Full-process risk assessment methods, devices, equipment and media based on artificial intelligence

Abstract

The embodiments of the present application provide a full-process risk assessment method, device, equipment and medium based on artificial intelligence, which relate to the technical field of road risk assessment. The method includes: collecting road data and road information through vehicle equipment; processing road data and road information, comprehensively processing the road data and road information, obtaining various road risk factors, and encoding the characteristics of various road risk factors according to set rules; inputting factor features, combining several set models to conduct risk assessment on various road risk factors, and calculating the risk parameters of each road section; obtaining risk assessment results and generating output results, and obtaining risk assessment reports and maps; automatically extracting all factors related to road risks, which greatly liberates manpower, financial resources and material resources, reduces errors, and at the same time, the assessment results are comprehensive and accurate, improving the assessment effect and efficiency.

Description

Technical Field

[0001] The present application relates to the technical field of road risk assessment, and specifically, to a full-process risk assessment method, device, equipment and medium based on artificial intelligence. Background Art

[0002] Road traffic safety risk assessment is a systematic analysis and assessment of the probability of traffic accidents, the impacts caused and the possible consequences, in order to identify and quantify the potential risks of traffic accidents. Combined with the results of road risk assessment, it can effectively reduce the risk of road accidents. Specifically, road traffic safety risk assessment is a means or technology to obtain the risk value, risk level, and potential risks of different sections of a road. Then, based on the results of road risk assessment, the sections can: quantitatively predict future risk changes and conduct effective prevention and control; improve high-risk sections to reduce road traffic risks and enhance road safety; guide the allocation of funds for road maintenance and repair, and determine the investment plan that maximizes benefits.

[0003] At present, there are two main methods for road risk assessment: the first is to conduct road risk assessment based on various methods of the IRAP (Information Resource Application Platform) model and its variants; the second is to calculate the risk value through some statistical methods and learning methods after collecting video, accident and other data. For the first method: the model parameters are derived from human summary, and there are many parameters, which are difficult to adjust, and the effect is poor in actual use; the extraction of elements is cumbersome, requiring manual identification and feature coding as well as a lot of expert experience, with huge manpower costs, low efficiency and quality of risk assessment, and subjective factors in the assessment process, resulting in differences in evaluation standards. For the second method, most of the research and application of road risk assessment is aimed at road risks in the macro direction, mainly using macro accident data and planning, design and other data to study road risks and safety levels. In the research, the integration of other factors related to traffic safety (traffic operation, road environment, etc.) is not high, the accident type is not considered comprehensively, and the customization of the assessment model is poor. There are still shortcomings in the evaluation efficiency and accuracy of the evaluation results, and most of them use traditional methods such as linear regression, logistic regression, digital simulation, and machine learning in the era of small sample data. Summary of the invention

[0004] The purpose of the embodiments of the present application is to provide a full-process risk assessment method, device, equipment and medium based on artificial intelligence, so as to solve the problem that the existing road risk assessment method relies on manpower, does not consider all factors, and leads to poor assessment results, low efficiency and quality.

[0005] In a first aspect, the present application provides a full-process risk assessment method based on artificial intelligence, including:

[0006] Collect road data and road information through vehicle equipment;

[0007] Processing road data and road information, comprehensively processing the road data and road information, obtaining various road risk factors, and encoding the characteristics of various road risk factors according to set rules;

[0008] Input element characteristics, combine several set models to conduct risk assessment on various road risk elements, and calculate the risk parameters of each road section;

[0009] Obtain risk assessment results and generate output results to obtain risk assessment reports and maps.

[0010] In the above implementation process, road data and road information are collected through vehicle equipment; the road data and road information are processed, and various road risk factors are obtained by comprehensively processing the road data and road information, and the characteristics of various road risk factors are encoded according to set rules; the characteristics of the factors are input, and the risk assessment of various road risk factors is carried out in combination with several set models, and the risk parameters of each road section are calculated; the risk assessment results are obtained and the output results are generated to obtain risk assessment reports and maps; all factors related to road risks are automatically extracted, which greatly liberates human, financial and material resources, reduces errors, and at the same time the assessment results are comprehensive and accurate, improving the assessment effect and efficiency.

[0011] Furthermore, the collecting of road data and road information includes:

[0012] Comprehensive collection of road data through vehicle-mounted equipment;

[0013] Collect road information; road information includes but is not limited to accident records, weather records, design drawings, completion drawings, and inspection data.

[0014] In the above implementation process, road data and other road information are collected through the vehicle status equipment to ensure the comprehensiveness of data collection.

[0015] Furthermore, the road data and road information are processed, and the processed road data and road information are integrated to obtain various road risk factors, and the various road risk factors are feature-coded according to set rules, including:

[0016] Automatically extract various elements related to road safety from road data based on algorithm models;

[0017] According to the road data, search, crawl, and extract various network-related data corresponding to the road data;

[0018] Documentation of road information;

[0019] After comprehensive processing, the road data and road information are obtained to obtain various road risk factors;

[0020] The various road risk factors are divided into feature codes based on the set mileage and converted into vector expressions.

[0021] In the above implementation process, the dependence on expertise and the error-proneness of repetitive work when manually processing large amounts of data were resolved. At the same time, the visual large model technology was integrated to greatly reduce the cost and cycle of model development, while solving the problem of automatic calculation of vehicle speed, pile number, and line shape.

[0022] Furthermore, the automatic extraction of various elements related to road safety from road data based on the algorithm model includes:

[0023] Unify and integrate various road data;

[0024] Obtain the outline of road safety-related elements based on the multi-task perception model and the SAM pixel-level perception model;

[0025] Based on calling the tracking model and combining the radar point cloud information and the vehicle's own speed, the speed and lane of other vehicles in the driving process are calculated;

[0026] Automatically generate the road stake number at every moment during driving based on longitude and latitude, lateral and longitudinal acceleration, video and video technology;

[0027] Based on the linear calculation model, the radius of curves in the entire highway is calculated using the longitude and latitude information collected along the entire line.

[0028] In the above implementation process, all factors related to road risks are automatically extracted, which greatly liberates human, financial and material resources, reduces errors, and improves assessment results and efficiency.

[0029] Furthermore, the unified fusion of various road data includes:

[0030] According to the timestamp assigned by the clock module, the point cloud takes the point cloud information captured from the current timestamp point cloud to the next video frame as the basic unit, and associates the video frame with the radar point cloud;

[0031] Bilinear interpolation is used to interpolate the latitude and longitude information and acceleration information for each frame of video;

[0032] The relationship between the two-dimensional image and the three-dimensional point cloud is associated through the homography matrix, and the correspondence between the video pixels and the radar point cloud is associated.

