Sun glare warning model construction system and method for road driving, sun glare warning system and method

Abstract

The present invention discloses a system and method for constructing a solar glare warning model for road driving, and a solar glare warning system and method. The solar glare warning model construction system includes a vehicle information acquisition module configured to acquire vehicle information; a road information acquisition module configured to acquire road geometric feature information based on geographic location information; a solar geometric feature determination module configured to calculate the spatial position information of the sun; a perception acquisition module configured to acquire hierarchical perception labels; a multidimensional feature generation module configured to generate multidimensional features of glare samples; and a training module configured to train a preset glare risk prediction decision tree model based on the multidimensional features of the glare samples and the corresponding hierarchical perception labels. The trained glare risk prediction decision tree model is then used as a solar glare warning model, enabling early prediction of glare risks at future vehicle trajectory nodes, significantly improving the accuracy of driving safety warnings.

Description

Technical Field

[0001] The present invention relates to the field of traffic safety control, and in particular to a system and method for constructing a sun glare warning model for road driving, and a sun glare warning system and method. Background Art

[0002] In the field of road traffic safety, the impact of environmental factors on driving safety has always been an important research direction. In the existing technology, studies have proposed various solutions for identifying various safety risks during road driving.

[0003] CN112015843B provides a driving risk situation assessment method and system based on the results of multi-vehicle intention interaction, including: step M1: obtaining the results of multi-vehicle intention interaction based on incomplete information dynamic game; step M2: establishing a random environment model based on the multi-vehicle intention interaction results and the parameters of current traffic elements and predicted parameters of future traffic elements; step M3: establishing a long- and short-time domain vehicle trajectory prediction model by fusing the kinematic model with the driving behavior cognitive model, and realizing comprehensive prediction of vehicle trajectory in the long and short time domains; step M4: performing collision probability analysis based on the random environment model and the vehicle trajectory prediction results; step M5: performing collision risk assessment based on the collision probability analysis, thereby building a driving risk situation assessment model.

[0004] WO2022233099A1 provides a method for investigating the spatiotemporal characteristics of road traffic violations based on networked ADAS, including: obtaining traffic violation information of forward vehicles perceived by all networked ADAS vehicles in the urban road traffic system during driving; encoding the traffic violation information and packaging it into structured data; uploading the structured data to a traffic data cloud platform, matching traffic violations on sections of roads in a selected area based on time information and latitude and longitude information, and outputting a spatiotemporal distribution map of traffic violations in each time period; performing statistical analysis on the spatiotemporal status of road traffic violations to obtain the frequency of traffic violations in each section; and performing cluster analysis on the spatiotemporal distribution status of road traffic violations to obtain the clustered sections and time periods of traffic violations.

[0005] However, there is currently no effective technical solution that can accurately predict the risk of sun glare while driving on the road, which makes it difficult to meet the actual needs of safety warnings in complex lighting environments in real driving scenarios, especially intelligent driving scenarios.

[0006] The content of this background technology description is only for facilitating understanding of the relevant technology in this field and is not regarded as an admission of the prior art. Summary of the Invention

[0007] Accordingly, the present invention provides a system and method for constructing a sun glare warning model for road driving, and a sun glare warning system and method, which at least partially solve the above technical problems.

[0008] In a first aspect, a system for constructing a sun glare warning model for road driving is provided, which may include:

[0009] A vehicle information acquisition module is configured to acquire vehicle information, wherein the vehicle information includes synchronously acquired geographic location information, driving information, and corresponding synchronization time information of the vehicle;

[0010] a road information acquisition module configured to acquire road geometric feature information corresponding to a road position based on the geographic location information;

[0011] a solar geometric feature determination module configured to calculate the solar spatial position information at the corresponding synchronization time according to a preset solar position calculation algorithm (SPA) based on the geographical location information and the synchronization time information;

[0012] a perception acquisition module configured to acquire a graded perception label of the driver's perception of sun glare at a corresponding synchronization time;

[0013] a multi-dimensional feature generation module configured to generate multi-dimensional features of glare samples based on the driving information, synchronization time information, road geometric feature information and sun spatial position information;

[0014] The training module is configured to train a preset glare risk prediction decision tree model based on the multidimensional features of the glare samples and the corresponding hierarchical perception labels, and use the trained glare risk prediction decision tree model as a solar glare warning model.

[0015] In a second aspect, a system for constructing a sun glare warning model for road driving is provided, which may include:

[0016] A vehicle information acquisition module is configured to acquire vehicle information, wherein the vehicle information includes synchronously acquired geographic location information, driving information, and corresponding synchronization time information of the vehicle;

[0017] a road information acquisition module configured to acquire road geometric feature information corresponding to a road position based on the geographic location information;

[0018] a perception acquisition module configured to acquire a graded perception label of the driver's perception of sun glare at a corresponding synchronization time;

[0019] a multi-dimensional feature generation module configured to generate multi-dimensional features of the glare sample based on the geographic location information, driving information, synchronization time information and road geometric feature information;

[0020] The end-to-end training module is configured to train an end-to-end integrated model based on the multidimensional features of the glare samples and the corresponding hierarchical perception labels, and use the trained end-to-end integrated model as a solar glare warning model. The end-to-end integrated model integrates a preset sun position calculation algorithm model and a glare risk prediction decision tree model.

[0021] In some embodiments, the vehicle information acquisition module is configured to acquire the vehicle information based on a daytime cycle with different sample collection density configurations in a first collection period corresponding to a preset time condition and a second collection period not corresponding to the preset time condition, wherein the sample collection density of the first collection period is higher than the sample collection density of the second collection period.

[0022] In some embodiments, the multidimensional feature generation module includes a sample balancing submodule, which is configured to perform data amplification processing on the samples collected in the second collection period based on the difference in sample collection density between the first collection period and the second collection period to compensate for the difference in sample collection density between the first collection period and the second collection period.

[0023] In some embodiments, the glare samples used for training are glare samples collected at the different collection densities and subjected to compensation processing.

[0024] In some embodiments, the vehicle information acquisition module is configured to acquire vehicle information based on a preset spatial sampling interval; the road information acquisition module and the perception acquisition module are configured to synchronously acquire the road geometric feature information and the driver's glare perception information based on the same spatial sampling interval.

[0025] In some embodiments, the sun glare warning model construction system further includes: an eye movement detector configured to detect the driver's eye movement data at corresponding synchronous times, wherein the eye movement data includes pupil changes and / or line of sight deviation.

[0026] In some embodiments, the system further comprises a label generation module configured to construct a dual perception label from the eye movement data and the hierarchical perception label;

[0027] The training module is configured to train a preset glare risk prediction decision tree model based on the multi-dimensional glare sample features and the dual perception labels.

[0028] In a third aspect, a sun glare warning system for road driving is provided, which may include:

[0029] a trajectory node determination module configured to determine future trajectory nodes of a user's vehicle;

[0030] a solar glare warning model constructed by the solar glare warning model construction system according to the first or second aspect, wherein the solar glare warning model is configured to output a glare risk prediction result for the future trajectory node;

[0031] The early warning control module is configured to selectively generate an early warning signal according to the glare risk prediction result.

[0032] In some embodiments, the sun glare warning system further comprises:

[0033] A filter is configured to identify a glare-risk-free road section and filter the future trajectory node and / or the glare risk prediction result based on the determined glare-risk-free road section, wherein the glare-risk-free road section is determined based on a tunnel section of the road.

[0034] In a fourth aspect, a method for constructing a sun glare warning model for road driving is provided, which may include:

[0035] Acquiring vehicle information, the vehicle information including synchronously acquired vehicle geographic location information, driving information, and corresponding synchronization time information;

[0036] Based on the geographic location information, obtaining road geometric feature information corresponding to the road location;

[0037] Based on the geographic location information and the synchronization time information, calculating the solar spatial position information at the corresponding synchronization time according to a preset solar position calculation algorithm (Solar Position Algorithm, SPA);

[0038] Obtaining the graded perception labels of the driver's sun glare perception at the corresponding synchronization time;

[0039] generating a multi-dimensional feature of the glare sample based on the driving information, synchronization time information, road geometric feature information and sun spatial position information;

[0040] A preset glare risk prediction decision tree model is trained based on the multidimensional features of the glare samples and the corresponding hierarchical perception labels, and the trained glare risk prediction decision tree model is used as a solar glare warning model.

