Auxiliary driving vehicle accident scene extraction method considering scene complexity

Abstract

The invention discloses an auxiliary driving vehicle accident scene extraction method considering scene complexity. The method comprises the following steps: setting accident scene information of an auxiliary driving vehicle accident scene; the accident scene information comprises scene complexity information, a scene index category corresponding to a typical scene category and feature information of the typical scene; the non-parametric feature information is subjected to one-hot coding, and the parametric feature information is kept in an original numerical form; performing clustering analysis on the feature information corresponding to the key index category of each accident through a Gaussian mixture model to obtain the complexity of three types of scenes; dividing the feature information corresponding to the scene index category corresponding to the complexity of each category of scene into s categories of typical scenes by using a DBSCAN clustering method; feature information of the typical scenes is given to each type of typical scenes; accident scenes of the auxiliary driving vehicle are classified by considering scene complexity, and a scientific basis is provided for improvement of an auxiliary driving system and accident prevention.

Description

Technical Field

[0001] The present invention relates to the field of accident scene extraction methods, and in particular to the field of accident scene extraction methods for assisted driving vehicles that take scene complexity into consideration. Background Art

[0002] In today's production and life, assisted driving has gradually become part of our daily lives. In order to make driving safer, it is necessary to classify the accidents that have occurred, so as to provide a scientific basis for the improvement of assisted driving systems and accident prevention.

[0003] Therefore, how to invent a method for extracting accident scenes of assisted driving vehicles to classify accident scenes of assisted driving vehicles taking into account the complexity of the scenes has become a difficult problem that needs to be solved urgently in this field. Summary of the invention

[0004] In order to solve the above technical problems, the present invention provides an assisted driving vehicle accident scene extraction method taking into account the scene complexity, so as to classify the assisted driving vehicle accident scenes taking into account the scene complexity. Each type of accident scene information includes: scene complexity information, scene indicator category corresponding to the typical scene category, and characteristic information of the typical scene.

[0005] According to the present invention, a method for extracting accident scenes of assisted driving vehicles taking into account scene complexity is provided, comprising the following steps:

[0006] Setting accident scenario information for an assisted driving vehicle accident scenario;

[0007] The accident scene information includes: scene complexity information, scene indicator categories corresponding to typical scene categories, and characteristic information of typical scenes;

[0008] Collect historical accident data of assisted driving vehicles to obtain an assisted driving vehicle accident data set;

[0009] Preprocess the assisted driving vehicle accident data set to obtain key indicator categories and their corresponding feature information;

[0010] The feature information includes parametric feature information and non-parametric feature information; the non-parametric feature information is one-hot encoded, and the parametric feature information is kept in the original numerical form;

[0011] The characteristic information corresponding to the key indicator categories of each accident was clustered and analyzed using the Gaussian mixture model to obtain three categories of scene complexity;

[0012] The complexity is scored according to the complexity scoring method, and the complexity of the three types of scenarios are marked as simple scenarios, general scenarios, and complex scenarios;

[0013] The characteristic information corresponding to the scene indicator category corresponding to the complexity of each type of scene is divided into s types of typical scenes using the DBSCAN clustering method; the characteristic information of the typical scene is assigned to each type of typical scene.

[0014] Compared with the prior art, the present invention has the following beneficial effects: by taking into account the complexity of the scene, the assisted driving vehicle accident scene extraction method is used to classify the assisted driving vehicle accident scene by taking into account the complexity of the vehicle accident, and at the same time, different categories of vehicle accident scenes include: scene complexity information, scene indicator categories corresponding to typical scene categories, and characteristic information of typical scenes;

[0015] By preprocessing the assisted driving vehicle accident data set, key indicator categories and their corresponding feature information are obtained, making subsequent classification closer to reality;

[0016] By processing the feature information including parametric feature information and non-parametric feature information, the non-parametric feature information is uniquely encoded, and the parametric feature information is kept in the original numerical form, so that the information reflecting the complexity can be digitized, analyzed, and classified;

[0017] By clustering the characteristic information corresponding to the key indicator categories of each accident through the Gaussian mixture model, the complexity of three categories of scenes is obtained, and the accident scenes are divided into three categories according to their complexity;

[0018] By scoring the complexity according to the complexity scoring method, a more accurate labeling of the complexity of this type of scene is achieved, which is close to the actual situation;

[0019] The characteristic information corresponding to the scene indicator category corresponding to the complexity of each type of scene is divided into s typical scenes using the DBSCAN clustering method; the characteristic information of the typical scene is assigned to each typical scene, so that the accident scenes of different complexity categories are further classified into several categories, and at the same time, the characteristic information of the typical scene is assigned to each category; thereby providing a scientific basis for the improvement of assisted driving systems and accident prevention.