[0033] In the above implementation process, various types of data are integrated to facilitate subsequent processing.

[0034] Furthermore, the step of acquiring the contours of road safety-related elements based on the multi-task perception model and the SAM pixel-level perception model includes:

[0035] Through the multi-task perception model, various road safety-related elements are obtained, and the elements are expressed in the form of rectangular boxes or dotted lines, and the contours of road safety-related elements are obtained by combining the SAM model;

[0036] Through morphological features and mapping elements from two-dimensional to three-dimensional, risk factors are obtained, among which risk factors include: guardrail height, sign layout size and sign occlusion.

[0037] Furthermore, the invoking of the tracking model and combining the radar point cloud information and the own vehicle speed to infer the speed and lane of other vehicles during driving include:

[0038] The tracking model is used to track vehicles on the road and distinguish vehicle types;

[0039] Combined with road data, determine the lane each tracked vehicle is in;

[0040] Map each tracked vehicle to the point cloud, obtain the position of the vehicle at each moment from the acquisition device, and calculate the vehicle speed by combining the change in the distance of the tracked vehicle relative to the vehicle within the set time and the speed of the vehicle;

[0041] Get the type, lane and speed of all vehicles.

[0042] Furthermore, the method of automatically generating the road stake number at each moment during the driving process according to the latitude and longitude, the lateral and longitudinal accelerations, the video and the video technology includes:

[0043] Initially give a starting pile number and driving direction;

[0044] Calculate the earth projection distance between the longitude and latitude of two adjacent moments, and correct it through acceleration information to obtain the vehicle's travel distance between the two moments;

[0045] According to the pile number information and driving direction at the previous moment, combined with the driving distance, the pile number information at the current moment is calculated;

[0046] During driving, whenever the road information collection module captures relevant information about the pile number plate, it transmits the pile number plate information to the pile number calculation part, identifies the pile number value through the OCR optical character recognition model, and calculates the distance between the pile number plate and the vehicle through the radar;

[0047] Check the current pile number information and make corrections.

[0048] Furthermore, the linear calculation model is based on which the radius of the curves existing in the entire highway is calculated through the longitude and latitude information collected along the entire highway, including:

[0049] Through the longitude and latitude sequence, a sliding window with actual distances of the first set distance, the second set distance and the third set distance is used to perform sliding calculations in both the uplink and downlink directions of the entire line;

[0050] Among them, a convex curve is fitted according to the GPS coordinates in each window, and the radius of the curve is calculated by taking the set key points;

[0051] The full-line radius set obtained by setting sliding windows is fused to obtain the full-line result.

[0052] Furthermore, the input element characteristics are combined with several set models to perform risk assessment on various road risk elements and calculate the risk parameters of each road section, including:

[0053] The risk reasoning model calculates the management risk, liability risk, accident risk level, risk factors and corresponding disposal measures of the road based on the input features and road characteristics, encodes the inference results, and outputs the automated processing results;

[0054] The IRAP information resource application platform model selects and converts the parameters required by the IRAP model based on the automated processing results, and calculates the risk value and risk factors of each road section;

[0055] The risk calculation model uses the element coding obtained from the automated processing results, the calculation results of the risk reasoning model and the IRAP model, and the original video sequence as input to comprehensively calculate the risk level of each road section and output a risk map;

[0056] The maintenance decision model is used to combine the treatment measures, risk levels, risk factors, and risk changes and costs after treatment of each road section to give final maintenance decision recommendations for each road section.

[0057] In the above implementation process, the knowledge reasoning part is added to make the model interpretable and constrained, and will not violate the basic criteria, ensuring the rationality and accuracy of the evaluation results. At the same time, treatment countermeasures are integrated, and treatment countermeasures can be given for each type of problem based on the situation; based on the attention mechanism, pictures, elements and other model results are integrated to make the model learning more sufficient and comprehensive, and the results more accurate.

[0058] In a second aspect, the embodiment of the present application provides a full-process risk assessment device based on artificial intelligence, including:

[0059] An information collection module is used to collect road data and road information through vehicle equipment;

[0060] A feature coding module is used to process road data and road information, comprehensively process the processed road data and road information, obtain various road risk factors, and perform feature coding on various road risk factors according to set rules;

[0061] The risk assessment module is used to input the element characteristics, combine several set models to conduct risk assessment on various road risk elements, and calculate the risk parameters of each road section;

[0062] The result output module is used to obtain risk assessment results and generate output results to obtain risk assessment reports and maps.

[0063] In a third aspect, an embodiment of the present application provides an electronic device, including:

[0064] A processor, a memory and a bus, wherein the processor is connected to the memory via the bus, and the memory stores computer-readable instructions. When the computer-readable instructions are executed by the processor, they are used to implement the full-process risk assessment method based on artificial intelligence as described above.

[0065] In a fourth aspect, an embodiment of the present application provides a computer-readable storage medium, on which a computer program is stored. When the computer program is executed by a server, the full-process risk assessment method based on artificial intelligence as described above is implemented.

[0066] In a fifth aspect, an embodiment of the present application provides a computer program product, which includes instructions, and when the instructions are executed by a computer, the computer implements the method as described above. BRIEF DESCRIPTION OF THE DRAWINGS

[0067] In order to more clearly illustrate the technical solutions of the embodiments of the present application, the drawings required for use in the embodiments of the present application will be briefly introduced below. It should be understood that the following drawings only show certain embodiments of the present application and therefore should not be regarded as limiting the scope. For ordinary technicians in this field, other related drawings can be obtained based on these drawings without paying creative work.

[0068] Figure 1 A schematic diagram of a process flow of a full-process risk assessment method based on artificial intelligence provided in an embodiment of the present application;

[0069] Figure 2 A schematic diagram of elements in the form of dots and lines of a full-process risk assessment method based on artificial intelligence provided in an embodiment of the present application;

[0070] Figure 3A schematic diagram of elements embodied in the form of a rectangular frame of a full-process risk assessment method based on artificial intelligence provided in an embodiment of the present application;

[0071] Figure 4 A schematic diagram of the area of ​​all objects in an image obtained by a SAM model in a full-process risk assessment method based on artificial intelligence provided in an embodiment of the present application;

[0072] Figure 5 A profile diagram of each type of element related to road safety in a full-process risk assessment method based on artificial intelligence provided in an embodiment of the present application;

[0073] Figure 6 A schematic diagram of an algorithm flow of a linear calculation model of a full-process risk assessment method based on artificial intelligence provided in an embodiment of the present application;

[0074] Figure 7 It is a structural schematic diagram of a full-process risk assessment device based on artificial intelligence provided in an embodiment of the present application;

[0075] Figure 8 It is a structural schematic diagram of an electronic device provided in an embodiment of the present application. DETAILED DESCRIPTION

[0076] The technical solutions in the embodiments of the present application will be described below in conjunction with the drawings in the embodiments of the present application.

