Vehicle behavior detection method, device and system based on vehicle AI platform

Abstract

The application relates to the technical field of road vehicle control systems, and discloses a vehicle behavior detection method, device and system based on a vehicle AI platform, which comprises the following steps: acquiring multiple types of vehicle driving information of a vehicle, weather information and highway map information of a location of the vehicle, and recording a plurality of traffic accident data; acquiring vehicle density of each vehicle based on similar performances between the vehicle driving information of the plurality of vehicles; quantifying road environment parameters of each vehicle; quantifying weather environment parameters of the vehicle; obtaining a driving environment coefficient of the vehicle; constructing a plurality of driving environment information groups of the vehicle, obtaining a plurality of related accident information groups of the vehicle; obtaining accident correlations of each type of vehicle driving information and each type of environment information of each vehicle; constructing an accident space graph of each vehicle; and acquiring a driving risk coefficient of the vehicle and performing a dangerous behavior early warning. The application aims to solve the problem that vehicle behavior detection lacks dynamic risk quantification capability based on road environment and traffic flow information.

Description

Technical Field

[0001] This invention relates to the field of road vehicle control system technology, and specifically to a vehicle behavior detection method, device, and system based on a vehicle AI platform. Background Technology

[0002] Since a large proportion of current road traffic accidents are related to dangerous driving behaviors, namely, vehicles failing to comply with traffic regulations and driving requirements while driving on highways, it is necessary to detect abnormal vehicle behavior in order to prevent traffic accidents and reduce the risk of accidents. However, current vehicle behavior detection methods rely on a single criterion, which means that the same driving behavior may be judged as abnormal in different scenarios. That is, the same driving behavior may correspond to completely different risk levels in highway and urban road scenarios. Existing technology has not yet effectively solved the mapping problem of "scenario-behavior-risk". Current vehicle behavior detection mostly adopts static risk assessment models and lacks the ability to dynamically quantify risks based on contextual information such as road environment and traffic flow. Summary of the Invention

[0003] This invention provides a vehicle behavior detection method, device, and system based on a vehicle AI platform to address the problem that existing vehicle behavior detection methods lack dynamic risk quantification capabilities based on information such as road environment and traffic flow. The specific technical solution adopted is as follows:

[0004] This invention proposes a vehicle behavior detection method based on a vehicle AI platform, which includes the following steps:

[0005] It acquires various types of vehicle driving information, as well as local weather and road map information, and transmits them to the vehicle AI platform, while also recording some traffic accident data.

[0006] Based on the similar performance of vehicle driving information of several vehicles, the vehicle density of each vehicle is obtained, and the road type weight of the vehicle's location is recorded; combined with the driving information of multiple types of vehicles and the local highway map information, the road environment parameters of each vehicle are quantified; the weather information of the vehicle's location is analyzed, and the weather environment parameters of the vehicle are quantified; combined with the vehicle's road environment parameters, the driving environment coefficient of the vehicle is obtained.

[0007] Based on vehicle driving information, as well as weather information and road map information of the vehicle's location as environmental information, several driving environment information groups for the vehicle are constructed. Combined with the similarity relationship between the same type of information in traffic accident data, several related accident information groups for the vehicle are obtained. The correlation between environmental information and vehicle driving information in the related accident information groups and their proportion in the traffic accident data are analyzed to obtain the accident correlation between various types of vehicle driving information and various types of environmental information for each vehicle.

[0008] Based on vehicle driving information, environmental information, and the aforementioned accident correlation, an accident space map for each vehicle is constructed; combined with the vehicle's driving environment coefficient, the vehicle's driving risk coefficient is obtained, and a warning of dangerous behavior is issued.

[0009] Optionally, the vehicle density of each vehicle is obtained using the following method:

[0010] Take any vehicle as the target vehicle, and construct the target vehicle's driving vector from the real-time positioning information and driving direction in the target vehicle's driving information; obtain the driving vectors of all other vehicles except the target vehicle, and obtain the cosine similarity between the driving vectors of all other vehicles and the driving vector of the target vehicle, as the driving similarity factor between the target vehicle and the driving vector of all other vehicles.

[0011] DBSCAN clustering is performed on all vehicles except the target vehicle. The distance metric is the absolute value of the difference between the driving similarity factors of each vehicle and the target vehicle, resulting in several clusters. The mean driving similarity factor between the vehicles in any cluster and the target vehicle is obtained, and the cluster with the largest mean driving similarity factor is taken as the target vehicle's peer cluster.

[0012] The product of the number of lanes on the road where the target vehicle is located and the minimum driving similarity factor between the target vehicle and the vehicles in the same lane cluster is obtained. The ratio of the number of vehicles in the same lane cluster to the product is used as the vehicle density of the target vehicle.

[0013] Optionally, the road environment parameters of each vehicle are obtained using the following method:

[0014] Based on the absolute value of the road slope, road curvature, road type weight, and road speed limit of the target vehicle's location, the road driving factor of the target vehicle is obtained; the average driving speed of all vehicles in the target vehicle's peer group is taken as the peer speed of the target vehicle.

[0015] Based on the vehicle density, speed of other vehicles, and road driving factors of the target vehicle, the road environment parameters of the target vehicle are obtained;

[0016] The road environment parameters are positively correlated with the road driving factor and the vehicle density, and negatively correlated with the travel speed; the road driving factor is positively correlated with the absolute value of the road slope and the road curvature, and negatively correlated with the road type weight and the road speed limit.