[0020] Furthermore, preprocessing the assisted driving vehicle accident data set includes the following steps:

[0021] Remove outliers, fill in missing values, extract key indicator categories and feature information included in key indicator categories;

[0022] Removing outliers includes determining outliers far from the mean through z-score standardization;

[0023] Filling missing values ​​includes using the majority to fill missing values ​​in categorical data and using interpolation to fill missing values ​​in numerical data.

[0024] Furthermore, the key indicator categories include:

[0025] Road type, road surface conditions, ambient lighting conditions, weather conditions, vehicle speed at the time of collision, driver behavior, enabled driver assistance features, vehicle type, collision party type, collision type, and relative direction of motion.

[0026] Furthermore, the non-parametric characteristic information includes characteristic information of road type, characteristic information of ambient lighting conditions, characteristic information of weather conditions, characteristic information of driver behavior, characteristic information of enabled auxiliary driving functions, characteristic information of vehicle type, characteristic information of collision party type, characteristic information of collision type, and characteristic information of relative motion direction;

[0027] The parameter-based characteristic information includes characteristic information of the vehicle's speed at the moment of collision.

[0028] Furthermore, the characteristic information of the road type includes: one of a tunnel, a sharp curve, a bridge, an intersection, a sidewalk, and an ordinary road section;

[0029] The characteristic information of the road condition includes: one of dry, wet, waterlogged, icy, and snowy;

[0030] The characteristic information of the ambient lighting conditions includes: one of daytime, nighttime with street lights, and nighttime without street lights;

[0031] The characteristic information of the weather condition includes: one of sunny day, rainy day, snowy day and foggy day;

[0032] The characteristic information of the driver's behavior includes one or more of normal driving, distraction, fatigue driving, speeding, and sudden braking;

[0033] The characteristic information of the enabled assisted driving functions includes: one or more of FCW, AEB, ACC, and LDW;

[0034] The characteristic information of the vehicle type includes: one of a sedan, a bus, a truck, and other types of vehicles;

[0035] The characteristic information of the collision party type includes: pedestrian, bus, car, truck, fixed object, or one of other collision party types;

[0036] The characteristic information of the collision type includes: hitting a person, scratching, hitting a fixed object, rollover, rear-end collision, frontal collision, and one of other collision types;

[0037] The characteristic information of the relative motion direction includes: one of the same direction and the side direction.

[0038] The beneficial effect of adopting the previous step is that when implementing the complexity classification, a more comprehensive consideration is given, and all relevant phenomena that affect the occurrence of accidents during the actual driving process are taken into account.

[0039] Furthermore, the complexity scoring method includes the following steps:

[0040] The feature information corresponding to the key indicator category is standardized using the minimum-maximum standardization method to ensure that all feature values ​​are within the range of [0,1];

[0041] The feature weights of each key indicator category are calculated using the entropy weight method;

[0042] By constructing a decision matrix, the complexity score of each accident sample is calculated based on the TOPS IS method;

[0043] The scene with the highest complexity score is marked as a simple scene, the scene with a middle complexity score is marked as a general scene, and the scene with the lowest complexity score is marked as a complex scene.

[0044] Furthermore, the standardization formula for standardizing the characteristic information corresponding to the key indicator category is:

[0045]

[0046] X is the original data of the feature information, min(X) and max(X) are the minimum and maximum values ​​of the feature information corresponding to the key indicator category, respectively. scaled It is the standardized data of characteristic information corresponding to the key indicator category.

[0047] Furthermore, the entropy weight method calculates the feature weights of each key indicator category, including the following steps:

[0048] The information entropy E of each key indicator category is calculated by formula 1 j ;

[0049] Formula 1:

[0050]

[0051] It represents the proportion of the standardized value of the characteristic information corresponding to the jth key indicator category of the i-th sample to the total characteristic information; n is the number of accident samples; E j is the information entropy of the jth key indicator category;

[0052] The characteristic weight of each key indicator category is calculated by formula 2;

[0053] Formula 2:

[0054]

[0055] w j is the feature weight of the jth key indicator category; m is the number of key indicator categories.