[0077] It should be noted that similar reference numerals and letters represent similar items in the following drawings, so once an item is defined in one drawing, it does not need to be further defined and explained in subsequent drawings. At the same time, in the description of this application, the terms "first", "second", etc. are only used to distinguish the description and cannot be understood as indicating or implying relative importance.

[0078] At present, there are two main methods for road risk assessment:

[0079] The first category is to conduct road risk assessment based on IRAP (and its variant methods) (the International Road Assessment Organization IRAP amplifies the collected traffic accident data of a certain section of the road into annual data for the road based on the national conditions, traffic habits and environment of each country, and uses the data to assess the safety level of the roads within the area where the road is located, so as to propose countermeasures to improve traffic safety and improve the facilities along the road). The specific process is as follows: Collect video information of the road and some other information. Through experts watching the collected road videos, extract various types of information in the rightmost column of the video below per kilometer (or 100 meters), and then multiply and add the extracted elements per kilometer (or 100 meters) with the preset parameters of the IRAP model to obtain the final risk value. According to the risk value obtained per kilometer (or 100 meters), the road is divided into 1-5 levels (1 is the safest and 5 is the most dangerous) according to the threshold.

[0080] The second category is to calculate the risk value through some statistical methods and learning methods after collecting data such as videos and accidents (this type of method mainly stays in the publication of papers and theoretical research, and has poor practicality), including: quantitative risk assessment methods, qualitative risk assessment methods, and comprehensive assessment methods combining qualitative and quantitative methods. Typical quantitative analysis methods include factor analysis, cluster analysis, time series model, regression model, equal risk graph method, decision tree method, etc. Typical qualitative analysis methods include factor analysis, logical analysis, historical comparison method, and Delph method. Among the comprehensive assessment methods combining qualitative and quantitative methods, the Analytic Hierarchy Process (AHP) proposed by Professor Satty of the University of Pittsburgh in the United States is the most commonly used method. In addition, risk assessment methods based on fuzzy comprehensive evaluation, risk assessment methods based on DS evidence theory, risk assessment methods based on machine learning, and risk assessment methods based on gray theory have also made a series of improvements and supplements to the deficiencies in risk assessment from different perspectives.

[0081] For the first type of method: 1. The model parameters are derived from manual summary, and there are many parameters, which are difficult to adjust, and the effect is poor in actual use; 2. The extraction of factors is cumbersome, requiring manual identification and feature coding as well as a lot of expert experience, with huge manpower costs, low risk assessment efficiency and quality, and subjective factors in the assessment process, which leads to differences in evaluation standards.

[0082] For the second type of methods, most of the research and application of road risk assessment are aimed at road risks in a macro sense, mainly using macro accident data and planning, design and other data to study road risks and safety levels. In the research, other factors related to traffic safety (traffic operation, road environment, etc.) are not integrated well, accident types are not considered comprehensively, and the customization of the assessment model is poor. There are still shortcomings in the assessment efficiency and accuracy of the assessment results, and most of them use traditional methods such as linear regression, logistic regression, digital simulation, and machine learning in the era of small sample data.

[0083] Based on the above problems, an embodiment of the present application proposes a full-process risk assessment method based on artificial intelligence to solve the above problems.

[0084] Please see Figure 1 , Figure 1 A schematic diagram of a process flow of a full-process risk assessment method based on artificial intelligence provided in an embodiment of the present application. The full-process risk assessment method based on artificial intelligence includes:

[0085] 100. Collect road data and road information through vehicle equipment.

[0086] Specifically, the road data is comprehensively collected through vehicle-mounted equipment; road information is collected; wherein the road information includes but is not limited to accident records, weather records, design drawings, completion drawings, and inspection data.

[0087] Among them, data collection is mainly carried out by driving the vehicle-mounted equipment in both directions along the evaluated section to collect road data (including radar), and at the same time collect some other road information (accident records, weather records, design drawings, completion drawings, test data, etc.).

[0088] For example, the equipment acquisition part is more critical, and supporting equipment has been developed. The equipment integrates laser radar, camera, Beidou chip and antenna, inertial navigation and clock module. The radar obtains the point cloud information of the road, the camera obtains the video of the road, the Beidou chip and antenna obtain the longitude and latitude information during driving, the inertial navigation obtains the lateral and longitudinal acceleration and vehicle turning angle during driving, and the clock module generates a timestamp every 10 milliseconds (allocated other data). Optionally, the device fixes the fuselage on the front cover or roof of the vehicle through four magnetic suction cups.

[0089] For example, other ways of collecting data are to collect several fixed types of data during road construction and operation and maintenance, including weather records, accident records, design drawings, completion drawings, and maintenance records.

[0090] 200. Process the road data and road information, obtain various road risk factors based on the processed road data and road information, and perform feature coding on the various road risk factors according to set rules.

[0091] 210. Automatically extract various elements related to road safety from road data based on algorithm models.

[0092] 211. Unify and integrate various road data.

[0093] Optionally, according to the timestamp assigned by the clock module and taking the video frame timestamp as the basic unit, the point cloud takes the point cloud information captured from the current timestamp point cloud to the next video frame, and the video frame and the radar point cloud are associated; specifically, according to the timestamp assigned by the clock module, the video frame, radar point cloud, longitude and latitude, and acceleration information are associated together: for the case of non-uniform acquisition frequency (video frame is 25 frames per second, point cloud is continuous data, longitude and latitude and acceleration information are 5 pieces of information per second), the video frame timestamp is taken as the basic unit, and the point cloud takes the point cloud information captured from the current timestamp point cloud to the next video frame; bilinear interpolation is used to interpolate longitude and latitude information and acceleration information for each frame of video.

[0094] In addition, the relationship between the two-dimensional image and the three-dimensional point cloud is associated through the homography matrix, and the correspondence between the video pixels and the radar point cloud is associated.

[0095] 212. Obtain the contours of road safety-related elements based on the multi-task perception model and the SAM pixel-level perception model.

[0096] Optionally, various road safety-related elements are obtained through a multi-task perception model, and the elements are expressed in the form of rectangular boxes or dotted lines. The SAM model is combined to obtain the outline of the road safety-related elements. The risk factors are obtained through morphological features and mapping of the elements from two-dimensional to three-dimensional, where the risk factors include: guardrail height, sign layout size and sign occlusion.