[0017] Optionally, the method for quantifying the vehicle's weather environment parameters and combining them with the vehicle's road environment parameters to obtain the vehicle's driving environment coefficient includes:

[0018] Set corresponding weights for different precipitation types; set corresponding weights for different road conditions;

[0019] The sum of the precipitation type weight and road condition weight of the target vehicle's location, and the ratio of the visibility of the target vehicle's location to the normalized result, are used as the weather environment parameters of the target vehicle.

[0020] The normalized product of the weather and road environmental parameters of the target vehicle is used as the driving environment coefficient of the target vehicle.

[0021] Optionally, the specific method for obtaining several sets of relevant accident information about the vehicle includes:

[0022] The vehicle's location is combined with various types of weather information and road map information to form the vehicle's multi-type environmental information; any type of vehicle driving information and any type of environmental information are combined into a driving environment information group, and several driving environment information groups are obtained for the target vehicle.

[0023] For any driving environment information group of the target vehicle, the element corresponding to the driving environment information group under each traffic accident data is obtained from several traffic accident data, and used as the accident information group of each traffic accident data in the driving environment information group.

[0024] Obtain the cosine similarity between each accident information group and the target vehicle's driving environment information group. If the cosine similarity is greater than or equal to the relevant threshold, the corresponding accident information group is taken as the relevant accident information group of the target vehicle's driving environment information group. Obtain the relevant accident information groups of each driving environment information group of the target vehicle.

[0025] Optionally, the specific methods for obtaining the accident correlation between various vehicle driving information and various environmental information include:

[0026] For any environmental information and any vehicle driving information, based on all relevant accident information groups of the target vehicle, the ratio of the number of relevant accident information groups containing the environmental information and the vehicle driving information to the number of relevant accident information groups containing the environmental information is used as the conditional probability of the vehicle driving information occurring under the environmental information.

[0027] The ratio of the number of relevant accident information groups containing environmental information and vehicle driving information to the total number of traffic accident data is used as the relative frequency of environmental information and vehicle driving information.

[0028] The product of the conditional probability of the occurrence of the vehicle driving information under the environmental information and the relative frequency of the environmental information and the vehicle driving information is used as the accident correlation factor between the target vehicle's driving information and the environmental information.

[0029] Obtain accident-related factors of various vehicle driving information and various environmental information of the target vehicle, perform linear normalization on all accident-related factors, and use the results as the accident correlation of various vehicle driving information and various environmental information of the target vehicle.

[0030] Optionally, the specific method for constructing the accident space map of each vehicle includes:

[0031] The target vehicle is taken as the central node, various types of vehicle driving information are taken as driving nodes, and various types of environmental information are taken as environmental nodes. The central node is connected to each driving node. For any driving node and any environmental node, the accident correlation between the vehicle driving information corresponding to the driving node and the environmental information corresponding to the environmental node is obtained. If the accident correlation is greater than or equal to the correlation threshold, the driving node and the environmental node are connected, and the accident correlation is taken as the edge value of the connection edge.

[0032] All driving nodes and environmental nodes are connected based on their corresponding accident correlations to complete the connections between the nodes and form an accident space map of the target vehicle.

[0033] Optionally, the driving risk coefficient of the vehicle is obtained using the following method:

[0034] The average value of the accident correlation between the vehicle driving information corresponding to each driving node and the environmental information corresponding to each environmental node in the accident space map of the target vehicle is obtained. The product of the average value and the driving environment coefficient of the target vehicle is used as the driving risk coefficient of the target vehicle.

[0035] This invention also proposes a vehicle behavior detection system based on a vehicle AI platform, the system comprising:

[0036] The vehicle information collection module is used to acquire various types of vehicle driving information, as well as local weather information and highway map information, and transmit them to the vehicle AI platform, and record some traffic accident data.

[0037] The vehicle information analysis module is used to obtain the vehicle density of each vehicle based on the similar performance of the vehicle driving information of several vehicles, and record the road type weight of the vehicle's location; combine the driving information of multiple types of vehicles and the local highway map information to quantify the road environment parameters of each vehicle; analyze the weather information of the vehicle's location to quantify the vehicle's weather environment parameters; and combine the vehicle's road environment parameters to obtain the vehicle's driving environment coefficient.

[0038] Based on vehicle driving information, as well as weather information and road map information of the vehicle's location as environmental information, several driving environment information groups for the vehicle are constructed. Combined with the similarity relationship between the same type of information in traffic accident data, several related accident information groups for the vehicle are obtained. The correlation between environmental information and vehicle driving information in the related accident information groups and their proportion in the traffic accident data are analyzed to obtain the accident correlation between various types of vehicle driving information and various types of environmental information for each vehicle.

[0039] The vehicle behavior detection module is used to construct an accident space map for each vehicle based on vehicle driving information, environmental information, and the aforementioned accident correlation; and to obtain the vehicle's driving risk coefficient and issue a warning for dangerous behavior by combining the vehicle's driving environment coefficient.

[0040] The present invention also proposes a vehicle behavior detection device based on a vehicle AI platform. The device includes a memory, a processor, and a computer program stored in the memory and running on the processor. When the processor executes the computer program, it implements the steps of the above method.