[0056] The beneficial effect of adopting the previous step is that when each accident sample is scored for complexity through the feature information of each key indicator category, the corresponding feature weights are combined for different key indicator categories, so that the complexity of calculation of each accident sample is close to the actual one.

[0057] Furthermore, the basic process of calculating the complexity score of each accident sample based on the TOPS IS method is:

[0058] Weighting of standardized data;

[0059]

[0060] X ij The standardized data of the characteristic information of the jth key indicator category of the i-th accident sample;

[0061] X i ' j The standardized data of the characteristic information of the jth key indicator category of the i-th accident sample after weighting;

[0062] n is a total of n accident samples;

[0063] The decision matrix D′ is constructed by using the standardized data of the characteristic information corresponding to the key indicator categories after weighting n accident samples;

[0064]

[0065] X nm The standardized data of the characteristic information of the mth key indicator category of the nth accident sample;

[0066]

[0067] For each ideal solution of the jth key indicator category, is the negative ideal solution of the jth key indicator category;

[0068]

[0069] C i Score the complexity of the i-th accident sample.

[0070] By adopting the beneficial effect of the previous step, each accident sample involved in the complexity classification is given a corresponding complexity score.

[0071] Furthermore, the complexity of each type of scene is divided into s typical scene categories using the DBSCAN clustering method; the scene indicator categories include road type, vehicle type, collision party type, relative motion direction, collision type, and enabled assisted driving functions;

[0072] The characteristic information corresponding to the scene indicator category of the accident sample in each scene complexity level is used as the data point;

[0073] Calculate the distance between each data point and other points and draw a k-distance graph. Select the best ε value through the "knee point" in the graph. The number of "knee points" in the graph is the number of typical scene categories s;

[0074] By calculating the average distance between each point and its first K nearest neighbors, the appropriate statistical 95% percentile point is selected to determine MinPts; for each pair of ε and MinPts combination, the DBSCAN clustering algorithm is applied to cluster and s clusters of data points are obtained, each cluster of data points corresponds to a typical scene; s typical scenes corresponding to the complexity of each scene are obtained;

[0075] and / or

[0076] The process of assigning characteristic information of typical scenes to each type of typical scene includes the following steps:

[0077] The local density estimation method is used to find the data point with the highest density in the cluster, which represents the typical scene corresponding to this cluster; the feature information corresponding to the scene indicator category of the highest data point in each cluster is the feature information of the corresponding typical scene.

[0078] By adopting the beneficial effects of the previous step, it is possible to assign characteristic information of typical scenes to each type of typical scenes, and to classify accident scenes of different complexity categories into several categories again, and at the same time assign characteristic information of typical scenes to each category; thereby providing a scientific basis for the improvement of assisted driving systems and accident prevention. DETAILED DESCRIPTION

[0079] In order to better understand the technical solution of the present invention, the present invention is further described below in conjunction with specific embodiments.

[0080] Embodiment 1:

[0081] According to this embodiment, a method for extracting accident scenes of assisted driving vehicles taking into account scene complexity is provided, which is characterized by comprising the following steps:

[0082] Setting accident scenario information for an assisted driving vehicle accident scenario;

[0083] The accident scene information includes: scene complexity information, scene indicator categories corresponding to typical scene categories, and characteristic information of typical scenes;

[0084] Collect historical accident data of assisted driving vehicles to obtain an assisted driving vehicle accident data set;

[0085] Preprocessing the assisted driving vehicle accident data set to obtain key indicator categories and their corresponding feature information; specifically including the following steps: removing outliers, filling missing values, extracting key indicator categories and feature information included in the key indicator categories;

[0086] Removing outliers includes determining outliers far from the mean through z-score standardization;

[0087] Filling missing values ​​includes using the majority to fill missing values ​​in categorical data and using interpolation to fill missing values ​​in numerical data.

[0088] The feature information includes parametric feature information and non-parametric feature information; the non-parametric feature information is one-hot encoded, and the parametric feature information is kept in the original numerical form;

[0089] The key indicator categories include: road type, road surface conditions, ambient lighting conditions, weather conditions, vehicle speed at the time of collision, driver behavior, enabled assisted driving functions, vehicle type, collision party type, collision type, and relative motion direction;

[0090] The non-parametric characteristic information includes characteristic information of road type, characteristic information of ambient lighting conditions, characteristic information of weather conditions, characteristic information of driver behavior, characteristic information of enabled auxiliary driving functions, characteristic information of vehicle type, characteristic information of collision party type, characteristic information of collision type, and characteristic information of relative motion direction;

[0091] The parameter-based characteristic information includes characteristic information of the vehicle's speed at the moment of collision.

