot and the image region are determined. The ratio between the first intersection region and the union region is used as the matching degree between the first cause information and the attention layer. The ratio between the first intersection region and the union region is the overlap parameter of the attention hotspot.

[0074] For example, the attention heat zone overlap parameter refers to the overlap between the explanatory entity and the attention heat zone of the decision model, where the entity in the first cause information... The image region mask is: Attention heatmap of decision-making model Select threshold Obtain attention hotspot mask: .based on and Computational Entity Attention hotspot overlap parameters: Furthermore, when the primary cause information contains N entities (the primary object), the attention hotspot overlap parameter... It can be represented by N entities, and the calculation formula is: .

[0075] In another example, the heatmap corresponding to the decision layer in the decision model is determined, and the second intersection region between the heatmap and the image region is determined. Based on the ratio between the intersection region and the image region, the matching degree between the first cause information and the attention layer is determined. The ratio between the intersection region and the image region is the overlap parameter of the decision heatmap, that is, the overlap parameter of the decision heatmap is used as the matching degree.

[0076] For example, the decision heatmap overlap parameter refers to the overlap between the image region where the entity is located in the first cause information and the decision layer heatmap in the decision model. The decision heatmap of the decision model is obtained according to Grad-CAM (Gradient-weighted Class Activation Mapping). Soldiers select threshold Obtain the decision-making level heat map mask: .based on and Computational Entity Matching rate between decision heatmap and the image region where the entity is located Furthermore, when the primary cause information contains N entities (the primary object), the decision hotspot overlap parameter... It is represented by N entities, and the calculation formula is: .

[0077] In one example, the first text vector corresponding to the first cause information and the image vector corresponding to the image region are determined. Based on the first similarity between the first text vector and the image vector, the matching degree between the first cause information and the attention layer is determined. The first similarity is the modality alignment strength parameter, that is, the modality alignment strength parameter is used as the matching degree.

[0078] For example, the modal alignment strength parameter refers to the degree of alignment between the text fragment in the first cause information and the relevant region in the image from a cross-modal alignment perspective, and is usually calculated using a cross-modal alignment model. The encoded vector of the text in the first cause information is obtained through a text encoder using a cross-modal alignment model. (First text vector), the encoded vector of the image region is obtained through the image encoder of the cross-modal alignment model. (Image vectors), calculate the cosine similarity between the two. Furthermore, if the first cause information contains N entities (the first object), then the multimodal alignment strength... Represented by N entities, the calculation formula is: .

[0079] In this embodiment, the explanation information for attention alignment is accurately determined by the image region where the first object mentioned in the first cause information is located.

[0080] Figure 4 A flowchart of a driving behavior analysis method provided in this application embodiment Figure 4 .based on Figures 1 to 3 In any of the embodiments shown, step S103 includes: Step S401: Obtain target parameters, which include at least one of model stability parameters and time consistency parameters.

[0081] In this embodiment, the first explanatory information includes explanatory information on model stability and robustness. The explanatory information on model stability and robustness is composed of at least one of model stability parameters and time consistency parameters. To this end, the device acquires target parameters, which include at least one of model stability parameters and time consistency parameters.

[0082] In one example, the target parameter is the model stability parameter. The device modifies the parameters in the target data by a target magnitude to obtain third data. The target magnitude is smaller than a preset magnitude, that is, the target data is slightly perturbed to obtain the third data. The third data is input into the decision model to obtain the third cause information corresponding to the second driving behavior output by the decision model. Then, the second text vector corresponding to the first cause information and the third text vector corresponding to the third cause information are determined. The model stability parameter is determined based on the second similarity between the second text vector and the third text vector.

[0083] For example, by analyzing the original input Obtain by making small perturbations The model explanation outputs were obtained respectively. and The text encoder of the CLIP (Contrastive Language-Image Pre-training) model was used to obtain the encoded vectors of the characters in the first cause information. The encoding vector of characters in the third reason information Calculate the consistency between two vectors using cosine similarity. The closer the value is to 1, the better the consistency of the model input perturbation, and the more stable the model. The formula for calculating the consistency of the model input perturbation is: .