[0041] The beneficial effects of this invention are as follows: This invention monitors vehicle driving information in real time through a vehicle AI platform, and dynamically assesses the risk of driving behavior by combining factors such as weather information, road type, and traffic flow. Specifically, it determines the real-time vehicle density of a vehicle by comparing the similarities in driving information of different vehicles and considering the number of lanes, and then quantifies the driving environment coefficient by combining this with the impact of weather information on road conditions. Furthermore, it considers the correlation between driving information and environmental information (weather information and highway map information) based on historical traffic accident data, quantifying the accident-related performance between driving behavior and environmental information, thereby constructing an accident space map and analyzing the driving risk coefficient. Compared with traditional methods, this invention can provide more accurate driving risk prediction. If high-risk driving behavior is detected, the system will automatically issue a warning to help the driver reduce the possibility of accidents. Through quantitative analysis of weather and road environment, it judges whether the driver's behavior is reasonable. For example, speeding in rainy weather may lead to a higher accident risk; the system will adjust the risk assessment standard according to the weather conditions. By introducing a dynamic mapping of scenario-behavior-risk, more accurate driving behavior detection is achieved. Attached Figure Description

[0042] To more clearly illustrate the technical solutions in the embodiments of the present invention 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 some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0043] Figure 1This is a schematic diagram of a vehicle behavior detection method based on a vehicle AI platform provided in one embodiment of the present invention;

[0044] Figure 2 A structural block diagram of a vehicle behavior detection system based on a vehicle AI platform, provided for another embodiment of the present invention;

[0045] Figure 3 This is a structural schematic diagram of the accident space diagram of the target vehicle of the present invention. Detailed Implementation

[0046] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0047] Please see Figure 1 The diagram illustrates a flowchart of a vehicle behavior detection method based on a vehicle AI platform according to an embodiment of the present invention. The method includes the following steps:

[0048] Step S001: Obtain various types of vehicle driving information, as well as local weather information and road map information, and transmit them to the vehicle AI platform, and record some traffic accident data.

[0049] It should be noted that in order for the vehicle AI platform to effectively understand the vehicle's status and perform driving behavior detection and analysis, it is necessary to acquire and transmit various aspects of the vehicle's location information to the vehicle AI platform to provide a data foundation. The vehicle's status is directly related to the geographical environment or road environment in which the vehicle is located during driving, as well as the driver's driving behavior. Moreover, since different weather conditions also affect vehicle driving, for example, in freezing rain or snow, driving at unreasonable speeds or engaging in other driving behaviors on highways may lead to serious traffic accidents. Therefore, it is also necessary to obtain the weather information of the vehicle's location.

[0050] Specifically, vehicle driving information of several vehicles, as well as weather information and road map information of the vehicle's location, are obtained through the corresponding API and vehicle module unit. Since the vehicle AI platform is required for transmission, a large number of vehicles need to collect data in real time. The vehicle driving information includes the vehicle's real-time positioning information (longitude, latitude, altitude), driving speed (unit: km / h), driving direction (unit: degrees), and acceleration data (unit: m / s²). By obtaining the vehicle driving information through the vehicle's built-in Global Positioning System (GPS) module and onboard inertial measurement unit (IMU), various types of vehicle driving information are obtained.

[0051] Furthermore, real-time weather information of the vehicle's location can be obtained through the third-party weather service application programming interface (API, such as the National Meteorological Administration OpenAPI or commercial services such as AccuWeather API) integrated into the vehicle AI platform. The weather information includes temperature (real-time air temperature, unit: °C), humidity (relative humidity, unit: %), precipitation type (enumerated values: no precipitation / light rain / moderate rain / heavy rain / freezing rain / snow / hail), visibility (unit: meters), and road conditions (enumerated values: dry / wet / water accumulation / snow accumulation / ice), thus obtaining various types of weather information.

[0052] Furthermore, highway map information is obtained through the map data interface of the vehicle AI platform (the map data interface connects to high-precision digital map services (such as Here Maps or Gaode Map API)). The highway map information includes road type (enumerated values: expressway, urban expressway, ordinary urban road, rural road), speed limit value of the current road segment (unit: km / h), number of lanes, road curvature (unit: 1 / meter, representing the curve radius, the larger the value, the sharper the curve) and road slope (unit: %, positive value is uphill, negative value is downhill), thus obtaining highway map information of various types.

[0053] Furthermore, the collected vehicle driving information of various types, along with local weather and road map information, is encapsulated into a unified data packet via a 5G communication module and transmitted to the vehicle AI platform via the 5G network. At the same time, a large amount of traffic accident data is collected and stored on the vehicle AI platform. The traffic accident data includes vehicle driving information, weather information, and road map information at the time of each historical accident.

[0054] Step S002: Based on the similar performance of vehicle driving information of several vehicles, obtain the vehicle density of each vehicle and record the road type weight of the vehicle's location; combine the driving information of multiple types of vehicles and the local highway map information to quantify the road environment parameters of each vehicle; analyze the weather information of the vehicle's location to quantify the vehicle's weather environment parameters; combine the vehicle's road environment parameters to obtain the vehicle's driving environment coefficient.