[0092] The characteristic information of the road type includes: one of a tunnel, a sharp curve, a bridge, an intersection, a sidewalk, and an ordinary road section;

[0093] The characteristic information of the road condition includes: one of dry, wet, waterlogged, icy, and snowy;

[0094] The characteristic information of the ambient lighting conditions includes: one of daytime, nighttime with street lights, and nighttime without street lights;

[0095] The characteristic information of the weather condition includes: one of sunny day, rainy day, snowy day and foggy day;

[0096] The characteristic information of the driver's behavior includes one or more of normal driving, distraction, fatigue driving, speeding, and sudden braking;

[0097] The characteristic information of the enabled assisted driving functions includes: one or more of FCW, AEB, ACC, and LDW;

[0098] The characteristic information of the vehicle type includes: one of a sedan, a bus, a truck, and other types of vehicles;

[0099] The characteristic information of the collision party type includes: pedestrian, bus, car, truck, fixed object, or one of other collision party types;

[0100] The characteristic information of the collision type includes: hitting a person, scratching, hitting a fixed object, rollover, rear-end collision, frontal collision, and one of other collision types;

[0101] The characteristic information of the relative motion direction includes: one of the same direction and the side direction.

[0102] The characteristic information corresponding to the key indicator categories of each accident was clustered and analyzed using the Gaussian mixture model to obtain three categories of scene complexity;

[0103] The complexity is scored according to the complexity scoring method, and the complexity of the three types of scenarios are marked as simple scenarios, general scenarios, and complex scenarios;

[0104] The complexity scoring method includes the following steps:

[0105] The feature information corresponding to the key indicator category is standardized using the minimum-maximum standardization method to ensure that all feature values ​​are within the range of [0,1];

[0106] The standardized formula for standardizing the characteristic information corresponding to the key indicator category is:

[0107]

[0108] X is the original data of the feature information, min(X) and max(X) are the minimum and maximum values ​​of the feature information corresponding to the key indicator category, respectively. scaled It is the standardized data of characteristic information corresponding to the key indicator categories;

[0109] The entropy weight method is used to calculate the feature weights of each key indicator category, which specifically includes the following steps:

[0110] The information entropy E of each key indicator category is calculated by formula 1 j ;

[0111] Formula 1:

[0112]

[0113] It represents the proportion of the standardized value of the characteristic information corresponding to the jth key indicator category of the i-th sample to the total characteristic information; n is the number of accident samples; E j is the information entropy of the jth key indicator category;

[0114] The characteristic weight of each key indicator category is calculated by formula 2;

[0115] Formula 2:

[0116]

[0117] w j is the feature weight of the jth key indicator category; m is the number of key indicator categories.

[0118] By constructing a decision matrix, the complexity score of each accident sample is calculated based on the TOPSI S method. The specific process is as follows:

[0119] Weighting of standardized data;

[0120]

[0121] X ij The standardized data of the characteristic information of the jth key indicator category of the i-th accident sample;

[0122] X i ' j The standardized data of the characteristic information of the jth key indicator category of the i-th accident sample after weighting;

[0123] n is a total of n accident samples;

[0124] The decision matrix D′ is constructed by using the standardized data of the characteristic information corresponding to the key indicator categories after weighting n accident samples;

[0125]

[0126] X nm The standardized data of the characteristic information of the mth key indicator category of the nth accident sample;

[0127]

[0128] For each ideal solution of the jth key indicator category, is the negative ideal solution of the jth key indicator category;

[0129]

[0130] Ci Score the complexity of the i-th accident sample.

[0131] The scene with the highest complexity score is marked as a simple scene, the scene with a middle complexity score is marked as a general scene, and the scene with the lowest complexity score is marked as a complex scene.