[0041] In a fifth aspect, a method for constructing a sun glare warning model for road driving is provided, which may include:

[0042] Acquiring vehicle information, the vehicle information including synchronously acquired vehicle geographic location information, driving information, and corresponding synchronization time information;

[0043] Based on the geographic location information, obtaining road geometric feature information corresponding to the road location;

[0044] Obtaining the graded perception labels of the driver's sun glare perception at the corresponding synchronization time;

[0045] generating multi-dimensional features of the glare sample based on the geographic location information, driving information, synchronization time information and road geometric feature information;

[0046] An end-to-end integrated model is trained based on the multidimensional features of the glare samples and the corresponding hierarchical perception labels, and the trained end-to-end integrated model is used as a solar glare warning model. The end-to-end integrated model integrates a preset sun position calculation algorithm model and a glare risk prediction decision tree model.

[0047] In a sixth aspect, a method for warning sun glare during road driving is provided, which may include:

[0048] Determine the future trajectory nodes of the user's vehicle;

[0049] Outputting a glare risk prediction result of the future trajectory node using the solar glare warning model constructed by the solar glare warning model construction method according to the fourth or fifth aspect;

[0050] A warning signal is selectively generated according to the glare risk prediction result.

[0051] An embodiment of the present invention provides a system for constructing a sun glare warning model for road driving, comprising: a vehicle information acquisition module configured to acquire vehicle information, including simultaneously acquired vehicle geographic location information, driving information, and corresponding synchronization time information; a road information acquisition module configured to acquire road geometry information corresponding to the road location based on the geographic location information; a solar geometry determination module configured to calculate the solar spatial position information at the corresponding synchronization time based on the geographic location information and synchronization time information according to a preset solar position calculation algorithm; a perception acquisition module configured to acquire a hierarchical perception label of the driver's solar glare perception at the corresponding synchronization time; a multidimensional feature generation module configured to generate multidimensional features of glare samples based on the driving information, synchronization time information, road geometry information, and solar spatial position information; and a training module configured to train a preset glare risk prediction decision tree model based on the multidimensional features of the glare samples and the corresponding hierarchical perception labels, and use the trained glare risk prediction decision tree model as a sun glare warning model. The present invention achieves accurate and advance prediction of glare risks at future trajectory nodes during road driving.

[0052] Accordingly, the solution of the embodiment of the present invention achieves accurate prediction of the glare risk of future trajectory nodes through multi-dimensional data fusion and machine learning modeling.

[0053] Furthermore, the solution of the embodiments of the present invention achieves the advantages of advanced prediction of road glare risks during deployment, and optimized integration with navigation systems and / or intelligent driving systems. Specifically, the solution of the embodiments of the present invention has the ability to predict solar glare in advance based on driving trajectories, pre-identifying risks before vehicles reach potentially glare sections and issuing warnings through the navigation system or intelligent driving system. This allows the technical solution of the embodiments of the present invention to be seamlessly integrated with existing navigation or intelligent driving ecosystems during deployment.

[0054] Furthermore, the solution of this embodiment integrates multi-dimensional features (including geography, time, solar geometry, and vehicle status) during training, and uses a cross-validation strategy based on driver labels to evaluate model generalization. This statistical learning based on a large amount of real-world driving scene data can handle the uncertainty of glare prediction in complex environments, enabling scalable application to a wider range of road networks based on training data collected from limited road sections and time periods. In other words, the roads, such as highways, used for training can be at least partially different from the roads, such as highways, used for early warning (risk prediction) purposes. Preferably, the roads, such as highways, used for early warning (risk prediction) purposes are larger than the roads, such as highways, used for training purposes. For example, a model trained on one or more highways can be directly applied to the entire regional highway network; a model trained on data collected on a specific road during a certain time period can be extended to apply to that road throughout the entire time period; and a model trained on a local section of a road can cover the entire road.

[0055] Furthermore, while the solution in this embodiment of the present invention fully integrates multi-dimensional features (including geography, time, solar geometry, and vehicle status) during training and uses a cross-validation strategy based on driver labels to evaluate model generalization, during deployment, the user only needs to provide basic information, namely navigation location. All complex calculations of solar position, road geometry parameters, and environmental factors are handled internally by the pre-trained model. Specifically, navigation service providers or intelligent driving providers can deploy the trained model and high-precision map data in the cloud. Simply obtaining the user's real-time location information can generate a glare risk warning and push it to the user's terminal, such as a navigation terminal or intelligent driving vehicle terminal. This significantly simplifies the hardware and computing requirements on the user side, particularly avoiding the need to deploy a large number of related sensors or detection devices in the user's vehicle.

[0056] Other optional features and technical effects of the embodiments of the present invention are partially described below, and partially can be understood by reading this document. BRIEF DESCRIPTION OF THE DRAWINGS

[0057] Hereinafter, embodiments of the present invention will be described in detail with reference to the accompanying drawings. The elements shown are not limited to the scale shown in the drawings. The same or similar reference numerals in the drawings represent the same or similar elements, wherein:

[0058] Figure 1 A module diagram of a system for constructing a sun glare warning model for road driving according to an embodiment of the present invention is shown;

[0059] Figure 2 A module diagram of a system for constructing a sun glare warning model for road driving according to an embodiment of the present invention is shown;

[0060] Figure 3 A module diagram of a sun glare warning system for road driving according to an embodiment of the present invention is shown;

[0061] Figure 4 A schematic diagram schematically illustrates the future trajectory nodes of the user's vehicle;

[0062] Figure 5 Schematically illustrates a sun glare risk warning signal provided in a navigation terminal or an intelligent driving terminal;

[0063] Figure 6 A flow chart showing a method for constructing a sun glare warning model for road driving according to an embodiment of the present invention is shown;

[0064] Figure 7 A flow chart showing a method for constructing a sun glare warning model for road driving according to an embodiment of the present invention is shown;

[0065] Figure 8 A flow chart of a sun glare warning method for road driving according to an embodiment of the present invention is shown;

[0066] Figure 9 The figure shows a structural diagram of an electronic device for implementing the method according to the embodiment of the present invention. DETAILED DESCRIPTION

[0067] In order to make the purpose, technical solutions and advantages of the present invention more clearly understood, the present invention will be further described in detail below in conjunction with specific embodiments and accompanying drawings. Here, the exemplary embodiments of the present invention and their descriptions are used to explain the present invention, but are not intended to limit the present invention.

[0068] Exemplary embodiments of the present invention will be described in more detail below with reference to the accompanying drawings. Although exemplary embodiments of the present invention are shown in the accompanying drawings, it should be understood that the present invention can be implemented in various forms and should not be limited by the embodiments set forth herein. Rather, these embodiments are provided to enable a more thorough understanding of the present invention and to fully convey the scope of the present invention to those skilled in the art.

[0069] As mentioned above, there are many solutions in the prior art for identifying various safety risks during road driving. However, there is currently no effective technical solution that can accurately predict the risk of sun glare during road driving.

[0070] As a significant environmental factor affecting driving safety, sun glare presents unique technical challenges in its prediction. During highway driving, sun glare is a significant environmental contributor to traffic accidents, particularly during periods of low solar altitude, such as sunrise and sunset. Intense sunlight enters the driver's field of view at low angles, causing visual obstruction, delayed perception, and even temporary blindness. The occurrence of sun glare exhibits distinct temporal periodicity and spatial localization, with its severity influenced by a combination of factors, including the sun's astronomical position, geographic latitude and longitude, road orientation, vehicle posture, and terrain slope.