[0097] Specifically, the multi-task perception model is used to perceive signs (including 80 types of signs and sign hanging methods), markings (including 20 types of markings, marking colors, damage status, etc.), guardrails (including more than ten types of guardrails, guardrail transition forms, guardrail transition sections, etc.), other traffic safety facilities (cones, contours, anti-glare panels, road spikes, etc.), environments (excavations, villages, buildings, overpasses, bridge piers, etc.), vehicles (different types of vehicles), mileage posts and other road safety-related factors from the video. The model structure adopts the idea of ​​multi-task learning. Through multi-task learning, a single network is used to extract multiple different types of risk factors, and the tasks promote each other. The model structure includes: feature extraction network, prediction head network, loss function and multi-task learning model architecture.

[0098] Among them, the feature extraction network uses the backbone (backbone network) and neck (connection part) of yolo-v8 as the feature extraction network. A series of convolution and deconvolution layers are used to extract features. At the same time, residual connections and bottleneck structures are used to reduce the size of the network and improve performance. The C2f module (processed by a convolution module with k=1, s=1, p=0, c=c_out, and then processed by a split process and processed by n DarknetBottleneck modules, and the results of the residual module and the backbone module are concat spliced ​​and output after processing by a convolution module) is used as the basic building block. The C2f module has fewer parameters and better feature extraction capabilities. In the Neck part, multi-scale feature fusion technology is used to fuse feature maps from different stages of the backbone to enhance feature representation capabilities. Specifically, it includes an SPPF module, a PAA module, and two PAN modules.

[0099] Among them, the prediction subnetwork is divided into four parts: traffic safety facility detection subnetwork, damaged marking segmentation subnetwork, lane line recognition subnetwork and road environment judgment subnetwork.

[0100] Traffic safety facilities and vehicle detection subnetwork: During the prediction process, the network performs target detection by sliding convolution. Specifically, the confidence score of each default box corresponding to the c+1 class (target class and background class) and the offset of the bounding box are calculated at each position of the feature map used for prediction. Therefore, for each feature map, ((c+1)+4)×6 convolution kernels are used to perform sliding convolution on the feature map to generate a feature map of size m×n×(((c+1)+4)×6). Among them, m and n are the sizes of the input feature map, and the values ​​on each pixel channel in the feature map correspond to the confidence scores of the default boxes associated with the position as the target class, the background class, and the regression values ​​of the bounding box.

[0101] Damaged marking segmentation subnetwork: It is divided into two prediction layers, which are used to predict the marking area in the complete state and the marking state in the actual situation respectively; among them, both prediction networks are composed of network units consisting of 3*3 deconvolution layers, 3*3 and 1*1 convolution layers. Finally, 1*1 convolution is used to predict the segmentation map of the marking in the complete state and the segmentation map in the actual state respectively, so as to calculate the damaged state of the marking.

[0102] Road environment discrimination subnetwork: It adopts the idea of ​​full convolution. For the feature map, it combines 3*3 convolution and 1*1 convolution. Finally, it divides the original image into multiple grid areas through a 3*3*N convolution layer. It predicts the probability of the existence of cliffs, water, bends, and village sections in each grid area. N represents the number of scene types that need to be divided.

[0103] Lane line recognition subnetwork: pre-define multiple row anchors in the image, divide each anchor into multiple grids, predict the existence probability of lane lines in each grid, combine the structural constraints of lane lines, and finally realize the detection of lane line positions.

[0104] Among them, the loss function: during joint training, the loss function adopts the weighted method of multiple sub-network losses, and each branch task is trained separately using the corresponding task loss.

[0105] Exemplarily, the traffic safety facility detection loss consists of two parts, namely, the cross entropy loss for determining the category of the traffic safety facility and the regression loss for correcting the location of the traffic safety facility:

[0106] ;

[0107] in, They represent the category to which the default box belongs, the confidence score predicted by the network, the regression value predicted by the network, and the true regression value respectively; is the cross entropy loss for classification, It is used for regional location and size regression loss; Represents the number of positive samples (identified as target classes) and negative samples used for training in each batch, is the number of samples to calculate the regression loss. The weight used to balance the two types of losses.

[0108] The cross entropy loss for classification is defined as follows:

[0109] ;

[0110] in, The value is {0,1}, indicating whether the label of the i-th default box matches the j-th annotation box of category p. If so, it is 1, otherwise it is 0; represents the probability that the network predicts the i-th default box as the p-th class, Represents the probability that the network predicts the i-th default box as the background class.

[0111] For the positioning loss, the regression loss of predicting the offset (cx, cy) between the default box and the center coordinates of the true region, the width (w) and height (h) of the default box are defined as follows:

[0112] ;

[0113] For example, the marking damage segmentation loss consists of three parts, namely, the marking integrity state prediction loss, the marking actual state prediction loss and the damage rate regression loss:

[0114] ;

[0115] in Dice loss is a metric function used to evaluate the similarity of two samples. Its value range is between 0 and 1. The larger the value, the more similar they are:

[0116] ;

[0117] in, is the intersection between X and Y, and Represents the number of elements of X and Y respectively.

[0118] For example, road environment discrimination loss: Considering that the normal environment accounts for the vast majority of road environments, the road environment uses grid-by-grid Focal loss as the loss to balance the difference between positive and negative samples. For a single image, the loss of the grids used is averaged:

[0119] ;

[0120] The Focal loss formula is:

[0121] ;

[0122] in, Set to 2, when a sample is misclassified, is very small, then the modulation factor Close to 1, the loss is not affected; when →1, factor If it is close to 0, the weight of well-classified samples will be lowered, thus achieving the purpose of focusing on difficult samples.

[0123] For example, the lane recognition loss is divided into three parts:

[0124] ;

[0125] , , They correspond to classification cross entropy loss, structural loss and segmentation loss respectively, where the structural loss is divided into similarity loss and shape loss:

[0126] ;

[0127] Among them, the similarity loss Used to constrain the similarity of the predicted probabilities between two adjacent row anchors, shape loss It is used to constrain the shape of the lane line:

[0128] ;

[0129] ;

[0130] Total loss of joint training:

[0131] ;

[0132] in, To detect damage to traffic safety facilities, The loss of road environment. The loss caused by the broken marking line. Loss for lane line recognition; , , , is a weight factor used to balance the weight losses between different tasks, which changes in a learning manner during the training process.