[0084] In another example, the target parameters include time consistency parameters. The device acquires historical cause information corresponding to the driving behavior output by the decision model at each historical moment, where each historical moment is temporally continuous with the current moment; it determines the historical text vector corresponding to each historical cause information, and determines the third similarity between the historical text vectors corresponding to the historical moments of the vectors, thereby determining the time consistency parameters through each third similarity.

[0085] For example, the time consistency parameter is calculated from a short-term (e.g., 1 second at 10 Hz) continuous test set of the scene, and continuous frames (continuous frames represent consecutive moments) are obtained through the model. Reason information The interpreted encoded vectors are obtained from the text encoder of the CLIP model. Calculate the cosine similarity between adjacent cause information. Finally, the time consistency parameters of this data segment are obtained, and the time consistency parameters are characterized as follows: .

[0086] Step S402: Determine the interpretation information of model stability and robustness based on the target parameters.

[0087] Once the target parameters are obtained, they can be used as explanatory information for the model's stability and robustness.

[0088] It should be noted that when the target parameters include model stability parameters and time consistency parameters, the mean of the model stability parameters and time consistency parameters is used as the explanatory information for model stability and robustness. The explanatory information for model stability and robustness is represented as follows: .

[0089] In this embodiment, the explanatory information of model stability and robustness is accurately determined by using model stability parameters and time consistency parameters.

[0090] Figure 5 A flowchart of a driving behavior analysis method provided in this application embodiment Figure 5 .based on Figures 1 to 4 In any of the embodiments shown, step S103 includes: Step S501: Based on the target data, determine the fourth objects that need to be considered when driving the vehicle and the first risk factors that exist when driving the vehicle.

[0091] In this embodiment, the first explanatory information includes assessment information of the model's risk perception. The assessment information of the model's risk perception can be determined through safety recall rate and risk omission rate.

[0092] The device first determines, based on the target data, the various fourth objects requiring attention for vehicle operation and the various first risk factors present for vehicle operation. For example, if the target data image contains 4 pedestrians, a traffic light, and 3 detected vehicles, then the various fourth objects requiring attention for vehicle operation are the 4 pedestrians, the traffic light, and the 3 detected vehicles; the various first risk factors determined based on the target data are: a pedestrian suddenly rushing out, a traffic light being red, a traffic light being green but with only 5 seconds remaining, and other vehicles turning without using their turn signals.

[0093] Step S502: Determine the first objects and second risk factors of interest to the decision model based on the first cause information.

[0094] The decision model outputs first-cause information. Within this first-cause information, the device identifies the primary objects and secondary risk factors that the decision model is interested in. Each primary object is the entity mentioned in the first-cause information. For example, if the first-cause information is "There are two vehicles turning ahead, four pedestrians crossing the crosswalk, and the traffic light is red; please stop before the crosswalk," then the two vehicles, four pedestrians, and the red light are all primary objects. The secondary risk factors are: two vehicles turning, four pedestrians crossing the crosswalk, and the traffic light being red.

[0095] Step S503: Determine the safety recall rate based on the ratio between the number of the first object and the number of the fourth object, and determine the risk omission rate based on each first risk factor and each second risk factor.

[0096] The device determines the ratio between the number of the first object and the number of the fourth object as the safety recall rate, which is characterized as: .

[0097] in, It refers to the collection of each first object. It refers to the collection of various fourth objects. For safety recall rate.

[0098] The device then determines the risk omission rate using each first risk factor and each second risk factor. For example, if each first risk factor and each second risk factor are the same, it indicates that the decision model has not omitted any risk factors, and therefore, the risk omission rate is 1. If at least one of the second risk factors is not any of the first risk factors, and / or the number of second risk factors is less than the number of first risk factors, then it can be determined that the decision model has omitted risk factors, and the risk omission rate is 0. The risk omission rate is characterized as follows: .