[0071] Currently, it is difficult to meet the actual needs of safety warnings in complex lighting environments in real driving scenarios, especially intelligent driving scenarios. In particular, it is difficult to accurately predict the risk of sun glare in real road driving scenarios such as highways.

[0072] Some potential solutions might determine the degree of sun glare based on real-time geometric calculations of the relative positions of the sun and the vehicle. However, such solutions require high-precision and complex real-time processing of multiple geometric parameters, and may also require extensive computationally intensive operations such as trigonometric calculations, coordinate system transformations, and multi-parameter coupling analysis. Furthermore, in real-world applications, due to environmental interference, sensor accuracy limitations, data synchronization constraints, and a lack of comprehensive consideration of individual driver differences and complex road environments, the results are often ineffective and inaccurate (with high false positive rates), particularly due to the coupled calculations of parameters spanning multiple spatial scales, from the astronomical sun position to the local road geometry and microscopic vehicle posture changes. This makes it difficult to meet the engineering deployment requirements of real-world scenarios. Furthermore, such real-time geometric calculations inherently only assess the current state, have inherent lags, and lack forward-looking predictive capabilities. Consequently, they cannot provide early warning of glare risks along the future travel path, and therefore cannot effectively reduce or eliminate the impact of sun glare on driving safety, especially during high-speed highway driving. Furthermore, existing solutions based on real-time geometric calculations face the engineering implementation challenge of the complex configuration of on-board sensor systems. To obtain all the parameters required for real-time geometric calculations, each vehicle typically requires multiple high-precision sensors. This complex sensor configuration not only significantly increases the hardware cost and system power consumption of each vehicle, but also introduces a series of engineering challenges, such as complex sensor calibration, heavy computational burdens for multi-sensor real-time data fusion, and difficulty ensuring long-term reliability. In summary, existing technical solutions based on real-time geometric calculations suffer from significant deficiencies in several key technical indicators, including computational complexity, prediction accuracy, system cost, and engineering reliability, making them difficult to meet the engineering deployment requirements of actual in-vehicle applications.

[0073] To this end, embodiments of the present invention provide a system and method for constructing a sun glare warning model for road driving, as well as a corresponding sun glare warning system for road driving. By implementing multidimensional data fusion modeling and a cross-validation strategy based on driver labels, the model's generalization capability is enhanced to achieve accurate prediction and proactive warning of sun glare risks in real-world road (especially highway) driving scenarios.

[0074] like Figure 1 As shown, the solar glare warning model construction system 100 provided by an embodiment of the present invention includes: a vehicle information acquisition module 110, a road information acquisition module 120, a solar geometric feature determination module 130, a perception acquisition module 140, a multi-dimensional feature generation module 150 and a training module 160.

[0075] like Figure 1As shown, the vehicle information acquisition module 110, the road information acquisition module 120, the solar geometry feature determination module 130 and the perception acquisition module 140 can be arranged on the vehicle side; the multi-dimensional feature generation module 150 and the training module 160 can be arranged on a cloud server or a data processing center. However, the road information acquisition module 120 and / or the solar geometry feature determination module 130 can optionally be arranged on the vehicle side or in the cloud, and the present invention does not limit this. The trained model can be deployed to a cloud server, and a corresponding interface can be provided on an on-board terminal such as a navigation terminal or an intelligent driving terminal to achieve early warning of solar glare risk prediction, as further described below. However, it is conceivable that the trained model can also be deployed on an on-board terminal.

[0076] In an embodiment of the present invention, the vehicle information acquisition module 110 may be configured to acquire vehicle information, where the vehicle information includes synchronously acquired geographic location information, driving information, and corresponding synchronization time information of the vehicle.

[0077] In some embodiments, the vehicle information acquisition module 110 may include a multi-mode GNSS receiver or use a multi-mode GNSS receiver to obtain the precise geographic coordinates of the vehicle. In some exemplary embodiments, the geographic location information includes the longitude and latitude of the vehicle. Optionally, the accuracy can reach 6 decimal places (approximately 1 meter level accuracy). In a specific embodiment, the multi-mode GNSS receiver can simultaneously receive signals from multiple satellite navigation systems such as GPS, GLONASS, Galileo, and Beidou, and improve positioning accuracy and reliability through a multi-system signal fusion algorithm. In a specific embodiment, the positioning update frequency is set to 10 Hz to ensure that the vehicle can still obtain continuous and stable position information even when driving at high speeds (e.g., 120 km / h).

[0078] As an optional embodiment, the vehicle information acquisition module 110 may also include an IMU inertial measurement unit (IMU) configured to obtain vehicle heading and attitude angle information. The IMU can, for example, measure the vehicle's linear acceleration and angular velocity using a three-axis accelerometer and a three-axis gyroscope, and fuse these information using a Kalman filter algorithm to obtain accurate vehicle attitude information. Optionally, when the vehicle passes through areas with weak GPS signals, such as tunnels, under overpasses, or in urban canyons, the vehicle information acquisition module 110 can employ an inertial navigation compensation algorithm to use IMU data for position estimation; after signal recovery, the accumulated error is automatically corrected to restore high-precision positioning.

[0079] In some embodiments, the driving information includes the speed and heading of the vehicle, such as the vehicle heading angle. In some embodiments, the vehicle information acquisition module 110 obtains the real-time driving status information of the vehicle through the on-board sensor network. In a specific embodiment, the vehicle speed is obtained by calculating the GPS Doppler effect. In a specific embodiment, the vehicle speed can also be obtained from the vehicle CAN bus as an alternative or supplement, for example, for cross-validation. The vehicle heading angle represents the angle of the vehicle's forward direction, and can be calculated, for example, by fusion of GPS heading and inertial sensors, preferably with due north as the reference of 0° and increasing clockwise to 360°.

[0080] In some embodiments, the vehicle information acquisition module 110 may also obtain the vehicle's pitch angle information through an IMU (Inertial Measurement Unit). The pitch angle reflects the vehicle's forward and backward tilt, taking a positive value when going uphill and a negative value when going downhill. In some embodiments, the pitch angle may be determined based on geographic location in conjunction with a high-precision map as described below. In some optional embodiments, the vehicle information acquisition module 110 may also obtain the vehicle's roll angle information through an IMU (Inertial Measurement Unit). The roll angle reflects the vehicle's left and right tilt, taking a positive value when leaning to the right and a negative value when leaning to the left.

[0081] In some embodiments, the synchronized time information includes year, month, day, and hour. Thus, the synchronized time information acquired by the vehicle information acquisition module 110 ensures that all data acquired on the vehicle side are accurately timestamped, and thereby enables the sample data to be time-synchronized, providing a reliable data basis for subsequent multi-dimensional feature fusion and model training. Preferably, to ensure the accuracy of data synchronization, the vehicle information acquisition module 110 can use a unified time base to timestamp all data acquired on the vehicle side, such as sensor data. For example, the synchronized time information can provide an accurate time base for calculating the position of the sun. Optionally, the vehicle information acquisition module 110 acquires UTC time through GPS timing. Optionally, time zone conversion can be automatically performed to obtain local time information.

[0082] In a preferred embodiment of the present invention, the vehicle information acquisition module 110 is configured to acquire vehicle information based on a diurnal cycle in a first acquisition period corresponding to a preset time condition and a second acquisition period not corresponding to the preset time condition with different sample acquisition density configurations. In an embodiment of the present invention, the sample acquisition density of the first acquisition period is greater than that of the second acquisition period. In an embodiment of the present invention, different sample acquisition densities based on time periods or (larger) or (smaller) sample acquisition densities are broadly interpreted, as long as the average number of samples collected per unit time is different, (larger) or (smaller). In a specific example, the first acquisition period includes 1 hour before and after sunrise and 1 hour before and after sunset, which can be regarded as a high glare risk period, for example. In a specific example, the second acquisition period is other time periods. Other data acquisition modules (vehicle-side modules) also have different sample acquisition densities accordingly.