[0132] The characteristic information corresponding to the scene indicator category corresponding to the complexity of each type of scene is divided into s types of typical scenes using the DBSCAN clustering method; the characteristic information of the typical scene is given to each type of typical scene;

[0133] , the complexity of each type of scene is divided into s typical scene categories using the DBSCAN clustering method; the scene indicator categories include road type, vehicle type, collision party type, relative motion direction, collision type, and enabled assisted driving functions;

[0134] The characteristic information corresponding to the scene indicator category of the accident sample in each scene complexity level is used as the data point;

[0135] Calculate the distance between each data point and other points and draw a k-distance graph. Select the best ε value through the "knee point" in the graph. The number of "knee points" in the graph is the number of typical scene categories s;

[0136] By calculating the average distance between each point and its first K nearest neighbors, the appropriate statistical 95% percentile point is selected to determine MinPts; for each pair of ε and MinPts combination, the DBSCAN clustering algorithm is applied to cluster and s clusters of data points are obtained, each cluster of data points corresponds to a typical scene; s typical scenes corresponding to the complexity of each scene are obtained;

[0137] The process of assigning characteristic information of typical scenes to each type of typical scene includes the following steps:

[0138] The local density estimation method is used to find the data point with the highest density in the cluster, which represents the typical scene corresponding to this cluster; the feature information corresponding to the scene indicator category of the highest data point in each cluster is the feature information of the corresponding typical scene.

[0139] The above description is only a preferred embodiment of the present application and an explanation of the technical principles used. Those skilled in the art should understand that the scope of the invention involved in the present application is not limited to the technical solution formed by a specific combination of the above technical features, but should also cover other technical solutions formed by any combination of the above technical features or their equivalent features without departing from the inventive concept. For example, the above features have similar functions to those disclosed in the present application (but not limited to).

Claims

1. A method for extracting accident scenes of assisted driving vehicles considering scene complexity, characterized in that: The following steps are involved: Setting accident scenario information for an assisted driving vehicle accident scenario; The accident scene information includes: scene complexity information, scene indicator categories corresponding to typical scene categories, and characteristic information of typical scenes; Collect historical accident data of assisted driving vehicles to obtain an assisted driving vehicle accident data set; Preprocess the assisted driving vehicle accident data set to obtain key indicator categories and their corresponding feature information; The feature information includes parametric feature information and non-parametric feature information; the non-parametric feature information is one-hot encoded, and the parametric feature information is kept in the original numerical form; The characteristic information corresponding to the key indicator categories of each accident was clustered and analyzed using the Gaussian mixture model to obtain three categories of scene complexity; The complexity is scored according to the complexity scoring method, and the complexity of the three types of scenarios are marked as simple scenarios, general scenarios, and complex scenarios; The characteristic information corresponding to the scene indicator category corresponding to the complexity of each type of scene is divided into s types of typical scenes using the DBSCAN clustering method; the characteristic information of the typical scene is assigned to each type of typical scene.

2. The method for extracting accident scenes of assisted driving vehicles considering scene complexity according to claim 1, characterized in that: The preprocessing of the assisted driving vehicle accident dataset includes the following steps: Remove outliers, fill in missing values, extract key indicator categories and feature information included in key indicator categories; Removing outliers includes determining outliers far from the mean through z-score standardization; Filling missing values ​​includes using the majority to fill missing values ​​in categorical data and using interpolation to fill missing values ​​in numerical data.

3. The method for extracting accident scenes of assisted driving vehicles considering scene complexity according to claim 1, characterized in that: The key indicator categories include: Road type, road surface conditions, ambient lighting conditions, weather conditions, vehicle speed at the time of collision, driver behavior, enabled driver assistance features, vehicle type, collision party type, collision type, and relative direction of motion.

4. The method for extracting accident scenes of assisted driving vehicles considering scene complexity according to claim 1, characterized in that: The non-parametric feature information includes feature information of road type, feature information of ambient lighting conditions, feature information of weather conditions, feature information of driver behavior, feature information of enabled assisted driving functions, feature information of vehicle type, feature information of collision party type, feature information of collision type, and feature information of relative motion direction; The parameter-based characteristic information includes characteristic information of the vehicle's speed at the moment of collision.