[0055] It should be noted that whether a vehicle's behavior constitutes dangerous driving depends on the driving environment. For example, regarding driving speed, on different levels of roads, if a vehicle is traveling at high speed within the speed limit of a highway, then it is not considered dangerous driving. However, traveling at high speed on a non-highway road is considered dangerous driving. Therefore, in order to accurately detect vehicle behavior, vehicle AI platforms need to comprehensively analyze local weather information, vehicle driving information, and road map information to determine whether the vehicle's driving behavior constitutes dangerous driving under the given driving environment.

[0056] It should be further explained that the driving environment of a vehicle refers to the external environment of the vehicle, which is related to the road environment and weather conditions. Generally speaking, whether a vehicle's behavior is dangerous depends on whether the environment allows the corresponding behavior to exist. Therefore, it is necessary to analyze the vehicle's driving environment.

[0057] Preferably, in one embodiment of the present invention, the vehicle density of each vehicle is obtained based on the similarity of the driving information of several vehicles, and the road type weight of the vehicle's location is recorded. The specific method includes:

[0058] It should be noted that the road environment in which a vehicle is traveling usually involves the current road conditions and congestion. The road conditions include the road type, speed limit, number of lanes, curvature, and gradient of the road. The congestion includes the density of vehicles and average speed on the road. Therefore, in order to understand the environment in which a vehicle is currently traveling, it is necessary to quantify the road environment parameters.

[0059] Specifically, any vehicle is taken as the target vehicle, and its real-time positioning information and driving direction are used to construct its driving vector. The driving vectors of all other vehicles are obtained, and the cosine similarity between these vectors and the target vehicle's driving vector is used as the driving similarity factor between the other vehicles and the target vehicle. DBSCAN clustering is performed on all other vehicles except the target vehicle, using the absolute value of the difference between the driving similarity factors of the other vehicles and the target vehicle as the distance metric, resulting in several clusters. The mean driving similarity factor between vehicles in any cluster and the target vehicle is obtained, and the cluster with the largest mean driving similarity factor is taken as the target vehicle's peer cluster. The product of the number of lanes on the road where the target vehicle is located and the minimum driving similarity factor between vehicles in the peer cluster and the target vehicle is obtained. The ratio of the number of vehicles in the peer cluster to the product is used as the target vehicle's vehicle density.

[0060] It should be noted that the more other vehicles in the same cluster, and the smaller the minimum driving similarity factor, the more other vehicles with similar driving vectors the target vehicle has, and they are likely to be distributed in the same road segment. The smaller the number of lanes, the greater the vehicle density in the location of the target vehicle.

[0061] Furthermore, the road types in the highway map information of the vehicle's location include highways, urban expressways, ordinary urban roads, and rural roads. In this embodiment, corresponding road type weights are set for each road type: highways are 1, urban expressways are 2, ordinary urban roads are 3, and rural roads are 4. It should be noted that the road type weights of highways, urban expressways, ordinary urban roads, and rural roads in this embodiment are preset according to the road usage requirements. Therefore, the specific weight values ​​can be adjusted according to actual needs or actual conditions, but the order of weights should be kept as follows: highways < urban expressways < ordinary urban roads < rural roads to ensure the subsequent calculation results.

[0062] Preferably, in one embodiment of the present invention, the road environment parameters of each vehicle are quantified by combining multiple types of vehicle driving information and local highway map information, including the following specific methods:

[0063] Based on the absolute value of the road slope, road curvature, road type weight, and road speed limit of the target vehicle's location, the road driving factor of the target vehicle is obtained; the average driving speed of all vehicles in the target vehicle's travel cluster is taken as the travel speed of the target vehicle; based on the target vehicle's vehicle density, travel speed, and road driving factor, the road environment parameters of the target vehicle are obtained; the road environment parameters are positively correlated with the road driving factor and the vehicle density, and negatively correlated with the travel speed; the road driving factor is positively correlated with the absolute value of the road slope and the road curvature, and negatively correlated with the road type weight and the road speed limit.

[0064] As an example, the specific calculation method for the road environment parameters of the target vehicle is as follows:

[0065]

[0066] in, Indicates the road environment parameters of the target vehicle; Indicates the vehicle density of the target vehicle; Indicates the speed of the target vehicle traveling in the same direction; Indicates the road curvature where the target vehicle is located; Indicates the road gradient where the target vehicle is located; Indicates the weight of the road type where the target vehicle is located; Indicates the speed limit value of the road where the target vehicle is located; This represents the road travel factor of the target vehicle; Represents the absolute value function; This represents the sigmoid function, which is used for normalization in this embodiment.

[0067] It should be noted that road environment parameters are used to describe the degree of restriction on vehicle movement by the road environment on the road it is currently traveling on. The higher the value of the road environment parameter, the higher the degree of restriction. That is, when the road environment is poor, the road type, speed limit, number of lanes, road curvature and slope restrict the vehicle from driving normally. At the same time, due to the high vehicle density and low average speed on the road, the vehicle cannot drive normally, resulting in a greater degree of restriction on vehicle movement, and the corresponding road environment parameter is larger.