[0083] In a preferred embodiment of the present invention, the vehicle information acquisition module 110 is configured to acquire vehicle information based on a preset spatial sampling interval. In one specific example, the spatial sampling interval can be adaptively set based on the road type, for example, sampling is performed every 10 meters on a highway section. As previously described, in the embodiments of the present invention, different sample collection densities based on time periods are broadly interpreted to encompass instances where different collection densities exist but the same sampling interval. For example, different collection densities can be used in the first collection period and the second collection period, but the sampling interval can remain the same, for example, every 10 meters.

[0084] In an embodiment of the present invention, the road information acquisition module 120 is configured to acquire road geometric feature information corresponding to the road location based on the geographic location information. In an embodiment of the present invention, the road information acquisition module 120 works closely with the vehicle information acquisition module 110 to query corresponding road parameters based on the vehicle's real-time location information, providing road geometric environment information for glare risk calculation. In a preferred embodiment, the road information acquisition module 120 and the vehicle information acquisition module 110 can be integrated into a single unit, or they can be partially integrated, such as their associated submodules are physically partially integrated, which falls within the scope of the present invention. For example, in some embodiments, the vehicle information acquisition module 110 and the road information acquisition module 120 can be integrated and implemented as a unified data acquisition module. In other embodiments, the vehicle information acquisition module 110 and the road information acquisition module 120 can also be implemented separately.

[0085] In some embodiments, the road information acquisition module 120 may include a high-precision map module or a high-precision map module interface, or be connected to a data interface of a high-precision map service provider through an on-board communication module, which is not limited by the present invention.

[0086] In an embodiment of the present invention, road geometric feature information includes road direction, slope, and curvature. By way of example and not limitation, the road information acquisition module 120 extracts the following parameters from the high-precision map:

[0087] Road strike angle: This indicates the direction of the road, with north as the base and increasing clockwise. For example, a direction vector can be constructed by taking the road centerline points 50 meters ahead and behind the current vehicle position, and then converted to an azimuth. The road strike angle directly affects the relative position of the vehicle and the sun and is a key parameter in glare risk assessment.

[0088] Road Slope Angle: This indicates the longitudinal inclination of the road, with positive values ​​for uphill slopes, negative values ​​for downhill slopes, and 0° for flat roads. For example, this can be calculated using the inverse tangent function based on the elevation difference within a given distance. This parameter is used to correct the solar incidence angle calculation and is important for glare prediction on mountainous roads.

[0089] Road curvature radius: This parameter indicates the severity of a road's curves. The curvature radius of a straight road tends to infinity, while the curvature radius of a sharp curve is relatively small. For example, the curvature radius can be derived from the coordinates of several consecutive (e.g., three) GPS points through geometric relationships. This parameter affects the driver's line of sight, necessitating more frequent glare risk assessments on curved roads.

[0090] As mentioned above, the road information acquisition module 120 may be configured to synchronously acquire road geometric feature information based on the same spatial sampling interval as the vehicle information acquisition module 110 to ensure a spatiotemporal correspondence with the vehicle information.

[0091] In an embodiment of the present invention, the solar geometry determination module 130 is configured to calculate the solar spatial position information at the corresponding synchronization time according to a preset solar position calculation algorithm (SPA) based on the geographic location information and the synchronization time information.

[0092] In a preferred embodiment, the solar geometry determination module 130 utilizes the high-precision Solar Position Algorithm (SPA) developed by the National Renewable Energy Laboratory (NREL). The SPA algorithm is a highly accurate standard for calculating solar position, with a calculation accuracy of ±0.0003°. In one specific example, the solar geometry determination module 130 is configured to implement the following process based on the SPA algorithm: First, the UTC time provided by the vehicle information acquisition module 110 is converted to Julian day number, a standard time reference in astronomical calculations that counts consecutive days since January 1, 4713 BC. Then, based on the geographic coordinates (longitude and latitude) of the observation point and the Julian day number, the Earth's position parameters in orbit are determined, including astronomical parameters such as geocentric longitude and solar declination. Furthermore, the solar altitude and azimuth angles at the vehicle's latitude on the current date are determined. In a specific application example, when the vehicle is at 40° north latitude and 116° east longitude, at 18:30 on June 21, 2024, the solar geometry feature determination module 130 calculates that the solar altitude angle is approximately 28.5° and the solar azimuth angle is approximately 284.7°, indicating that the sun is located in the northwest-west direction and the altitude angle is moderate. The glare risk can be further judged in combination with the vehicle's driving direction.

[0093] In a preferred embodiment of the present invention, the solar altitude angle and azimuth angle will be used as direct factors together with driving information such as vehicle speed and heading, and road geometric feature information such as road direction, slope and curvature to perform sample multidimensional characterization as described below for training.

[0094] In certain embodiments, the solar geometry determination module 130 may also include a relative orientation determination function that determines the vehicle's relative orientation information relative to the sun based on driving information, road geometry information, and solar spatial position information. The relative orientation information includes a relative angle and a slope angle of incidence. The relative angle determination function determines the relative angle between the solar azimuth and the vehicle's forward direction based on the vehicle's travel direction, vehicle heading angle, and solar azimuth. The relative angle is defined as the minimum angular difference between the vehicle's heading angle and the solar azimuth. In this embodiment, the relative angle can be normalized to a preset angle range of 0°-90° by dividing by 90°. The slope angle of incidence determination function determines the slope angle of incidence at the vehicle's road location based on the vehicle's travel direction, road slope, and solar angle of incidence. In this embodiment, the relative angle and / or slope angle of incidence can be used as a supplement to or alternative to the vehicle heading angle and / or road slope as factors used in the multidimensional characterization of samples as described below.

[0095] In an embodiment of the present invention, the perception acquisition module 140 is configured to acquire a graded perception label of the (experimental) driver's perception of solar glare at the corresponding synchronization time. In this embodiment of the present invention, the graded perception label may include multiple glare levels, such as, but not limited to, three levels: no glare or permissible glare, interference glare, and disabling glare. No glare or permissible glare corresponds to a label value of 0. No glare indicates that forward vision is essentially unaffected by light. Permissible glare indicates a slight glare, but forward vision is essentially unaffected by light, but the driver can observe the road and traffic conditions ahead normally without taking special protective measures. Interference glare corresponds to a label value of 0.5, indicating glare that the driver wants to avoid, affecting driving comfort but not seriously affecting safe driving. Disabling glare corresponds to a label value of 1, indicating severe glare that makes it difficult to open the eyes. In some embodiments, more or fewer label levels may be used, and / or different label values ​​may be used. In other embodiments, a score may be used instead.

[0096] In this embodiment of the present invention, the perception acquisition module 140 collects data using a subjective annotation method, based on the driver's actual perception. During data collection, the driver can use an onboard annotation device (such as a labeling button or voice command) to mark their current glare perception level in real time. To ensure accuracy and consistency in annotation, standardized driver training can be conducted prior to data collection, demonstrating typical scenarios and descriptions of different glare levels.

[0097] In some embodiments, perception acquisition can similarly be performed based on synchronized time and / or spatial intervals. In this case, the test driver can be prompted to label their glare experience. In another embodiment, perception acquisition can also utilize an event-triggered mechanism, proactively labeling the driver's glare experience when the driver experiences it. This information is then associated with corresponding sample data (such as vehicle information, road information, and solar spatial position information (if any)) based on the synchronized time information. For example, if a glare event is triggered by the test driver within a time interval and / or spatial interval (e.g., 10 meters) before and / or after the synchronized time and / or spatial position corresponding to a sample, the highest glare level is recorded. If no glare event is triggered, the data is recorded as no glare. As a supplement or alternative to the above embodiment, it is also conceivable that when the driver labels their glare experience, the sample data (such as vehicle information, road information, and solar spatial position information (if any)) at the time of the labeling, as well as the time of the labeling, are triggered to be acquired. In this case, the time of the labeling is the synchronized time.