[0099] in, This represents the risk omission rate.

[0100] Step S504: Determine the assessment information of the model's risk perception based on the safety recall rate and the risk omission rate.

[0101] After obtaining the safety recall rate and the risk rejection rate, the evaluation information of the model's risk perception can be determined based on these two rates. The evaluation information of the model's risk perception is represented as follows: .

[0102] It is understandable that the assessment information of the model's risk perception is actually the average of the safety recall rate and the risk loss rate.

[0103] In this embodiment, the assessment information of the model's risk perception is accurately determined by using the safety recall rate and the risk omission rate.

[0104] Figure 6 A flowchart of a driving behavior analysis method provided in this application embodiment Figure 6 .based on Figures 1 to 5 In any of the embodiments shown, step S103 includes: Step S601: Obtain multiple target information, including at least two of the following: explanatory information on behavioral consistency, explanatory information on attention alignment, explanatory information on model stability and robustness, and evaluation information on model risk perception.

[0105] In this embodiment, the first explanatory information includes descriptive information about the model's safety performance. This descriptive information about the model's safety performance is determined using at least two of the following: consistent explanatory information, attention alignment explanatory information, model stability and robustness explanatory information, and model risk perception assessment information.

[0106] In response, the device acquires multiple target information, including at least two of the following: explanatory information on behavioral consistency, explanatory information on attention alignment, explanatory information on model stability and robustness, and assessment information on model risk perception.

[0107] The process for determining explanatory information on behavioral consistency, attention alignment, model stability and robustness, and assessment information on model risk perception is detailed in the above description and will not be repeated here.

[0108] Step S602: Determine the vehicle's current first driving scenario based on the target data, and determine the first safety performance parameters of the decision model under the first driving scenario based on the first driving scenario and various target information.

[0109] The device determines the current driving scenario of the vehicle based on the target data. This driving scenario is defined as the first driving scenario. Based on the first driving scenario and various target information, the device determines the safety performance parameters of the decision model under the first driving scenario. These safety performance parameters are defined as the first safety performance parameters.

[0110] For example, based on the first driving scenario, a corresponding weight is configured for each target information, and a weighted calculation is performed based on each target information and its corresponding weight to obtain the first safety performance parameter.

[0111] Step S603: Obtain the second safety performance parameters of the decision model in the second driving scenario, and determine the descriptive information of the model's safety performance based on the first and second safety performance parameters. The first driving scenario is different from the second driving scenario.

[0112] The device acquires the second safety performance parameters of the decision model in the second driving scenario. The second driving scenario is different from the first driving scenario, and the second safety performance parameters are the performance parameters previously calculated for the vehicle.

[0113] After obtaining the first safety performance parameter and the second safety performance parameter, the descriptive information of the model's safety performance is determined based on the first safety performance parameter and the second safety performance parameter.

[0114] For example, the descriptive information of the model's safety performance is characterized as follows:

[0115] in, This refers to the weighting in typical driving scenarios; This refers to explanatory information on behavioral consistency in typical driving scenarios, explanatory information on attention alignment, explanatory information on model stability and robustness, and assessment information on model risk perception. This refers to the weighting in long-tail driving scenarios. This refers to explanatory information on behavioral consistency in typical driving scenarios, explanatory information on attention alignment, explanatory information on model stability and robustness, and assessment information on model risk perception.

[0116] In this embodiment, the descriptive information of the decision-making model's performance is accurately determined by using explanatory information from different driving scenarios.

[0117] Corresponding to the driving behavior analysis method described above, this application also provides a driving behavior analysis device. Figure 7 This is a schematic diagram of a driving behavior analysis device provided in an embodiment of this application. The driving behavior analysis device 700 provided in this embodiment includes: The acquisition module 710 is used to acquire target data for vehicle detection. The input module 720 is used to input target data into the decision model to obtain the target driving behavior to be performed by the vehicle and the first cause information output by the decision model. The first cause information is used to describe the reason why the decision model determines the target driving behavior. Analysis module 730 is used to perform safety analysis on target driving behavior and obtain at least one first explanation information of target driving behavior, the first explanation information being used to describe the safety of the vehicle performing the target driving behavior; The output module 740 is used to output the target driving behavior, the first cause information, and at least one first explanation information.