[0068] Preferably, in one embodiment of the present invention, the weather information of the vehicle's location is analyzed to quantify the vehicle's weather environment parameters; combined with the vehicle's road environment parameters, the driving environment coefficient of the vehicle is obtained, including the following specific methods:

[0069] In the weather information of the vehicle's location, this embodiment sets corresponding weights for precipitation types according to the degree of precipitation, specifically: no precipitation = 1, light rain = 2, moderate rain = 3, heavy rain = 4, freezing rain = 5, snow = 6, and hail = 7. At the same time, corresponding weights are set for road conditions, specifically: dry = 1, wet = 2, water accumulation = 3, snow accumulation = 4, and icing = 5. It should be noted that, similar to the road type weights, the precipitation type weights and road condition weights in this embodiment are set according to their influence on the road slipperiness. Therefore, the magnitude of the precipitation type weights and road condition weights can be adjusted according to actual needs or actual conditions, but the following order should be maintained: no precipitation < light rain < moderate rain < heavy rain < freezing rain < snow < hail, and dry < wet < water accumulation < snow accumulation < icing.

[0070] Furthermore, the sum of the precipitation type weight and road condition weight of the target vehicle's location, and the ratio of the visibility of the target vehicle's location to the normalized result, are used as the weather environment parameters of the target vehicle; in this embodiment, the sigmoid function is used for normalization.

[0071] It should be noted that weather environment parameters are used to describe the degree to which the weather environment at the current location restricts the vehicle's driving process. The higher the value of the weather environment parameter, the greater the restriction. That is, when the weather conditions are poor, such as low visibility, slippery roads and other adverse weather conditions that restrict the vehicle's driving, the greater the restriction on the vehicle's driving process, and the higher the corresponding weather environment parameter.

[0072] Furthermore, the product of the weather environment parameters and road environment parameters of the target vehicle and the result of normalization are used as the driving environment coefficient of the target vehicle; in this embodiment, the sigmoid function is used for normalization.

[0073] It should be noted that the driving environment coefficient is used to describe the degree of restriction a target vehicle is subject to under the combined constraints of road traffic conditions and weather conditions when it is driving. The higher the value of the driving environment coefficient, the more the target vehicle's driving process is restricted by the current environmental conditions. When analyzing the target vehicle's driving behavior, the more abnormal the vehicle's behavior becomes when it exceeds the limitations of the current environmental conditions.

[0074] Thus, the vehicle's driving environment coefficient is obtained.

[0075] Step S003: Based on the vehicle's driving information, as well as the weather information and road map information of the vehicle's location as environmental information, construct several driving environment information groups for the vehicle, and combine the similarity relationship between the same type of information in the traffic accident data to obtain several related accident information groups for the vehicle; analyze the correlation between environmental information and vehicle driving information in the related accident information groups and their proportion in the traffic accident data to obtain the accident correlation between various types of vehicle driving information and various types of environmental information for each vehicle.

[0076] It should be noted that in order to accurately obtain vehicle behavior detection results, it is necessary to further utilize the vehicle's driving information for behavior analysis. Based on the vehicle's driving environment as a criterion, the vehicle behavior is analyzed to determine whether the behavior shown in the vehicle's driving information is within the range allowed by the current driving environment.

[0077] Preferably, in one embodiment of the present invention, based on the vehicle's driving information and the weather information and road map information of the vehicle's location as environmental information, several driving environment information groups for the vehicle are constructed, and combined with the similarity relationship between similar types of information in traffic accident data, several related accident information groups for the vehicle are obtained. The specific method includes:

[0078] The vehicle's location is used to generate various types of environmental information, including weather and road map data. Any type of vehicle driving information and any type of environmental information are grouped together as a driving environment information group (vehicle driving information first, environmental information second, forming a vector; the values ​​of each dimension of the vector are the element values ​​of the corresponding type; specifically, for precipitation type, road condition, road type, etc., the corresponding element values ​​are weighted accordingly). Several driving environment information groups (element values ​​under the corresponding type of information for the target vehicle) are obtained for the target vehicle. For any driving environment information group of the target vehicle, the corresponding elements of that driving environment information group are obtained from several traffic accident data sets, serving as the accident information group for each traffic accident data set in that driving environment information group. The cosine similarity between each accident information group and the target vehicle's driving environment information group is obtained (cosine similarity is calculated using vectors). A preset threshold is used; in this embodiment, the threshold is described as 0.8. If the cosine similarity is greater than or equal to the threshold, the corresponding accident information group is considered as the relevant accident information group for the target vehicle's driving environment information group. The relevant accident information groups for each driving environment information group of the target vehicle are obtained using the above method.

[0079] Preferably, in one embodiment of the present invention, the correlation between environmental information and vehicle driving information in the relevant accident information group and their proportion in the traffic accident data are analyzed to obtain the accident correlation between various types of vehicle driving information and various types of environmental information. The specific method includes:

[0080] For any environmental information and any vehicle driving information, based on all relevant accident information groups of the target vehicle, the ratio (percentage) of the number of relevant accident information groups containing the environmental information and the vehicle driving information (corresponding type information, non-corresponding type element values) to the total number of relevant accident information groups containing the environmental information is used as the conditional probability of the vehicle driving information occurring under the environmental information; the ratio of the number of relevant accident information groups containing the environmental information and the vehicle driving information to the total number of traffic accident data is used as the relative frequency of the environmental information and the vehicle driving information; the product of the conditional probability of the vehicle driving information occurring under the environmental information and the relative frequency of the environmental information and the vehicle driving information is used as the accident correlation factor between the target vehicle's vehicle driving information and the environmental information; the accident correlation factors of various types of vehicle driving information and various types of environmental information of the target vehicle are obtained according to the above method, and all accident correlation factors are linearly normalized, and the result is used as the accident correlation between various types of vehicle driving information and various types of environmental information of the target vehicle.