[0098] In some optional embodiments, the system 100 may also include an eye movement detector 170, which is configured to detect the driver's eye movement data at corresponding synchronous times, wherein the eye movement data includes pupil changes and / or line of sight deviation. The eye movement data provides objective physiological indicator verification for the driver's glare perception. More specifically, the detection parameters of the eye movement detector include pupil diameter change rate, gaze point coordinate offset, and blink frequency change. The pupil diameter change rate is obtained by continuously measuring the pupil size and calculating the change rate. Under normal lighting, the pupil diameter is within a preset range and may shrink when encountering strong light. A rate of change per unit time exceeding a preset threshold may indicate that a strong light stimulus has been encountered; the gaze point coordinate offset reflects the change in the eyeball's gaze direction. During normal driving, the gaze point is concentrated on the road area ahead. When affected by glare, the gaze point will deviate from the center area.

[0099] In an optional embodiment including an eye movement detector, system 100 may also include a label generation module (not shown) that can be configured to construct dual-perception labels from eye movement data and hierarchical perceptual labels. This dual-perception labeling system combines subjective perceptual annotations with objective physiological indicators, providing more accurate and reliable training labels. For example, the label generation module can combine subjective annotations and objective detection results using a weighted fusion algorithm, such as a weighted average based on a preset weight ratio.

[0100] In various embodiments of the present invention, system 100 may further include one or more optional modules or components. In further embodiments, system 100 may further include a brightness acquisition module. The brightness acquisition module may, for example, include an inductive brightness acquisition device, such as an onboard camera or other brightness sensor, configured to acquire camera brightness values ​​as information about the ambient light intensity at that time, i.e., sunlight illumination. In another embodiment, the brightness acquisition module is configured to acquire local sunlight illumination information from an external data service based on geographic location information and synchronized time information.

[0101] Data Acquisition Experimental Protocol Example

[0102] In a specific embodiment, based on the aforementioned vehicle information acquisition module 110, road information acquisition module 120, solar geometry determination module 130, and perception acquisition module 140, an experimental protocol for data collection during vehicle driving can be implemented to obtain high-quality training sample data. This experimental protocol aims to collect glare sample data covering different geographical environments, different time conditions, and different road types, providing a sufficient and high-quality data foundation for subsequent model training.

[0103] Based on the diverse characteristics of China's geographical environment, the experiment selected three representative geographical areas for data collection to ensure that the training data can cover typical scenarios at different latitudes, terrains and road directions.

[0104] Eastern Coastal Cities Collection Area: The Shanghai-Hangzhou-Ningbo triangle was selected as a representative collection area for the eastern coastal areas. The road network includes east-west arterial roads such as the Shanghai-Hangzhou Expressway and the Hangzhou-Ningbo Expressway, as well as typical north-south roads in coastal areas.

[0105] North China Plain Collection Area: The Beijing-Tianjin-Shijiazhuang triangle was selected as a representative collection area for the North China Plain. The road network includes several radial expressways, such as the Beijing-Tianjin-Tangshan Expressway and the Beijing-Shijiazhuang Expressway, with roads radiating out from Beijing.

[0106] Southwest Mountainous Area: The Chengdu-Chongqing-Kunming triangle was selected as a representative collection area for the southwest mountainous area. The road network includes expressways such as the Chengdu-Chongqing Expressway and the Chengdu-Kunming Expressway, which traverse complex mountainous terrain. The roads are characterized by winding paths due to the constraints of the terrain.

[0107] Vehicle Configuration and Personnel Arrangement: Deploy 3-5 experimental vehicles equipped with a complete data collection system in each collection area. Each vehicle is equipped with the aforementioned vehicle information acquisition module 110 (multi-mode GNSS + IMU), road information acquisition module 120 (HD map interface), solar geometry determination module 130 (SPA algorithm unit), and perception acquisition module 140 (annotation device). A certain number or all of the vehicles will be equipped with an optional eye tracking device to implement dual-annotation verification.

[0108] Each vehicle was staffed with two experimenters: a primary (experimental) driver responsible for safe driving and subjective glare perception annotation, and a secondary driver responsible for monitoring equipment operation and data quality inspection. All drivers received training on the unified glare grading standard before the experiment to ensure consistent annotation.

[0109] Data collection may be performed according to the time window calculated by the solar geometry determination module 130 .

[0110] In a vehicle equipped with an eye tracking detector, the perception acquisition module 140 implements a complete dual labeling mechanism, combining subjective labeling and objective physiological indicator detection:

[0111] The driver uses a three-button labeling device to label glare in real time: the green button indicates no glare or permissible glare (label value 0), the yellow button indicates distracting glare (label value 0.5), and the red button indicates disabling glare (label value 1). Labeling can be triggered in two modes: active labeling (immediately when the driver experiences glare) and reminder labeling (glare perception is labeled with a voice reminder at preset driving intervals (such as every 10 meters). Each time a label is marked, or when only marking is performed at preset driving intervals, relevant sample data at the time of marking can be obtained or determined in real time: vehicle position (latitude and longitude), timestamp, sun position (altitude, azimuth), road parameters (direction, slope, curvature), vehicle status (speed, heading), etc. These parameters provide a basis for subsequent labeling rationality verification.

[0112] For samples with both subjective annotations and objective detection, the system uses a weighted fusion algorithm to generate the final label: Final label = 0.7 × subjective annotation + 0.3 × objective detection result.

[0113] Continue to refer Figure 1 In an embodiment of the present invention, the multidimensional feature generation module 150 is configured to generate multidimensional features for glare samples based on driving information, synchronization time information, road geometry information, and solar spatial position information. Optionally, the system 100 or the multidimensional feature generation module 150 may include a preprocessing unit, such as a cleaning unit, to preliminarily clean and format the relevant sample data to form a standardized data set. Each sample may thus include multiple feature raw values ​​(such as road information (road direction, slope, curvature), time characteristics (month, day, hour), solar geometry information (altitude angle, azimuth), and vehicle information (speed, heading)) for generating corresponding multidimensional features.

[0114] In various embodiments of the present invention, the multidimensional feature generation module 150 can use any suitable method to generate or convert multiple feature raw values ​​into corresponding multidimensional features. In some embodiments, the multidimensional feature generation module 150 employs a direct factor feature scheme, generating multidimensional glare sample features directly based on the road strike angle, road slope angle, and road curvature radius as feature components. The road geometric feature information includes the road strike angle, road slope angle, and road curvature radius, and the solar spatial position information includes the solar altitude angle and solar azimuth angle. Accordingly, the multidimensional feature generation module 150 generates multidimensional glare sample features directly based on the vehicle speed and heading angle in the driving information, the month, day, and time in the synchronized time information (the year information in this synchronized information may not be used as a feature raw value), the road strike angle, road slope angle, road curvature radius, solar altitude angle, and solar azimuth angle as independent feature components. In other embodiments, the multidimensional feature generation module 150 generates multidimensional glare sample features based on relative position information in addition to or in place of at least a portion of the road geometric feature information.

[0115] In some embodiments, the multi-dimensional feature generation module 150 may include a normalization unit, thereby normalizing the original feature values ​​before generating the multi-dimensional features.

[0116] In some embodiments, the multidimensional feature generation module 150 also includes a sample balancing submodule configured to perform data augmentation on samples collected during the second acquisition period based on the difference in sample collection density between the first and second acquisition periods to compensate for the difference in sample collection density between the first and second acquisition periods. The sample balancing submodule employs the Synthetic Minority Oversampling Technique (SMOTE) method for data augmentation, generating new synthetic samples by interpolating between existing minority class samples, effectively increasing the number of minority class samples (here, samples from the second acquisition period). During specific implementations, the sample balancing submodule can identify the acquisition period to which sample data belongs based on time information (timestamp).