[0133] As mentioned above, the multi-task perception model can obtain various road safety-related elements such as signs (including 80 types of signs and sign hanging methods), markings (including 20 types of markings, marking colors, damage status, etc.), guardrails (including more than ten types of guardrails, guardrail transition forms, guardrail transition sections, etc.), other traffic safety facilities (cones, contours, anti-glare plates, road spikes, etc.), environment (excavation, villages, buildings, overpasses, bridge piers, etc.), vehicles (by vehicle type), mileage posts, etc. These elements are mainly expressed in the form of rectangular frames or dotted lines, such as Figure 2 and Figure 3 The rectangular frame and dotted lines lack the outline and shape of various facilities, and lack boundary information when matching with radar.

[0134] Specifically, the SAM pixel-level perception model SAM is the Segment Anything Model (SAM) proposed by Meta, which can divide all pixels of the same category in the image. It includes a powerful image encoder (calculating image embedding), a hint encoder (calculating hint embedding), and a lightweight mask decoder (real-time prediction of masks).

[0135] Among them, the image encoder: based on a scalable and powerful pre-training method, using ViT pre-trained with MAE, is minimally suitable for processing high-resolution inputs; the image encoder runs once for each image and is applied before the prompt model.

[0136] Among them, the prompt encoder: consider two sets of prompts: sparse (points, boxes, text) and dense (masks). Points and boxes are represented by position encoding, and the learned embeddings and free-form text for each prompt type are added to the ready-made text encoding in CLIP. Dense prompts (i.e., masks) are embedded using convolutions and element-wise summed over image embeddings.

[0137] Among them, Mask Decoder: The mask decoder efficiently maps image embedding, hint embedding, and output token to masks.

[0138] like Figure 4 As shown in the figure, the SAM model can obtain the regions of all objects in the image (and split the image into different colors, the same color represents the same object), but the SAM model can only distinguish different objects, but cannot determine what these objects are. The multi-task perception model can obtain the categories of different objects (what they are), but cannot obtain their specific regions and contours (can only be wrapped by bounding boxes). Therefore, the two models can be combined to obtain the contours of each type of elements related to road safety. Please refer to Figure 5 shown.

[0139] In addition, through morphological features and the previous 2D to 3D mapping, risk factors such as guardrail height, sign layout size, and sign occlusion can be obtained.

[0140] 213. Based on calling the tracking model and combining the radar point cloud information and the vehicle's own speed, the speed and lane of other vehicles in the driving process are calculated.

[0141] Specifically, the tracking model YOLOV8 is used to track vehicles on the way and distinguish vehicle types; combined with road data, the lane of each tracked vehicle is determined; each tracked vehicle is mapped to the point cloud, and the position of the vehicle at each moment from the collection device of the vehicle is obtained, and the vehicle speed is calculated by combining the change in the distance of the tracked vehicle relative to the vehicle within the set time and the speed of the vehicle; the type, driving lane and speed of all vehicles are obtained.

[0142] 214. According to longitude and latitude, lateral and longitudinal acceleration, video and video technology, the road stake number at every moment in the driving process is automatically generated.

[0143] Specifically, a starting pile number and a driving direction are initially given, and optionally, it can be set to increase driving along the pile number or decrease driving along the pile number, etc.; the earth projection distance between the longitude and latitude of two adjacent moments is calculated, and corrected by acceleration information to obtain the driving distance of the vehicle between the two moments; the pile number information at the previous moment is calculated based on the driving direction and the driving distance; during the driving process, whenever the road information acquisition module captures the relevant information of the pile number plate, the pile number plate information is transmitted to the pile number calculation part, the pile number value is identified by the OCR optical character recognition model, and the distance between the pile number plate and the vehicle is calculated by radar; the current pile number information is verified and corrected.

[0144] 215. Based on the linear calculation model, the radius of the curves in the entire highway is calculated using the longitude and latitude information collected along the entire line.

[0145] Specifically, through the longitude and latitude sequence, sliding calculations are performed on the entire line in both the uplink and downlink directions in combination with sliding windows whose actual distances are the first set distance, the second set distance and the third set distance respectively; wherein a convex curve is fitted according to the GPS coordinates in each window, and the curve radius is calculated by taking a set number of key points; the set of full-line radius obtained for the set number of sliding windows is fused to obtain the full-line result.

[0146] For example, please refer to Figure 6 As shown in the algorithm flow, sliding calculations are performed in both the uplink and downlink directions of the entire line through sliding windows with actual distances of 100m, 200m and 300m respectively. A convex curve is fitted according to the GPS coordinates in each window, and three key points are taken to calculate the curve radius; the radius sets of the entire line obtained from the three sliding windows are fused to obtain the final full-line result.

[0147] As mentioned above, the automatic extraction of all factors related to road risks has greatly liberated human, financial and material resources, reduced errors, and improved assessment results and efficiency.

[0148] 220. Based on the road data, search, crawl and extract various network-related data corresponding to the road data.

[0149] Specifically, by collecting various types of road-related information, various types of relevant data are collected and crawled from the Internet to supplement it, including: obtaining historical accident information from the navigation platform based on longitude and latitude information; obtaining historical weather information from the meteorological platform based on longitude and latitude information; designing regional nouns based on road names and accident records to crawl on several social networking sites, crawl related topic articles, and judge the topic and sentiment through LDA (Latent Dirichlet Allocation).

[0150] 230. Document the road information.

[0151] Optionally, a unified formatted document is set for information such as accidents, weather, and drawings, a fixed format conversion method is set for word, excel and other files, and PDF, pictures, etc. are converted into formatted documents using OCR and document parsing models.

[0152] 240. After comprehensive processing, the road data and road information are obtained to obtain various road risk factors.

[0153] 250. Feature encoding is performed on various road risk factors based on set mileage and converted into vector expression.

[0154] Optionally, various road safety-related elements extracted by the automated processing part are divided into whole kilometers (or hundred meters) for feature encoding and converted into vector expressions.

[0155] As described above, the embodiments of the present application solve the dependence on expertise and the error-proneness of repetitive work when manually processing large amounts of data, while incorporating visual large model technology to greatly reduce the cost and cycle of model development, while solving the problem of automatic calculation of vehicle speed, pile number, and line shape.

[0156] 300. Input feature characteristics, combine several set models to conduct risk assessment on various road risk factors, and calculate the risk parameters of each road section.

[0157] Optionally, the risk assessment is divided into four parts: risk reasoning model, IRAP model, risk calculation model and maintenance decision model.

[0158] 310. The risk reasoning model calculates the management risk, liability risk, accident risk level, risk factors and corresponding disposal measures of the road based on the input features and road characteristics, encodes the inference results, and outputs the automated processing results.