[0081] It should be noted that by obtaining accident correlation, the strength of the association between specific driving behavior and traffic accident occurrence under specific environmental conditions is quantified. Conditional probability reflects the probability of an accident in the past for a target vehicle's driving behavior in the corresponding environment. The higher the probability, the less safe the behavior is, the more abnormal the driving behavior is, and the higher the probability of an accident. Relative frequency represents the proportion of vehicle driving information and environmental information combinations in all historical accidents. For example, if the number of relevant accident information groups is larger (such as "heavy rain + speeding" frequently occurring in accidents), the relative frequency is larger, indicating that the combination is a common cause of accidents. Physically, it reflects the representativeness of the behavior-environment combination in historical accidents. The higher the value, the more typical the combination is in high-risk scenarios. Therefore, the product of conditional probability and relative frequency combines the probability of the behavior occurring in the environment and the historical representativeness of the combination in accidents, reflecting the dual dependence of risk. That is, whether a vehicle behavior is dangerous depends not only on the behavior itself, but also on whether the environment "amplifies" the risk of the behavior. For example, snowy weather makes speeding more fatal.

[0082] At this point, the accident correlations between various vehicle driving information and various environmental information are obtained.

[0083] Step S004: Based on vehicle driving information, environmental information, and the aforementioned accident correlation, construct an accident space map for each vehicle; combine the vehicle's driving environment coefficient to obtain the vehicle's driving risk coefficient and issue a warning for dangerous behavior.

[0084] It should be noted that after obtaining the accident correlation between various vehicle driving information and environmental information, the accident correlation is compared by using a threshold, and this is used as a boundary value to connect the driving nodes corresponding to various vehicle driving information and the environmental nodes corresponding to various environmental information. The driving nodes, environmental nodes and the connecting edges constitute an accident space graph. The accident space graph intuitively reflects the correlation between various vehicle driving information and environmental information, providing a foundation for the subsequent comprehensive quantification of driving risk coefficient based on the correlation performance and driving environment coefficient.

[0085] Preferably, in one embodiment of the present invention, the method for constructing an accident spatial map of each vehicle based on vehicle driving information, environmental information, and the aforementioned accident correlation is as follows:

[0086] Using the target vehicle as the central node, various vehicle driving information as driving nodes, and various environmental information as environmental nodes, the central node is connected to each driving node. For any driving node and any environmental node, the accident correlation between the vehicle driving information corresponding to that driving node and the environmental information corresponding to that environmental node is obtained. A preset correlation threshold is used; in this embodiment, the correlation threshold is described as 0.7. If the accident correlation is greater than or equal to the correlation threshold, the driving node and the environmental node are connected, and the accident correlation is used as the edge value of the connection edge. Connection judgments are made for all driving nodes and environmental nodes based on their corresponding accident correlations to complete the connections between nodes, thus constructing an accident space graph of the target vehicle. Figure 3 As shown, this is a schematic diagram of the accident space map structure of a vehicle. For ease of display, the boundary values ​​are not labeled. The center of the accident space map is the target vehicle, the first layer of nodes are driving nodes, and the second layer of nodes are environment nodes. It should be noted that the schematic diagram of the accident space map structure is only used to intuitively show the connection method between nodes and is not intended to limit the connection relationship between nodes. That is, the connection relationship between nodes is constructed according to the specific method in this embodiment.

[0087] It should be noted that the corresponding edges between driving nodes and environment nodes in the accident space graph reflect the correlation between the corresponding nodes in terms of the probability of an accident. Through topology processing, the vehicle AI platform can understand the risk of vehicle behavior based on the environmental context based on the accident space graph in the subsequent process, thereby effectively detecting vehicle behavior.

[0088] It should be further explained that the accident relationship diagram intuitively reflects the correlation between vehicle driving information and environmental information. In the process of quantifying vehicle driving risk, the overall performance of the accident correlation between vehicle driving information and environmental information is used, and combined with the driving environment coefficient for comprehensive quantification. That is, in vehicle behavior detection, the environment is the basis of risk and behavior is the core logic of risk triggering conditions. By using the topological structure of the accident space diagram, historical traffic accident data is combined with the real-time environment and driving behavior of the target vehicle, avoiding the problem of unsatisfactory detection results caused by the rigidity of traditional vehicle behavior detection methods.

[0089] Preferably, in one embodiment of the present invention, the method for obtaining the vehicle's driving risk coefficient and issuing a warning for dangerous behavior by combining the vehicle's driving environment coefficient includes:

[0090] The average value of the accident correlation between the vehicle driving information corresponding to each driving node and the environmental information corresponding to each environmental node in the accident space map of the target vehicle is obtained. The product of the average value and the driving environment coefficient of the target vehicle is used as the driving risk coefficient of the target vehicle. A risk threshold is preset. In this embodiment, the risk threshold is described as 0.7. If the driving risk coefficient of the target vehicle is greater than or equal to the risk threshold, the target vehicle has dangerous behavior, and the vehicle AI platform sends a dangerous behavior warning information to the target vehicle.