[0117] Through the above-described configuration, the embodiments of the present invention achieve outstanding results. By way of explanation, not limitation, the present invention recognizes that solar glare exhibits a significant temporal concentration, primarily occurring during specific periods of low solar altitude. Based on this characteristic, the present invention employs a biased data collection strategy, namely, high-density data collection during the first collection period, when glare is most prevalent. However, the present invention further discovered that while this biased data collection strategy addresses the scarcity of glare samples, it introduces new technical challenges. The temporal distribution of the training data significantly deviates from the all-weather distribution characteristics of actual road scenes. To address this issue, the present invention implements a reverse compensation strategy within the sample balancing submodule, performing data augmentation on samples from the second collection period, artificially increasing their proportion in the training set and thereby restoring a balanced temporal distribution of the training data. Although the present invention proactively collects an unbalanced data distribution during the acquisition process and then augments seemingly inefficient data during non-critical periods, the present invention finds that this "biased data collection + reverse compensation" approach ensures the model's ability to fully learn glare characteristics while maintaining its generalization performance in all-weather application scenarios.

[0118] In an embodiment of the present invention, the training module 160 is configured to train a preset glare risk prediction decision tree model based on the multidimensional features of glare samples and the corresponding hierarchical perception labels, and use the trained glare risk prediction decision tree model as a solar glare warning model. In some embodiments, the training module 160 uses the LightGBM (Light Gradient Boosting Machine) gradient boosting decision tree algorithm. LightGBM is a gradient boosting framework based on decision trees, and its working principle involves constructing a strong learner by integrating multiple weak learners (decision trees). In a specific embodiment, the glare risk prediction decision tree model based on LightGBM can be trained in this way, starting with an initial prediction, for example, an initial value set based on experience or default settings, such as an initial prediction probability of 0.3. Then, an iterative training process is entered, and each iteration adds a new decision tree to correct the prediction error that still exists after all the previous trees are combined. For example, the first tree analyzes which feature combinations lead to initial prediction errors. By learning the relationship between the sample's feature values ​​and the true label, it constructs decision rules to reduce prediction errors. Based on the prediction results of the first tree, the second tree continues to analyze the remaining prediction errors and learn more fine-grained feature combination patterns. This process continues iteratively until a preset number of trees is reached or the prediction accuracy meets the requirements. In an optional embodiment, the feature space can also be optimized based on a histogram, discretizing the feature values ​​into a finite number of intervals. Histogram statistics can then be used to accelerate the search for the optimal split point. For example, for the solar altitude feature in glare prediction, the algorithm might divide the range of 0°-90° into 180 intervals of 0.5° each, using a histogram to quickly find the split point that maximizes information gain. For the final prediction, the output values ​​of all decision trees are summed to obtain a total score, which is then converted to a probability value between 0 and 1 using a normalization function, such as a sigmoid function. In one specific embodiment, when the probability exceeds a preset threshold (e.g., 0.65), the system 100 determines that a glare risk exists.

[0119] In this embodiment of the present invention, training module 160 can use multiple metrics to evaluate model performance, including AUC-ROC, precision, recall, and F1-score. AUC-ROC reflects the overall classification capability of the model. Precision is defined as the proportion of samples predicted as glare that are actually glare, reflecting the accuracy of the warning. Recall is defined as the proportion of correct predictions of actual glare situations, reflecting the ability to capture real glare events. F1-score is the harmonic mean of precision and recall, which comprehensively evaluates model performance.

[0120] As an explanation but not limitation, since LightGBM adopts a leaf-to-leaf growth strategy, each time the leaf node with the largest loss reduction is selected for splitting, no matter which layer of the tree the leaf is located, this strategy can achieve lower loss and better accuracy under the same limit on the number of leaves, which is particularly suitable for dealing with the problem of uneven feature space distribution in glare prediction.

[0121] Optionally, the system 100 or the training module 160 may further include a testing (sub)module, for example, for conducting a cross-regional and cross-seasonal generalization ability test after the training is completed.

[0122] like Figure 2 As shown, another embodiment of the present invention provides a system 200 for constructing a sun glare warning model for road driving. The system 200 may include a vehicle information acquisition module 210, a road information acquisition module 220, a perception acquisition module 240, a multi-dimensional feature generation module 250, and an end-to-end training module 260. Figure 2 The system 200 of the illustrated embodiment is Figure 1 The main difference between the system 100 of the embodiment shown is that the system 200 may not include an independent solar geometry feature determination module. Instead, the model adopts an end-to-end integrated model architecture and integrates the solar position calculation function into the end-to-end integrated model. Here, the functions and implementation methods of the vehicle information acquisition module 210, the road information acquisition module 220 and the perception acquisition module 240 are the same as those of the vehicle information acquisition module 210, the road information acquisition module 220 and the perception acquisition module 240. Figure 1 The corresponding modules 110, 120, and 140 are basically the same and will not be described again here.

[0123] Optionally, the system 200 may also include an eye movement detector for detecting the driver's eye movement data at the corresponding synchronization time. In other preferred embodiments, the system 200 or its modules may also similarly include additional modules or configurations such as a multi-mode GNSS receiver, an IMU inertial measurement unit, a brightness acquisition module (on-board camera system), or other submodules or units. The specific functions and implementation methods of these optional additional modules or configurations can be referred to. Figure 1 Description of the corresponding module in .

[0124] In an embodiment of the present invention, the multi-dimensional feature generation module 250 is configured to generate multi-dimensional features of the glare sample based on geographic location information, driving information, synchronization time information and road geometric feature information. Figure 1 Unlike the multi-dimensional feature generation module 150 in FIG, the multi-dimensional feature generation module 250 does not receive the solar space position information as input, but receives Figure 1In the embodiment shown, the geographical location information (longitude, latitude) is not used as a feature factor. Specifically, the multidimensional feature vector generated by the multidimensional feature generation module 250 may include: geographical location features (such as longitude, latitude), road geometry features (road heading angle, road slope angle, road curvature), time features (year, month, day, hour), vehicle state features (vehicle speed, vehicle heading angle), and other multidimensional feature vectors. It should be noted that Figure 1 The embodiments are different. Figure 2 The time feature in the illustrated embodiment contains complete year information, and the end-to-end integrated model will use the year information to internally calculate the astronomical parameters of the sun's position.

[0125] In an embodiment of the present invention, the end-to-end training module 260 is configured to train an end-to-end integrated model based on the multi-dimensional features of the glare samples and the corresponding hierarchical perception labels (or dual perception labels of the included hierarchical perception labels), and use the trained end-to-end integrated model as a solar glare warning model. The end-to-end integrated model integrates a preset solar position calculation algorithm model and a glare risk prediction decision tree model. Here, the end-to-end training module 260 integrates the calculation logic of the Solar Position Algorithm (SPA) algorithm into the model architecture, and directly learns glare risk prediction from geographic location information and synchronized time information through end-to-end learning, wherein the solar geometric features participate in the prediction process in the form of hidden layer features. Specifically, the end-to-end integrated model internally contains network layers that implement the functions of the SPA algorithm. These network layers can calculate the altitude and azimuth of the sun based on the input latitude, longitude and time information, but these intermediate calculation results do not need to be explicitly output, but are directly passed to the subsequent glare risk prediction layer as hidden layer features. Figure 1 Similar to the embodiment, the decision tree part for glare risk prediction in the end-to-end integrated model also adopts the gradient boosting decision tree architecture, but Figure 1 Unlike the standalone LightGBM model, the decision tree model here is deeply integrated with the sun position calculation model. In some embodiments, the glare risk prediction decision tree model of the end-to-end integrated model is implemented using a differentiable decision tree, enabling end-to-end optimization of the entire model through backpropagation.

[0126] During the training process, the end-to-end training module 260 uses a gradient descent algorithm to simultaneously optimize the parameters of the sun position calculation part and the glare risk prediction decision tree model part, thereby achieving end-to-end optimization from the original input to the final prediction.