[0159] Optionally, the risk reasoning model manually extracts road safety-related standards, specifications, road safety-related expert experience, and project experience, converts this knowledge into a basic knowledge graph, and uses graph representation, embedding and other methods to complete the entire graph to construct a risk reasoning knowledge graph; when in use, the factor feature coding of the road is input, combined with information such as the road construction age, to activate the nodes and links of the graph, and then use rule reasoning, graph convolution, and graph topology methods to perform link prediction, which is then converted into existing management risks, liability risks, and accident risk levels and risk factors and corresponding disposal measures.

[0160] Specifically, the embodiments of the present application build knowledge systems such as vehicles, roads, and fields based on standard specifications related to roads, vehicles, communications, and other fields in different periods and regions; convert the nodes (entities) in the knowledge system into vectors through an embedding model to establish an interconnected and sound knowledge graph; learn node representation and prediction through R-GCN, HetGNN, HAN heterogeneous graph neural network and heterogeneous graph attention network; and evaluate road safety through knowledge reasoning.

[0161] 320. The IRAP information resource application platform model selects and converts the parameters required by the IRAP model based on the automated processing results, and calculates the risk value and risk factors of each road section.

[0162] It can be understood that the automated processing part obtains the risk factor characteristic code, which includes the elements required by the IRAP model. In this part, the characteristic code is converted into the format required by the IRAP model, and the risk value and risk factor of the road section are obtained through the IRAP model.

[0163] 330. The risk calculation model uses the element coding obtained from the automated processing results, the calculation results of the risk reasoning model and the IRAP model, and the original video sequence as input to comprehensively calculate the risk level of each road section and output a risk map.

[0164] Optionally, the risk calculation model has three parts for input: video frames, automated processing to obtain feature encoding, risk reasoning model, and IRAP model results. Based on the Transformer technology structure, a trainable risk assessment model is constructed, which ultimately comprehensively outputs the risk levels and factors of each road section.

[0165] Specifically, the video information encoding process is as follows: according to the extracted risk factors, N 224*224 block areas are intercepted with each factor position as the center, and N 224*224 block areas are randomly cropped at the same time, and a total of block-shaped area, To specify the number of key elements, when the number of key elements of a single class is less than N, a 224*224 noise that follows a normal distribution is used. For each image block to be input, it is first passed through a CNN backbone network for feature extraction. Suppose the feature of the last convolutional layer is , where W, H and C are the width, height and number of channels of the convolutional feature map respectively. Further stretching into a feature sequence ,in Represents the length of the sequence. Since the convolution operation can maintain the spatial position relationship of the image, each dimension in L can be mapped back to the corresponding area of ​​the input image.

[0166] The results of feature coding, risk reasoning model, and IRAP model obtained from the automated processing are unified to the same length L, and the insufficient parts are filled with zero values. A feature vector of length L.

[0167] The construction length is The feature sequence of , where the dimension of each feature is L; the self-attention mechanism can be used to extract the correlation between different feature factors in the input sequence, which has multiple implementation forms, using the formula:

[0168] ;

[0169] Among them, Q, K and V are , and The query, key, and value matrices of ; On this basis, we can further introduce multi-head attention mechanism and position encoding, and construct the corresponding Transforme network by superimposing multiple self-attention calculations; after the network modeling is completed, we can learn the relationship and influence matrix between images of different factors.

[0170] Model training: The output features of the Transformer network are connected in parallel to the risk level classification branch, the risk cause classification branch and the smoothed accident casualty regression branch, and three types of constraint optimization are performed. Only the risk level and risk cause are retained in the prediction stage.

[0171] 340. The maintenance decision model is used to combine the treatment measures, risk level, risk factors, and risk changes and costs after treatment of each road section to provide final maintenance decision recommendations for each road section.

[0172] Therefore, the risk assessment adds the knowledge reasoning part, making the model interpretable and constrained, and not violating the basic criteria, ensuring the rationality and accuracy of the assessment results. At the same time, it incorporates treatment countermeasures, and can give treatment countermeasures for each type of problem in combination with the situation; based on the attention mechanism, it integrates pictures, elements and other model results, making the model learning more sufficient and comprehensive, and making the results more accurate. In the maintenance decision part of risk assessment, through reinforcement learning, the trade-off between maintenance funds and road risk increase can be comprehensively considered, and the optimal reference suggestions can be given.

[0173] 400. Obtain risk assessment results and generate output results, thereby obtaining risk assessment reports and maps.

[0174] As described above, the embodiments of the present application collect road data and road information through vehicle equipment; process the road data and road information, comprehensively process the road data and road information, obtain various road risk factors, and perform feature encoding on various road risk factors according to set rules; input feature characteristics, conduct risk assessment on various road risk factors in combination with several set models, and calculate risk parameters of each road section; obtain risk assessment results and generate output results to obtain risk assessment reports and maps; automatically extract all factors related to road risks, greatly liberate human, financial and material resources, reduce errors, and at the same time, the assessment results are comprehensive and accurate, improving the assessment effect and efficiency.

[0175] The above steps are not to be performed in a strict order as described in the numbers, but should be understood as an overall solution.

[0176] In the second aspect, based on the above embodiments, Figure 7 This is a schematic diagram of the structure of a full-process risk assessment device based on artificial intelligence provided in an embodiment of the present application. Figure 7 The full-process risk assessment device based on artificial intelligence provided in this embodiment specifically includes: an information collection module 701, a feature encoding module 702, a risk assessment module 703 and a result output module 704.

[0177] Among them, the information collection module 701 is used to collect road data and road information through vehicle equipment; the feature encoding module 702 is used to process road data and road information, comprehensively process the road data and road information, obtain various road risk factors, and perform feature encoding on various road risk factors according to set rules; the risk assessment module 703 is used to input feature features, conduct risk assessment on various road risk factors in combination with several set models, and calculate the risk parameters of each road section; the result output module 704 is used to obtain risk assessment results and generate output results to obtain risk assessment reports and maps.

[0178] As described above, the embodiments of the present application collect road data and road information through vehicle equipment; process the road data and road information, comprehensively process the road data and road information, obtain various road risk factors, and perform feature encoding on various road risk factors according to set rules; input feature characteristics, conduct risk assessment on various road risk factors in combination with several set models, and calculate risk parameters of each road section; obtain risk assessment results and generate output results to obtain risk assessment reports and maps; automatically extract all factors related to road risks, greatly liberate human, financial and material resources, reduce errors, and at the same time, the assessment results are comprehensive and accurate, improving the assessment effect and efficiency.

[0179] The artificial intelligence-based full-process risk assessment device provided in the embodiment of the present application can be used to execute the artificial intelligence-based full-process risk assessment method provided in the above embodiment, and has corresponding functions and beneficial effects.