[0091] This concludes the embodiment.

[0092] Please see Figure 2 This illustrates another embodiment of the vehicle behavior detection system based on a vehicle AI platform provided by the present invention, the system comprising:

[0093] Vehicle Information Collection Module 101: Acquires various types of vehicle driving information, as well as local weather information and highway map information, and transmits them to the vehicle AI platform, and records some traffic accident data;

[0094] Vehicle Information Analysis Module 102: Based on the similar performance of vehicle driving information of several vehicles, obtain the vehicle density of each vehicle and record the road type weight of the vehicle's location; combine the driving information of multiple types of vehicles and the local highway map information to quantify the road environment parameters of each vehicle; analyze the weather information of the vehicle's location to quantify the vehicle's weather environment parameters; combine the vehicle's road environment parameters to obtain the vehicle's driving environment coefficient.

[0095] Based on vehicle driving information, as well as weather information and road map information of the vehicle's location as environmental information, several driving environment information groups for the vehicle are constructed. Combined with the similarity relationship between the same type of information in traffic accident data, several related accident information groups for the vehicle are obtained. The correlation between environmental information and vehicle driving information in the related accident information groups and their proportion in the traffic accident data are analyzed to obtain the accident correlation between various types of vehicle driving information and various types of environmental information for each vehicle.

[0096] Vehicle behavior detection module 103: Based on vehicle driving information and environmental information, as well as the aforementioned accident correlation, it constructs an accident space map for each vehicle; combined with the vehicle's driving environment coefficient, it obtains the vehicle's driving risk coefficient and issues a warning for dangerous behavior.

[0097] Another embodiment of the present invention proposes a vehicle behavior detection device based on a vehicle AI platform. The device includes a memory, a processor, and a computer program stored in the memory and running on the processor. When the processor executes the computer program, it implements the above-described method steps S001 to S004.

[0098] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A vehicle behavior detection method based on a vehicle AI platform, characterized in that, The method includes the following steps: It acquires various types of vehicle driving information, as well as local weather and road map information, and transmits them to the vehicle AI platform, while also recording some traffic accident data. Based on the similar performance of vehicle driving information of several vehicles, the vehicle density of each vehicle is obtained, and the road type weight of the vehicle's location is recorded; combined with the driving information of multiple types of vehicles and the local highway map information, the road environment parameters of each vehicle are quantified; the weather information of the vehicle's location is analyzed, and the weather environment parameters of the vehicle are quantified; combined with the vehicle's road environment parameters, the driving environment coefficient of the vehicle is obtained. Based on vehicle driving information, as well as weather information and road map information of the vehicle's location as environmental information, several driving environment information groups for the vehicle are constructed. Combined with the similarity relationship between the same type of information in traffic accident data, several related accident information groups for the vehicle are obtained. The correlation between environmental information and vehicle driving information in the related accident information groups and their proportion in the traffic accident data are analyzed to obtain the accident correlation between various types of vehicle driving information and various types of environmental information for each vehicle. Based on vehicle driving information, environmental information, and the aforementioned accident correlation, an accident space map for each vehicle is constructed; combined with the vehicle's driving environment coefficient, the vehicle's driving risk coefficient is obtained, and a warning of dangerous behavior is issued. The vehicle density of each vehicle is obtained using the following method: Take any vehicle as the target vehicle, and construct the target vehicle's driving vector from the real-time positioning information and driving direction in the target vehicle's driving information; obtain the driving vectors of all other vehicles except the target vehicle, and obtain the cosine similarity between the driving vectors of all other vehicles and the driving vector of the target vehicle, as the driving similarity factor between the target vehicle and the driving vector of all other vehicles. DBSCAN clustering is performed on all vehicles except the target vehicle. The distance metric is the absolute value of the difference between the driving similarity factors of each vehicle and the target vehicle, resulting in several clusters. The mean driving similarity factor between the vehicles in any cluster and the target vehicle is obtained, and the cluster with the largest mean driving similarity factor is taken as the target vehicle's peer cluster. The product of the number of lanes on the road where the target vehicle is located and the minimum driving similarity factor between the target vehicle and the vehicles in the same lane cluster is obtained. The ratio of the number of vehicles in the same lane cluster to the product is used as the vehicle density of the target vehicle.

2. The vehicle behavior detection method based on a vehicle AI platform according to claim 1, characterized in that, The specific method for obtaining the road environment parameters of each vehicle is as follows: Based on the absolute value of the road slope, road curvature, road type weight, and road speed limit of the target vehicle's location, the road driving factor of the target vehicle is obtained; the average driving speed of all vehicles in the target vehicle's peer group is taken as the peer speed of the target vehicle. Based on the vehicle density, speed of other vehicles, and road driving factors of the target vehicle, the road environment parameters of the target vehicle are obtained; The road environment parameters are positively correlated with the road driving factor and the vehicle density, and negatively correlated with the travel speed; the road driving factor is positively correlated with the absolute value of the road slope and the road curvature, and negatively correlated with the road type weight and the road speed limit.