5. The method for extracting accident scenes of assisted driving vehicles considering scene complexity according to claim 1, characterized in that: The characteristic information of the road type includes: one of a tunnel, a sharp curve, a bridge, an intersection, a sidewalk, and an ordinary road section; The characteristic information of the road condition includes: one of dry, wet, waterlogged, icy, and snowy; The characteristic information of the ambient lighting conditions includes: one of daytime, nighttime with street lights, and nighttime without street lights; The characteristic information of the weather condition includes: one of sunny day, rainy day, snowy day and foggy day; The characteristic information of the driver's behavior includes one or more of normal driving, distraction, fatigue driving, speeding, and sudden braking; The characteristic information of the enabled assisted driving functions includes: one or more of FCW, AEB, ACC, and LDW; The characteristic information of the vehicle type includes: one of a sedan, a bus, a truck, and other types of vehicles; The characteristic information of the collision party type includes: pedestrian, bus, car, truck, fixed object, or one of other collision party types; The characteristic information of the collision type includes: hitting a person, scratching, hitting a fixed object, rollover, rear-end collision, frontal collision, and one of other collision types; The characteristic information of the relative motion direction includes: one of the same direction and the side direction.

6. The method for extracting accident scenes of assisted driving vehicles considering scene complexity according to claim 1, characterized in that: The complexity scoring method includes the following steps: The feature information corresponding to the key indicator category is standardized using the minimum-maximum standardization method to ensure that all feature values ​​are within the range of [0,1]; The feature weights of each key indicator category are calculated using the entropy weight method; By constructing a decision matrix, the complexity score of each accident sample is calculated based on the TOPS IS method; The scene with the highest complexity score is marked as a simple scene, the scene with a middle complexity score is marked as a general scene, and the scene with the lowest complexity score is marked as a complex scene.

7. The method for extracting accident scenes of assisted driving vehicles considering scene complexity according to claim 6, characterized in that: The standardized formula for standardizing the characteristic information corresponding to the key indicator category is: X is the original data of the feature information, min(X) and max(X) are the minimum and maximum values ​​of the feature information corresponding to the key indicator category, respectively. scaled It is the standardized data of characteristic information corresponding to the key indicator category.

8. The method for extracting accident scenes of assisted driving vehicles considering scene complexity according to claim 6, characterized in that: The entropy weight method calculates the feature weights of each key indicator category, including the following steps: The information entropy E of each key indicator category is calculated by formula 1 j ; Formula 1: It represents the proportion of the standardized value of the characteristic information corresponding to the jth key indicator category of the i-th sample to the total characteristic information; n is the number of accident samples; E j is the information entropy of the jth key indicator category; The characteristic weight of each key indicator category is calculated by formula 2; Formula 2: w j is the feature weight of the jth key indicator category; m is the number of key indicator categories.

9. The method for extracting accident scenes of assisted driving vehicles considering scene complexity according to claim 6, characterized in that: The basic process of calculating the complexity score of each accident sample based on the TOPSIS method is: Weighting of standardized data; X ij The standardized data of the characteristic information of the jth key indicator category of the i-th accident sample; X i ' j The standardized data of the characteristic information of the jth key indicator category of the i-th accident sample after weighting; n is a total of n accident samples; The decision matrix D′ is constructed by using the standardized data of the characteristic information corresponding to the key indicator categories after weighting n accident samples; X nm The standardized data of the characteristic information of the mth key indicator category of the nth accident sample; For each ideal solution of the jth key indicator category, is the negative ideal solution of the jth key indicator category; C i Score the complexity of the i-th accident sample.

10. The method for extracting accident scenes of assisted driving vehicles considering scene complexity according to claim 1, characterized in that: The complexity of each type of scene is divided into s typical scene categories using the DBSCAN clustering method; the scene indicator categories include road type, vehicle type, collision party type, relative motion direction, collision type, and enabled assisted driving functions; The characteristic information corresponding to the scene indicator category of the accident sample in each scene complexity level is used as the data point; Calculate the distance between each data point and other points and draw a k-distance graph. Select the best ε value through the "knee point" in the graph. The number of "knee points" in the graph is the number of typical scene categories s; By calculating the average distance between each point and its first K nearest neighbors, the appropriate statistical 95% percentile point is selected to determine MinPts; for each pair of ε and MinPts combination, the DBSCAN clustering algorithm is applied to cluster and obtain s clusters of data points, each cluster of data points corresponds to a typical scene; and s typical scenes corresponding to the complexity of each scene are obtained; and / or The process of assigning characteristic information of typical scenes to each type of typical scene includes the following steps: The local density estimation method is used to find the data point with the highest density in the cluster, which represents the typical scene corresponding to this cluster; the feature information corresponding to the scene indicator category of the highest data point in each cluster is the feature information of the corresponding typical scene.

Images