[0118] In some implementations, the driving behavior analysis device 700 is also used for: The first object in the target data is processed to obtain the second data containing the second object. The first object is the object mentioned in the first reason information, and the second object is the first object that is occluded, the first object after the behavior is modified, or the preset object that replaces the first object. The second data is input into the decision model to obtain the first driving behavior and the second cause information corresponding to the first driving behavior. Based on the first object, the third object mentioned in the second cause information, the first driving behavior, and the target driving behavior, determine the explanatory information for behavioral consistency.

[0119] In some implementations, the driving behavior analysis device 700 is also used for: Based on the target data, determine the image region where the first object mentioned in the first cause information is located; Based on the image region, the matching degree between the primary cause information and the attention layer in the decision model is determined as explanatory information for attention alignment.

[0120] In some implementations, the driving behavior analysis device 700 is also used for: Identify the attention hotspots corresponding to the attention layers in the decision-making model; Determine the first intersection region and the union region between the attention hotspot and the image region; The matching degree between the first cause information and the attention layer is determined based on the ratio between the first intersection region and the union region.

[0121] In some implementations, the driving behavior analysis device 700 is also used for: Determine the heatmap corresponding to the decision-making level in the decision-making model; Determine the second intersection region between the heatmap and the image region; The matching degree between the first cause information and the attention layer is determined based on the ratio between the second intersection region and the image region.

[0122] In some implementations, the driving behavior analysis device 700 is also used for: Determine the first text vector corresponding to the first cause information and the image vector corresponding to the image region; The matching degree between the first cause information and the attention layer is determined based on the first similarity between the first text vector and the image vector.

[0123] In some implementations, the driving behavior analysis device 700 is also used for: Obtain the target parameters, which include at least one of the model stability parameters and time consistency parameters; Based on the target parameters, determine the explanatory information for the model's stability and robustness.

[0124] In some implementations, the driving behavior analysis device 700 is also used for: The parameters in the target data are modified to a target amplitude to obtain the third data, where the target amplitude is less than the preset amplitude; Input the third data into the decision model to obtain the third cause information corresponding to the second driving behavior output by the decision model; Determine the second text vector corresponding to the first cause information and the third text vector corresponding to the third cause information; The model stability parameters are determined based on the second similarity between the second and third text vectors.

[0125] In some implementations, the driving behavior analysis device 700 is also used for: Obtain the historical cause information corresponding to the driving behavior output by the decision model at each historical moment, and each historical moment is continuous in time with the current moment; Determine the historical text vector corresponding to each historical cause information, and determine the third similarity between the historical text vectors corresponding to adjacent historical moments; Based on each third similarity, determine the time consistency parameter.

[0126] In some implementations, the driving behavior analysis device 700 is also used for: Based on the target data, identify the fourth objects that need to be considered when driving the vehicle and the first risk factors that exist when driving the vehicle. Based on the first cause information, determine the primary objects and secondary risk factors that the decision-making model focuses on; The safety recall rate is determined based on the ratio between the number of the first object and the number of the fourth object, and the risk omission rate is determined based on each first risk factor and each second risk factor. Based on the safety recall rate and risk omission rate, the assessment information for the model's risk perception is determined.

[0127] In some implementations, the driving behavior analysis device 700 is also used for: Acquire multiple target information, including at least two of the following: explanatory information on behavioral consistency, explanatory information on attention alignment, explanatory information on model stability and robustness, and evaluation information on model risk perception. Based on the target data, determine the vehicle's current first driving scenario, and based on the first driving scenario and various target information, determine the first safety performance parameters of the decision model under the first driving scenario; The decision model obtains the second safety performance parameters in the second driving scenario, and determines the descriptive information of the model's safety performance based on the first and second safety performance parameters. The first and second driving scenarios are different.