[0180] On the third aspect, the embodiments of the present application also provide an electronic device that can integrate the full-process risk assessment device based on artificial intelligence provided by the embodiments of the present application. Figure 8 Schematic diagram of the structure of an electronic device provided by an embodiment of the present application. Figure 8 The electronic device includes: an input device 83, an output device 84, a memory 82 and one or more processors 81; the memory 82 is used to store one or more programs; when the one or more programs are executed by the one or more processors 81, the one or more processors 81 implement the full-process risk assessment method based on artificial intelligence as provided in the above embodiment. The input device 83, the output device 84, the memory 82 and the processor 81 can be connected by a bus or other means. Figure 8 The example of connecting through bus is taken in the following.

[0181] The processor 81 executes various functional applications and data processing of the device by running the software programs, instructions and modules stored in the memory 82, thereby realizing the above-mentioned full-process risk assessment method based on artificial intelligence.

[0182] The electronic device provided above can be used to execute the full-process risk assessment method based on artificial intelligence provided in the above embodiment, and has corresponding functions and beneficial effects.

[0183] In a fourth aspect, an embodiment of the present application also provides a computer-readable storage medium, which includes a stored computer program; wherein, when the computer program is running, the device where the computer-readable storage medium is located is controlled to execute the full-process risk assessment method based on artificial intelligence as described above, and can achieve the same beneficial effects.

[0184] Of course, the storage medium containing computer-executable instructions provided in the embodiments of the present application is not limited to the full-process risk assessment method based on artificial intelligence as described above, but can also execute relevant operations in the full-process risk assessment method based on artificial intelligence provided in any embodiment of the present application.

[0185] In a fifth aspect, the embodiments of the present application also provide a computer program product, and the methods described in the various embodiments of the present application can be implemented in whole or in part by software, hardware, firmware, or any combination thereof. When implemented using software, it can be implemented in whole or in part in the form of a computer program product. The computer program product includes one or more computer programs or instructions. When the computer program or instruction is loaded and executed on a computer, the processes or functions described in the various embodiments of the present application are executed in whole or in part. The computer can be a general-purpose computer, a special-purpose computer, a computer network, a network device, a user device, a core network device, an OAM (Open Application Model) or other programmable device.

[0186] The computer program or instructions may be stored in a computer-readable storage medium, or transmitted from one computer-readable storage medium to another computer-readable storage medium, for example, the computer program or instructions may be transmitted from one website, computer, server or data center to another website, computer, server or data center by wired or wireless means. The computer-readable storage medium may be any available medium that can be accessed by a computer or a data storage device such as a server or data center that integrates one or more available media. The available medium may be a magnetic medium, such as a floppy disk, a hard disk, or a magnetic tape; it may also be an optical medium, such as a digital video disk; it may also be a semiconductor medium, such as a solid-state drive. The computer-readable storage medium may be a volatile or non-volatile storage medium, or may include both volatile and non-volatile types of storage media.

[0187] In several embodiments provided in the present application, it should be understood that the disclosed devices and methods can also be implemented in other ways. The device embodiments described above are merely schematic. For example, the flowcharts and block diagrams in the accompanying drawings show the possible architecture, functions and operations of the devices, methods and computer program products according to multiple embodiments of the present application. In this regard, each box in the flowchart or block diagram can represent a module, a program segment or a part of a code, and the module, a program segment or a part of a code contains one or more executable instructions for implementing the specified logical function. It should also be noted that in some alternative implementations, the functions marked in the box can also occur in a different order from the order marked in the accompanying drawings. For example, two consecutive boxes can actually be executed substantially in parallel, and they can sometimes be executed in the opposite order, depending on the functions involved. It should also be noted that each box in the block diagram and / or flowchart, and the combination of boxes in the block diagram and / or flowchart can be implemented with a dedicated hardware-based system that performs a specified function or action, or can be implemented with a combination of dedicated hardware and computer instructions.

[0188] In addition, the functional modules in the various embodiments of the present application may be integrated together to form an independent part, or each module may exist separately, or two or more modules may be integrated to form an independent part.

[0189] If the function is implemented in the form of a software function module and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present application can be essentially or partly embodied in the form of a software product that contributes to the prior art. The computer software product is stored in a storage medium and includes several instructions for an electronic device (which can be a personal computer, server, or network device, etc.) to perform all or part of the steps of the method described in each embodiment of the present application. The aforementioned storage medium includes: U disk, mobile hard disk, read-only memory (ROM, Read-Only Memory), random access memory (RAM, Random Access Memory), disk or optical disk, and other media that can store program codes.

[0190] The above description is only an embodiment of the present application and is not intended to limit the scope of protection of the present application. For those skilled in the art, the present application may have various changes and variations. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present application should be included in the scope of protection of the present application. It should be noted that similar reference numerals and letters represent similar items in the following drawings, so once an item is defined in one drawing, it does not need to be further defined and explained in the subsequent drawings.

[0191] The above is only a specific implementation of the present application, but the protection scope of the present application is not limited thereto. Any technician familiar with the technical field can easily think of changes or substitutions within the technical scope disclosed in the present application, which should be included in the protection scope of the present application. Therefore, the protection scope of the present application should be based on the protection scope of the claims.

[0192] It should be noted that, in the embodiments of the present application, relational terms such as first and second, etc. are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply that there is any such actual relationship or order between these entities or operations. Moreover, the terms "include", "comprise" or any other variants thereof are intended to cover non-exclusive inclusion, so that the process, method, article or equipment including a series of elements includes not only those elements, but also includes other elements that are not clearly listed, or also includes elements inherent to such process, method, article or equipment. In the absence of more restrictions, the elements limited by the sentence "comprise one..." do not exclude the existence of other identical elements in the process, method, article or equipment including the elements.