[0127] and Figure 1Compared to System 100, System 200 reduces the complexity of inter-module interfaces and avoids the accumulation of intermediate calculation errors. Furthermore, the end-to-end integrated model automatically adjusts the internal sun position calculation accuracy based on the requirements of the glare prediction task, optimizing computational efficiency while ensuring prediction accuracy.

[0128] During the deployment phase, the trained end-to-end integrated model can be deployed as a single model file, simplifying the complexity of system integration. It will be appreciated that different input parameters may vary depending on the trained model, as further described below.

[0129] like Figure 3 As shown, the embodiment of the present invention provides a sun glare warning system 300 for road driving, comprising: a trajectory node determination module 310, a sun glare warning model 330, and a warning control module 340. In a preferred embodiment, the system 300 further comprises a filter 320. The sun glare warning system 300 of the embodiment of the present invention can be constructed based on the embodiment of the present invention, as described above. Figure 1 or Figure 2 The deployment system of the solar glare warning model constructed by the model construction system shown or the construction method described later in actual application scenarios can be integrated and applied to a navigation system or terminal, an intelligent driving system or terminal, or an in-vehicle assisted driving system or terminal.

[0130] In an embodiment of the present invention, the trajectory node determination module 310 is configured to determine the future trajectory nodes of the user's vehicle. Figure 4 As shown, when the user vehicle 410 is traveling on the road, the trajectory node determination module 310 predicts the location points that the vehicle will pass through in the future time window, such as the future trajectory node 420, based on the vehicle's current location, driving direction and navigation path planning.

[0131] In one specific embodiment, trajectory node determination module 310 can be integrated with an in-vehicle navigation system or intelligent driving path planning module to obtain a planned driving route. In some embodiments, for each future trajectory node, module 310 can not only determine the future trajectory node's geographic location (latitude and longitude) but also predict the time it will take for the vehicle to arrive at that node. This time can be used as time information for subsequent model input. For example, based on the current vehicle speed and road conditions, it is predicted that user vehicle 410 will arrive at the location of future trajectory node 420 in 30 seconds. Alternatively or additionally, the time information used as subsequent model input can be determined based on the current time and a preset vehicle speed (e.g., a prescribed road speed).

[0132] In embodiments of the present invention, trajectory nodes can be determined in a variety of ways. For example, for a vehicle using navigation, future trajectory points can be directly extracted from the navigation path. For example, for an intelligent driving vehicle, an accurate future trajectory sequence can be obtained from a path planning module.

[0133] Optionally, vehicle driving information of the user's vehicle may also be obtained, such as vehicle location, driving direction, geographic information (latitude and longitude), etc. Since the system 300 of the embodiment of the present invention can be integrated into a navigation or intelligent driving system or terminal, such information can be easily obtained.

[0134] The solar glare warning model 330 is a system or method according to the present invention, such as the aforementioned Figure 1 System 100 or Figure 2 The trained model is constructed using system 200 or the construction method described below. This model is configured to output glare risk prediction results for future trajectory nodes. For each future location point provided by trajectory node determination module 310, solar glare warning model 330 determines the corresponding glare risk probability of future trajectory node 420 based on the vehicle information, road information, and solar position information (if available) at that location. In actual operation, solar glare warning model 330 can be deployed on a cloud server or an onboard edge computing platform, which is not limited by the present invention.

[0135] In some embodiments, the sun glare warning model 330 may be a non-end-to-end model, for example, based on Figure 1 or Figure 7 Accordingly, the solar glare warning model 330 can obtain multi-dimensional feature inputs, such as corresponding to Figure 1These features or raw values ​​can be obtained as needed, for example, based on the aforementioned future trajectory node information corresponding to the user vehicle in combination with a high-precision map, and optionally in combination with the previously acquired vehicle information of the user vehicle. Here, for the purpose of explanation and not limitation, the above-mentioned road information (road direction, slope, curvature), time characteristics (month, day, hour), solar geometry information (altitude angle, azimuth angle) and vehicle information (speed, heading) refer to information at future trajectory nodes, so that the road information (road direction, slope, curvature) can be determined based on the position information of the future trajectory node in combination with the high-precision map and optionally in combination with the vehicle's trajectory (i.e., the direction of travel), and the vehicle information (speed, heading) can be determined based on prediction or approximation, for example, the vehicle speed can be approximated by the current vehicle speed or by a preset vehicle speed (such as the road specified speed), and the heading can be approximated or simulated based on the future trajectory node, for example, it can be assumed that the vehicle will be in the middle lane when traveling at the future trajectory node.

[0136] In this embodiment, the system 300 may include an additional solar geometry feature determination module (not shown) for calculating the solar altitude and azimuth angles at each future trajectory node in real time. Figure 1 The same SPA algorithm is implemented as in module 130. Accordingly, the system 300 may include an optional multi-dimensional feature generation module, such as Figure 1 However, it is conceivable that the system 300 may not include the solar geometry feature determination module and / or the multi-dimensional feature generation module, and the relevant data, features or information may be provided by another device or data source, which falls within the scope of the present invention.

[0137] In some embodiments, the sun glare warning model 330 may be an end-to-end model, for example based on Figure 2 or Figure 8 The end-to-end model constructed in the embodiment shown. Accordingly, the sun glare warning model 330 can obtain multi-dimensional feature inputs, such as corresponding to Figure 2 These include multidimensional feature vectors such as those described in the embodiments, such as geographic location features (e.g., longitude and latitude), road geometry features (road heading angle, road slope angle, road curvature), time features (year, month, day, hour), and vehicle status features (vehicle speed, vehicle heading angle). These features or raw values ​​can be obtained as needed and are not described in detail here. The end-to-end model already integrates the sun position calculation function, so system 300 does not require a separate sun position calculation module. Similarly, system 300 may include a multidimensional feature generation module.

[0138] In an embodiment of the present invention, the warning control module 340 is configured to selectively generate a warning signal based on the glare risk prediction result. When the glare risk probability of a future trajectory node exceeds a preset threshold (such as 0.65), the warning control module 340 generates a corresponding warning message. Figure 5 As shown, on the display interface of the navigation terminal or intelligent driving terminal, the warning information can be displayed as: "There is a risk of sun glare XX meters ahead, please take XX measures". Here, the warning method can include visual warning (such as Figure 5 The present invention does not limit this herein. In some embodiments, the warning method may also provide additional or alternative measures based on the integrated terminal or system. For example, when integrated into a navigation system or terminal, an intelligent driving system or terminal, or an on-board assisted driving system or terminal, the warning method may include changing the planned path (such as a navigation path or an intelligent driving path), and / or, when integrated into a navigation system or terminal, an intelligent driving system or terminal, or an on-board assisted driving system or terminal, the warning method may include changing the intelligent driving or assisted driving mode, such as automatically or semi-automatically triggering the takeover or exiting the intelligent or assisted driving mode.

[0139] In a preferred embodiment, system 300 may further include a filter 320 configured to identify sections of road without glare risk and filter future trajectory nodes and / or glare risk prediction results based on the identified sections without glare risk. Filter 320 may determine sections of road that will not generate glare based on the tunnel properties of the road. By way of explanation and not limitation, a certain distance (e.g., 50 meters) near a tunnel exit is not considered to be within the tunnel, as the light-dark adaptation process upon exiting the tunnel may exacerbate the effects of glare. Filter 320 may operate outside of this distance range (hereinafter referred to as the tunnel interior), or may be ineffective within this distance range.

[0140] In some embodiments, filter 320 is a pre-filter positioned before solar glare warning model 330. Filter 320 can screen future trajectories and / or future trajectory nodes, directly marking trajectories and / or trajectory nodes located inside the tunnel (except for a certain distance from the exit) as risk-free and excluding them from the prediction range. Only trajectory nodes outside the tunnel that may pose a glare risk are passed to solar glare warning model 330 for prediction, thereby reducing unnecessary computational overhead.