[0128] The driving behavior analysis device and the driving behavior analysis method provided in the above embodiments of this application belong to the same application concept and can execute the driving behavior analysis method provided in any of the above embodiments of this application. They have the corresponding functional modules and beneficial effects of executing the driving behavior analysis method. Technical details not described in detail in this embodiment can be found in the specific processing content of the driving behavior analysis method provided in the above embodiments of this application, and will not be repeated here.

[0129] The functions implemented by each module in the driving behavior analysis device can be implemented by the same or different processors, and this application embodiment does not limit this.

[0130] It should be understood that the modules in the above driving behavior analysis device can be implemented in the form of processor calling firmware. For example, the system includes a processor connected to a memory containing instructions. The processor calls the instructions stored in the memory to implement any of the above methods or to implement the functions of each module of the device. The processor can be a general-purpose processor, such as a CPU or microprocessor, and the memory can be internal to the device or external to the system. Alternatively, the modules in the system can be implemented in the form of hardware circuits. By designing the hardware circuits, some or all of the module functions can be implemented. The hardware circuits can be understood as one or more processors. For example, in one implementation, the hardware circuit is an ASIC, and the functions of some or all of the above modules are implemented by designing the logical relationships of the components within the circuit. In another implementation, the hardware circuit can be implemented using a PLD, such as an FPGA, which can include a large number of logic gates. The connection relationships between the logic gates are configured through a configuration file, thereby implementing the functions of some or all of the above modules. All modules of the above driving behavior analysis device can be implemented entirely by processor calling firmware, entirely by hardware circuits, or partially by processor calling firmware with the remaining parts implemented by hardware circuits.

[0131] In this application embodiment, a processor is a circuit with signal processing capabilities. In one implementation, the processor can be a circuit with instruction reading and execution capabilities, such as a CPU, microprocessor, GPU, or DSP. In another implementation, the processor can implement certain functions through the logical relationships of hardware circuits. These logical relationships are fixed or reconfigurable. For example, the processor may be a hardware circuit implemented as an ASIC or PLD, such as an FPGA. In a reconfigurable hardware circuit, the process of the processor loading a configuration document and configuring the hardware circuit can be understood as the processor loading instructions to implement the functions of some or all of the above modules. Furthermore, it can also be a hardware circuit designed for artificial intelligence, which can be understood as an ASIC, such as an NPU, TPU, or DPU.

[0132] As can be seen, each module in the above driving behavior analysis device can be one or more processors (or processing circuits) configured to implement the above methods, such as: CPU, GPU, NPU, TPU, DPU, microprocessor, DSP, ASIC, FPGA, or a combination of at least two of these processor types.

[0133] Furthermore, the modules in the above-mentioned driving behavior analysis device can be integrated in whole or in part, or they can be implemented independently. In one implementation, these modules are integrated together and implemented in the form of a System-on-a-Chip (SoC). The SoC may include at least one processor for implementing any of the above methods or implementing the functions of the modules of the device. The at least one processor can be of different types, such as CPU and FPGA, CPU and artificial intelligence processor, CPU and GPU, etc.

[0134] This application provides a structural schematic diagram of a vehicle, see [link / reference] Figure 8 As shown, the vehicle includes a memory 800 and a processor 810; wherein the memory 800 is connected to the processor 810 and is used to store programs; the processor 810 is used to implement the driving behavior analysis method disclosed in any of the above embodiments by running the programs stored in the memory 800.

[0135] Specifically, the vehicle may also include: a bus, a communication interface 820, an input device 830, and an output device 840. The vehicle may also include a data transceiver module, an image monitoring module, and a signal monitoring module.

[0136] The processor 810, memory 800, communication interface 820, input device 830, and output device 840 are interconnected via a bus. Among them: A bus can include a pathway for transmitting information between various components in a vehicle.