Claims

1. A full-process risk assessment method based on artificial intelligence, characterized in that: include: Collect road data and road information through vehicle equipment; Processing road data and road information, comprehensively processing the road data and road information, obtaining various road risk factors, and encoding the characteristics of various road risk factors according to set rules; Input element characteristics, combine several set models to conduct risk assessment on various road risk elements, and calculate the risk parameters of each road section; Get risk assessment results and generate output results, and obtain risk assessment reports and maps; The processing of the road data and the road information, the integrated processed road data and the road information, to obtain various road risk factors, and feature encoding of the various road risk factors according to set rules, includes: Automatically extract various elements related to road safety from road data based on algorithm models; Among them, the automatic extraction of various elements related to road safety in road data based on the algorithm model includes: Unify and integrate various road data; Obtain the outline of road safety-related elements based on the multi-task perception model and the SAM pixel-level perception model; Based on calling the tracking model and combining the radar point cloud information and the vehicle's own speed, the speed and lane of other vehicles in the driving process are calculated; Automatically generate the road stake number at every moment during driving based on longitude and latitude, lateral and longitudinal acceleration, video and video technology; Based on the linear calculation model, the radius of curves in the entire highway is calculated through the longitude and latitude information collected along the entire line; The method of automatically generating the road stake number at each moment during the driving process according to the latitude and longitude, lateral and longitudinal acceleration, video and video technology includes: Initially give a starting pile number and driving direction; Calculate the earth projection distance between the longitude and latitude of two adjacent moments, and correct it through acceleration information to obtain the vehicle's travel distance between the two moments; According to the pile number information and driving direction at the previous moment, combined with the driving distance, the pile number information at the current moment is calculated; During driving, whenever the road information collection module captures relevant information about the pile number plate, it transmits the pile number plate information to the pile number calculation part, identifies the pile number value through the OCR optical character recognition model, and calculates the distance between the pile number plate and the vehicle through the radar; Check the current pile number information and make corrections.

2. The full-process risk assessment method based on artificial intelligence according to claim 1 is characterized in that: The collecting of road data and road information includes: Comprehensive collection of road data through vehicle-mounted equipment; Collect road information; road information includes but is not limited to accident records, weather records, design drawings, completion drawings, and inspection data.

3. The full-process risk assessment method based on artificial intelligence according to claim 1 is characterized in that: The road data and road information are processed, and the processed road data and road information are integrated to obtain various road risk factors, and feature coding of various road risk factors is performed according to set rules, including: According to the road data, search, crawl, and extract various network-related data corresponding to the road data; Documentation of road information; After comprehensive processing, the road data and road information are obtained to obtain various road risk factors; The various road risk factors are divided into feature codes based on the set mileage and converted into vector expressions.

4. The full-process risk assessment method based on artificial intelligence according to claim 1 is characterized in that: The unified fusion of various road data includes: According to the timestamp assigned by the clock module, the point cloud takes the point cloud information captured from the current timestamp point cloud to the next video frame as the basic unit, and associates the video frame with the radar point cloud; Bilinear interpolation is used to interpolate the latitude and longitude information and acceleration information for each frame of video; The relationship between the two-dimensional image and the three-dimensional point cloud is associated through the homography matrix, and the correspondence between the video pixels and the radar point cloud is associated.

5. The full-process risk assessment method based on artificial intelligence according to claim 1 is characterized in that: The method of acquiring the contours of road safety-related elements based on the multi-task perception model and the SAM pixel-level perception model includes: Through the multi-task perception model, various road safety-related elements are obtained, and the elements are expressed in the form of rectangular boxes or dotted lines, and the contours of road safety-related elements are obtained by combining the SAM model; Through morphological features and mapping elements from two-dimensional to three-dimensional, risk factors are obtained, among which risk factors include: guardrail height, sign layout size and sign occlusion.

6. The full-process risk assessment method based on artificial intelligence according to claim 1 is characterized in that: The method of invoking the tracking model and combining the radar point cloud information and the vehicle's own speed to infer the speed and lane of other vehicles during driving includes: The tracking model is used to track vehicles on the road and distinguish vehicle types; Combined with road data, determine the lane each tracked vehicle is in; Map each tracked vehicle to the point cloud, obtain the position of the vehicle at each moment from the acquisition device, and calculate the vehicle speed by combining the change in the distance of the tracked vehicle relative to the vehicle within the set time and the speed of the vehicle; Get the type, lane and speed of all vehicles.

7. The full-process risk assessment method based on artificial intelligence according to claim 1 is characterized in that: The linear calculation model is based on which the radius of the curve existing in the entire highway is calculated through the longitude and latitude information collected along the entire highway, including: Through the longitude and latitude sequence, a sliding window with actual distances of the first set distance, the second set distance and the third set distance is used to perform sliding calculations in both the uplink and downlink directions of the entire line; Among them, a convex curve is fitted according to the GPS coordinates in each window, and the radius of the curve is calculated by taking the set key points; The full-line radius set obtained by setting sliding windows is fused to obtain the full-line result.

8. A full-process risk assessment device based on artificial intelligence, characterized in that: include: An information collection module is used to collect road data and road information through vehicle equipment; A feature coding module is used to process road data and road information, comprehensively process the processed road data and road information, obtain various road risk factors, and perform feature coding on various road risk factors according to set rules; The risk assessment module is used to input the element characteristics, combine several set models to conduct risk assessment on various road risk elements, and calculate the risk parameters of each road section; A result output module is used to obtain risk assessment results and generate output results to obtain risk assessment reports and maps; The processing of the road data and the road information, the integrated processed road data and the road information, to obtain various road risk factors, and feature encoding of the various road risk factors according to set rules, includes: Automatically extract various elements related to road safety from road data based on algorithm models; Among them, the automatic extraction of various elements related to road safety in road data based on the algorithm model includes: Unify and integrate various road data; Obtain the outline of road safety-related elements based on the multi-task perception model and the SAM pixel-level perception model; Based on calling the tracking model and combining the radar point cloud information and the vehicle's own speed, the speed and lane of other vehicles in the driving process are calculated; Automatically generate the road stake number at every moment during driving based on longitude and latitude, lateral and longitudinal acceleration, video and video technology; Based on the linear calculation model, the radius of curves in the entire highway is calculated through the longitude and latitude information collected along the entire line; The method of automatically generating the road stake number at each moment during the driving process according to the latitude and longitude, lateral and longitudinal acceleration, video and video technology includes: Initially give a starting pile number and driving direction; Calculate the earth projection distance between the longitude and latitude of two adjacent moments, and correct it through acceleration information to obtain the vehicle's travel distance between the two moments; According to the pile number information and driving direction at the previous moment, combined with the driving distance, the pile number information at the current moment is calculated; During driving, whenever the road information collection module captures relevant information about the pile number plate, it transmits the pile number plate information to the pile number calculation part, identifies the pile number value through the OCR optical character recognition model, and calculates the distance between the pile number plate and the vehicle through the radar; Check the current pile number information and make corrections.

9. An electronic device, characterized in that: include: A processor, a memory and a bus, wherein the processor is connected to the memory via the bus, and the memory stores computer-readable instructions. When the computer-readable instructions are executed by the processor, they are used to implement the full-process risk assessment method based on artificial intelligence as described in any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores a computer program, which, when executed by a server, implements the full-process risk assessment method based on artificial intelligence as described in any one of claims 1 to 7.

11. A computer program product, characterized in that The computer program product comprises instructions which, when executed by a computer, cause the computer to implement the method according to any one of claims 1-7.

Images