3. The vehicle behavior detection method based on a vehicle AI platform according to claim 1, characterized in that, The method for quantifying the vehicle's weather environment parameters and combining them with the vehicle's road environment parameters to obtain the vehicle's driving environment coefficient includes the following specific methods: Set corresponding weights for different precipitation types; set corresponding weights for different road conditions; The sum of the precipitation type weight and road condition weight of the target vehicle's location, and the ratio of the visibility of the target vehicle's location to the normalized result, are used as the weather environment parameters of the target vehicle. The normalized product of the weather and road environmental parameters of the target vehicle is used as the driving environment coefficient of the target vehicle.

4. The vehicle behavior detection method based on a vehicle AI platform according to claim 1, characterized in that, The specific method for obtaining several sets of relevant accident information about the vehicle is as follows: The vehicle's location is combined with various types of weather information and road map information to form the vehicle's multi-type environmental information; any type of vehicle driving information and any type of environmental information are combined into a driving environment information group, and several driving environment information groups are obtained for the target vehicle. For any driving environment information group of the target vehicle, the element corresponding to the driving environment information group under each traffic accident data is obtained from several traffic accident data, and used as the accident information group of each traffic accident data in the driving environment information group. Obtain the cosine similarity between each accident information group and the target vehicle's driving environment information group. If the cosine similarity is greater than or equal to the relevant threshold, the corresponding accident information group is taken as the relevant accident information group of the target vehicle's driving environment information group. Obtain the relevant accident information groups of each driving environment information group of the target vehicle.

5. The vehicle behavior detection method based on a vehicle AI platform according to claim 1, characterized in that, The specific methods for obtaining the accident correlation between various vehicle driving information and various environmental information are as follows: For any environmental information and any vehicle driving information, based on all relevant accident information groups of the target vehicle, the ratio of the number of relevant accident information groups containing the environmental information and the vehicle driving information to the number of relevant accident information groups containing the environmental information is used as the conditional probability of the vehicle driving information occurring under the environmental information. The ratio of the number of relevant accident information groups containing environmental information and vehicle driving information to the total number of traffic accident data is used as the relative frequency of environmental information and vehicle driving information. The product of the conditional probability of the occurrence of the vehicle driving information under the environmental information and the relative frequency of the environmental information and the vehicle driving information is used as the accident correlation factor between the target vehicle's driving information and the environmental information. Obtain accident-related factors of various vehicle driving information and various environmental information of the target vehicle, perform linear normalization on all accident-related factors, and use the results as the accident correlation of various vehicle driving information and various environmental information of the target vehicle.

6. The vehicle behavior detection method based on a vehicle AI platform according to claim 1, characterized in that, The specific methods for constructing the accident space map for each vehicle are as follows: The target vehicle is taken as the central node, various types of vehicle driving information are taken as driving nodes, and various types of environmental information are taken as environmental nodes. The central node is connected to each driving node. For any driving node and any environmental node, the accident correlation between the vehicle driving information corresponding to the driving node and the environmental information corresponding to the environmental node is obtained. If the accident correlation is greater than or equal to the correlation threshold, the driving node and the environmental node are connected, and the accident correlation is taken as the edge value of the connection edge. All driving nodes and environmental nodes are connected based on their corresponding accident correlations to complete the connections between the nodes and form an accident space map of the target vehicle.

7. The vehicle behavior detection method based on a vehicle AI platform according to claim 1, characterized in that, The driving risk coefficient of the vehicle is obtained using the following method: The average value of the accident correlation between the vehicle driving information corresponding to each driving node and the environmental information corresponding to each environmental node in the accident space map of the target vehicle is obtained. The product of the average value and the driving environment coefficient of the target vehicle is used as the driving risk coefficient of the target vehicle.

8. A vehicle behavior detection system based on a vehicle AI platform, characterized in that, The system implements the steps of the vehicle behavior detection method based on a vehicle AI platform as described in any one of claims 1-7 during operation, and the system includes: The vehicle information collection module is used to acquire various types of vehicle driving information, as well as local weather information and highway map information, and transmit them to the vehicle AI platform, and record some traffic accident data. The vehicle information analysis module is used to obtain the vehicle density of each vehicle based on the similar performance of the vehicle driving information of several vehicles, and record the road type weight of the vehicle's location; combine the driving information of multiple types of vehicles and the local highway map information to quantify the road environment parameters of each vehicle; analyze the weather information of the vehicle's location to quantify the vehicle's weather environment parameters; and combine the vehicle's road environment parameters to obtain the vehicle's driving environment coefficient. Based on vehicle driving information, as well as weather information and road map information of the vehicle's location as environmental information, several driving environment information groups for the vehicle are constructed. Combined with the similarity relationship between the same type of information in traffic accident data, several related accident information groups for the vehicle are obtained. The correlation between environmental information and vehicle driving information in the related accident information groups and their proportion in the traffic accident data are analyzed to obtain the accident correlation between various types of vehicle driving information and various types of environmental information for each vehicle. The vehicle behavior detection module is used to construct an accident space map for each vehicle based on vehicle driving information, environmental information, and the aforementioned accident correlation; and to obtain the vehicle's driving risk coefficient and issue a warning for dangerous behavior by combining the vehicle's driving environment coefficient.

9. A vehicle behavior detection device based on a vehicle AI platform, comprising a memory, a processor, and a computer program stored in the memory and running on the processor, characterized in that, When the processor executes the computer program, it implements the steps of the vehicle behavior detection method based on the vehicle AI platform as described in any one of claims 1-7.

Images