[0137] The processor 810 can be a general-purpose processor, such as a general-purpose central processing unit (CPU), a microprocessor, etc., or an application-specific integrated circuit (ASIC), or one or more integrated circuits used to control the execution of the program of the present invention. It can also be a digital signal processor (DSP), an application-specific integrated circuit (ASIC), an off-the-shelf programmable gate array (FPGA), or other programmable logic devices, discrete gate or transistor logic devices, or discrete hardware components.

[0138] The processor 810 may include a main processor, as well as a baseband chip, modem, etc.

[0139] The memory 800 stores a program that executes the technical solution of this invention, and may also store an operating system and other key business functions. Specifically, the program may include program code, which includes computer operation instructions. More specifically, the memory 800 may include read-only memory (ROM), other types of static storage devices capable of storing static information and instructions, random access memory (RAM), other types of dynamic storage devices capable of storing information and instructions, disk storage, flash memory, etc.

[0140] Input device 830 may include a device for receiving user input data and information, such as a keyboard, mouse, camera, scanner, light pen, voice input device, touch screen, pedometer, or gravity sensor.

[0141] Output device 840 may include devices that allow information to be output to a user, such as a display screen, printer, speaker, etc.

[0142] The communication interface 820 may include a device that uses any transceiver to communicate with other devices or communication networks, such as Ethernet, Radio Access Network (RAN), Wireless Local Area Network (WLAN), etc.

[0143] The processor 810 executes the program stored in the memory 800 and calls other devices, which can be used to implement each step of any of the driving behavior analysis methods provided in the above embodiments of this application.

[0144] It should be noted that the vehicle can be an in-vehicle terminal, mobile phone, wearable device or server, etc.; or it can be a vehicle that includes an in-vehicle terminal, etc.

[0145] This application also proposes a chip, which includes a processor and a data interface. The processor reads and runs a program stored in the memory through the data interface to execute the driving behavior analysis method described in any of the above embodiments. For the specific processing procedure and its beneficial effects, please refer to the embodiments of the driving behavior analysis method described above.

[0146] In addition to the methods and devices described above, embodiments of this application may also be computer program products, which include computer program instructions that, when executed by a processor, cause the processor to perform the steps in the driving behavior analysis methods according to various embodiments of this application as described in any of the foregoing embodiments of this specification.

[0147] Computer program products can be written in any combination of one or more programming languages ​​to perform the operations of the embodiments of this application. The programming languages ​​include object-oriented programming languages ​​such as Java and C++, as well as conventional procedural programming languages ​​such as C or similar languages. The program code can be executed entirely on the user's computing device, partially on the power device, as a standalone firmware package, partially on the user's computing device and partially on a remote computing device, or entirely on a remote computing device or server.

[0148] Furthermore, embodiments of this application may also be storage media storing a computer program, which is executed by a processor to perform the steps of the driving behavior analysis method according to various embodiments of this application described in any of the above embodiments of this specification, specifically implementing the steps of the driving behavior analysis method as described above.

[0149] For the foregoing method embodiments, in order to simplify the description, they are all described as a series of actions. However, those skilled in the art should understand that this application is not limited to the described order of actions, because according to this application, some steps can be performed in other orders or simultaneously. Furthermore, those skilled in the art should also understand that the embodiments described in the specification are all preferred embodiments, and the actions and modules involved are not necessarily essential to this application.

[0150] It should be noted that the various embodiments in this specification are described in a progressive manner, with each embodiment focusing on the differences from other embodiments. Similar or identical parts between embodiments can be referred to interchangeably. For apparatus embodiments, since they are basically similar to method embodiments, the description is relatively simple; relevant parts can be referred to the descriptions in the method embodiments.

[0151] The steps in the methods of the various embodiments of this application can be adjusted, merged, or deleted in order according to actual needs, and the technical features described in each embodiment can be replaced or combined.

[0152] The units of the apparatus in the various embodiments of this application can be merged, divided, and deleted according to actual needs.

[0153] It should be understood that the disclosed terminals, devices, and methods can be implemented in other ways, given the several embodiments provided in this application. For example, the terminal embodiments described above are merely illustrative. For instance, the division of modules or sub-modules is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple sub-modules or modules may be combined or integrated into another module, or some features may be ignored or not executed. Furthermore, the coupling or direct coupling or communication connection shown or discussed may be indirect coupling or communication connection through some interfaces, devices, or modules, and may be electrical, mechanical, or other forms.

[0154] The modules or submodules described as separate components may or may not be physically separate. The components that constitute a module or submodule may or may not be physical modules or submodules; that is, they may be located in one place or distributed across multiple network modules or submodules. Some or all of the modules or submodules can be selected to achieve the purpose of this embodiment's solution, depending on actual needs.

[0155] Furthermore, the functional modules or sub-modules in the various embodiments of this application can be integrated into one processing module, or each module or sub-module can exist physically separately, or two or more modules or sub-modules can be integrated into one module. The integrated modules or sub-modules described above can be implemented in hardware or as firmware functional modules or sub-modules.

[0156] Those skilled in the art will further recognize that the units and algorithm steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, computer firmware, or a combination of both. To clearly illustrate the interchangeability of hardware and firmware, the components and steps of the various examples have been generally described in terms of functionality in the foregoing description. Whether these functions are implemented in hardware or firmware depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of this application.

[0157] The steps of the methods or algorithms described in conjunction with the embodiments disclosed herein can be implemented directly using hardware, firmware units executed by a processor, or a combination of both. The firmware unit can be located in random access memory (RAM), main memory, read-only memory (ROM), electrically programmable ROM, electrically erasable programmable ROM, registers, hard disk, removable disk, CD-ROM, or any other form of storage medium known in the art.

[0158] Finally, it should be noted that in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitations, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes the element.

[0159] The above description of the disclosed embodiments enables those skilled in the art to make or use this application. Various modifications to these embodiments will be readily apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of this application. Therefore, this application is not to be limited to the embodiments shown herein, but is to be accorded the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. A method for analyzing driving behavior, characterized in that, include: Acquire target data for vehicle inspection; The target data is input into the decision model to obtain the target driving behavior to be performed by the vehicle and the first reason information output by the decision model. The first reason information is used to describe the reason why the decision model determines the target driving behavior. A safety analysis is performed on the target driving behavior to obtain at least one first explanatory information of the target driving behavior, wherein the first explanatory information is used to describe the safety of the vehicle performing the target driving behavior; Output the target driving behavior, the first cause information, and the at least one first explanation information.

2. The driving behavior analysis method according to claim 1, characterized in that, The first explanatory information includes explanatory information on behavioral consistency, and the step of performing a safety analysis on the target driving behavior includes: The first object in the target data is processed to obtain the second data containing the second object, wherein the first object is the object mentioned in the first reason information, and the second object is the first object that is occluded, the first object after the behavior is modified, or a preset object that replaces the first object. The second data is input into the decision model to obtain the first driving behavior and the second cause information corresponding to the first driving behavior; Based on the first object, the third object mentioned in the second reason information, the first driving behavior, and the target driving behavior, determine the explanation information for behavioral consistency.

3. The driving behavior analysis method according to claim 1, characterized in that, The first explanatory information includes explanatory information on attention alignment, and the step of performing safety analysis on the target driving behavior includes: Based on the target data, determine the image region where the first object mentioned in the first reason information is located; Based on the image region, the matching degree between the first cause information and the attention layer in the decision model is determined as explanatory information for attention alignment.

4. The driving behavior analysis method according to claim 3, characterized in that, Determining the matching degree between the first cause information and the attention layer in the decision model based on the image region includes: Determine the attention hotspots corresponding to the attention layers in the decision model; Determine the first intersection region and the union region between the attention hotspot and the image region; The matching degree between the first cause information and the attention layer is determined based on the ratio between the first intersection region and the union region.

5. The driving behavior analysis method according to claim 3, characterized in that, Determining the matching degree between the first cause information and the attention layer in the decision model based on the image region includes: Determine the heatmap corresponding to the decision layer in the decision model; Determine the second intersection region between the heatmap and the image region; The matching degree between the first cause information and the attention layer is determined based on the ratio between the second intersection region and the image region.

6. The driving behavior analysis method according to claim 3, characterized in that, Determining the matching degree between the first cause information and the attention layer in the decision model based on the image region includes: Determine the first text vector corresponding to the first cause information and the image vector corresponding to the image region; The matching degree between the first cause information and the attention layer is determined based on the first similarity between the first text vector and the image vector.

7. The driving behavior analysis method according to claim 1, characterized in that, The first explanatory information includes explanatory information on model stability and robustness, and the safety analysis of the target driving behavior includes: Obtain target parameters, which include at least one of model stability parameters and time consistency parameters; Based on the target parameters, determine the explanatory information for model stability and robustness.

8. The driving behavior analysis method according to claim 7, characterized in that, The target parameters include model stability parameters, and obtaining the target parameters includes: The parameters in the target data are modified by a target amplitude to obtain third data, wherein the target amplitude is less than a preset amplitude; The third data is input into the decision model to obtain the third cause information corresponding to the second driving behavior output by the decision model; Determine the second text vector corresponding to the first reason information and the third text vector corresponding to the third reason information; The model stability parameter is determined based on the second similarity between the second text vector and the third text vector.

9. The driving behavior analysis method according to claim 7, characterized in that, The target parameters include time consistency parameters, and obtaining the target parameters based on the target data includes: Obtain the historical cause information corresponding to the driving behavior output by the decision model at each historical moment, wherein each historical moment is continuous in time with the current moment; Determine the historical text vector corresponding to each of the historical cause information, and determine the third similarity between the historical text vectors corresponding to adjacent historical moments; Based on each of the aforementioned third similarities, the time consistency parameters are determined.

10. The driving behavior analysis method according to claim 1, characterized in that, The first explanatory information includes assessment information on model risk perception, and the safety analysis of the target driving behavior includes: Based on the target data, determine the fourth objects that need to be considered when driving the vehicle and the first risk factors that exist when driving the vehicle. Based on the first cause information, determine the first objects and second risk factors that the decision model focuses on; The safety recall rate is determined based on the ratio between the number of the first objects and the number of the fourth objects, and the risk omission rate is determined based on each of the first risk factors and each of the second risk factors. The risk perception assessment information of the model is determined based on the safety recall rate and the risk omission rate.

11. The driving behavior analysis method according to claim 1, characterized in that, The first explanatory information includes descriptive information about the model's safety performance, and the step of performing safety analysis on the target driving behavior includes: Acquire multiple target information, including at least two of the following: explanatory information on behavioral consistency, explanatory information on attention alignment, explanatory information on model stability and robustness, and evaluation information on model risk perception. The first driving scenario of the vehicle is determined based on the target data, and the first safety performance parameter of the decision model in the first driving scenario is determined based on the first driving scenario and each of the target information. The decision model is obtained in the second driving scenario, and the safety performance description information of the model is determined based on the first safety performance parameter and the second safety performance parameter. The first driving scenario is different from the second driving scenario.

12. A driving behavior analysis device, characterized in that, Including memory and processor, among which, The memory is connected to the processor and is used to store programs; The processor is used to implement the driving behavior analysis method as described in any one of claims 1-11 by running the program in the memory.

13. A vehicle, characterized in that, The vehicle includes a driving behavior analysis device, which implements the driving behavior analysis method as described in any one of claims 1-11.

14. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a computer program, which, when executed by a processor, implements the driving behavior analysis method as described in any one of claims 1-11.

15. A computer program product, characterized in that, When the computer program is executed by the processor, it implements the driving behavior analysis method as described in any one of claims 1-11.

Images