[0141] In some embodiments, the filter 320 is a post-filter, which is located between the solar glare warning model 330 and the warning control module 340. After all trajectory nodes are predicted by the model, the filter 320 performs a secondary check on the prediction results and forcibly sets the prediction results of the nodes inside the tunnel (except the short distance to the exit) to no risk, so as to avoid false alarms in the tunnel due to model errors.

[0142] like Figure 6 , which shows a flow chart of a method for constructing a sun glare warning model for road driving according to an embodiment of the present invention, comprising the following steps:

[0143] S610: Acquire vehicle information, including the vehicle's geographical location information, driving information, and corresponding synchronization time information acquired simultaneously;

[0144] S620: Acquire road geometric feature information corresponding to the road location based on the geographic location information;

[0145] S630: Calculating the solar spatial position information at the corresponding synchronization time according to a preset solar position calculation algorithm based on the geographical location information and the synchronization time information;

[0146] S640: Obtaining a graded perception label of the driver's sun glare experience at the corresponding synchronization time;

[0147] S650: Generate multi-dimensional features of the glare sample based on driving information, synchronization time information, road geometric feature information, and sun spatial position information;

[0148] S660: A preset glare risk prediction decision tree model is trained based on the multi-dimensional features of the glare samples and the corresponding hierarchical perception labels, and the trained glare risk prediction decision tree model is used as a solar glare warning model.

[0149] Here, the system and its components, modules, units and features described in the embodiments of the present invention can be combined with the method of the embodiments of the present invention in a non-contradictory manner, and the method and its steps, sub-steps and features described in the embodiments of the present invention can also be combined with the system of the embodiments of the present invention in a non-contradictory manner. Figure 6 The specific implementation of the method shown can refer to the aforementioned Figure 1 The functions of the corresponding modules in the system 100 are shown as being implemented.

[0150] like Figure 7 FIG. 2 is a flowchart showing a method for constructing a sun glare warning model for road driving according to another embodiment of the present invention, comprising the following steps:

[0151] S710: Acquire vehicle information, including the vehicle's geographical location information, driving information, and corresponding synchronization time information acquired simultaneously;

[0152] S720: Acquire road geometric feature information corresponding to the road location based on the geographic location information;

[0153] S730: Obtaining a graded perception label of the driver's sun glare experience at the corresponding synchronization time;

[0154] S740: Generate multi-dimensional features of the glare sample based on the geographic location information, driving information, synchronization time information, and road geometric feature information;

[0155] S750: Train an end-to-end integrated model based on the multi-dimensional features of glare samples and the corresponding hierarchical perception labels, and use the trained end-to-end integrated model as a solar glare warning model.

[0156] Among them, the end-to-end integrated model shown integrates a preset sun position calculation algorithm model and a glare risk prediction decision tree model.

[0157] Here, the system and its components, modules, units and features described in the embodiments of the present invention can be combined with the method of the embodiments of the present invention in a non-contradictory manner, and the method and its steps, sub-steps and features described in the embodiments of the present invention can also be combined with the system of the embodiments of the present invention in a non-contradictory manner. Figure 7 The specific implementation of the method shown can refer to the aforementioned Figure 2 The functions of the corresponding modules in the system 200 are shown as being implemented.

[0158] like Figure 8 , which shows a flow chart of a sun glare warning method for road driving according to an embodiment of the present invention, comprising the following steps:

[0159] S810: Determine future trajectory nodes of the user's vehicle;

[0160] S820: Outputting glare risk prediction results for future trajectory nodes using a solar glare warning model constructed using the solar glare warning model construction method;

[0161] S830: Selectively generate a warning signal based on the glare risk prediction result.

[0162] Here, the system and its components, modules, units and features described in the embodiments of the present invention can be combined with the method of the embodiments of the present invention in a non-contradictory manner, and the method and its steps, sub-steps and features described in the embodiments of the present invention can also be combined with the system of the embodiments of the present invention in a non-contradictory manner. Figure 8 The specific implementation of the method shown can refer to the aforementioned Figure 3 The functions of the corresponding modules in the system 300 are implemented. In step S810, the future trajectory nodes can be predicted based on the current position, driving direction and navigation path planning of the vehicle. In step S820, the sun glare warning model can be an embodiment of the present invention, such as Figure 6 or Figure 7 The model constructed by the method of the illustrated embodiment. In step S830, a corresponding warning message is generated when the glare risk probability exceeds a preset threshold. Optionally, a filtering step may be included. The filtering step may include identifying road sections without glare risk and filtering the future trajectory nodes and / or the glare risk prediction results based on the determined road sections without glare risk. Specifically, the future trajectory nodes in the form of the user's vehicle (including filtering out future trajectories) and / or the glare risk prediction results may be filtered based on preset conditions, wherein the preset conditions are determined based on road tunnel sections.

[0163] In an embodiment of the present invention, an electronic device may be provided, including: a processor and a memory storing a computer program, wherein the processor is configured to execute any method of the embodiment of the present invention when running the computer program.

[0164] Figure 9 A schematic diagram of an electronic device 900 that can implement a method or realize an embodiment of the present invention is shown. In some embodiments, the method may include more or fewer electronic devices than shown. In some embodiments, the method may be implemented using a single electronic device or multiple electronic devices. In some embodiments, the method may be implemented using cloud-based or distributed electronic devices.

[0165] like Figure 9 As shown, electronic device 900 includes a processor 901, which can perform various appropriate operations and processes based on programs and / or data stored in read-only memory (ROM) 902 or programs and / or data loaded from storage 908 into random access memory (RAM) 903. Processor 901 can be a multi-core processor or include multiple processors. In some embodiments, processor 901 can include a general-purpose main processor and one or more specialized coprocessors, such as a graphics processing unit (GPU), a neural network processing unit (NPU), a digital signal processor (DSP), etc. RAM 903 also stores various programs and data required for the operation of electronic device 900. Processor 901, ROM 902, and RAM 903 are interconnected via a bus 904. An input / output (I / O) interface 905 is also connected to bus 904.

[0166] The processor and the memory are used together to execute the program stored in the memory. When the program is executed by the computer, the steps or functions of the methods described in the above embodiments can be implemented.

[0167] The following components are connected to the I / O interface 905: an input section 906 including a keyboard, mouse, touch screen, and the like; an output section 907 including devices such as a cathode ray tube (CRT), a liquid crystal display (LCD), and speakers; a storage section 908 including devices such as a hard disk; and a communication section 909 including a network interface card such as a LAN card or a modem. The communication section 909 performs communication processing via a network such as the Internet. A drive 910 is also connected to the I / O interface 905 as needed. Removable media 911, such as a magnetic disk, an optical disk, a magneto-optical disk, or a semiconductor memory, is installed in the drive 910 as needed, so that computer programs read therefrom can be installed in the storage section 908 as needed. Figure 9 Only some components are shown schematically, which does not mean that the computer system 900 only includes Figure 9 Components shown.

[0168] In some embodiments, the electronic device includes a mobile terminal or a computer, including a mobile phone, a car terminal, a smart TV, etc. Taking a mobile phone as an example, the electronic device also includes a display screen with a touch function, an external speaker, a gyroscope, a camera, a 4G / 5G antenna and other device modules.

[0169] The systems, devices, modules, or units described in the above embodiments may be implemented by a computer or its associated components. The computer may be, for example, a mobile terminal, a smartphone, a personal computer, a laptop computer, an in-vehicle human-computer interaction device, a personal digital assistant, a media player, a navigation device, a game console, a tablet computer, a wearable device, a smart TV, an Internet of Things system, a smart home, an industrial computer, a server, or a combination thereof.

[0170] Although not shown, in an embodiment of the present invention, a program product is provided. The program product includes a computer program configured to implement any method of the embodiments of the present invention when executed.

[0171] Although not shown, in an embodiment of the present invention, a storage medium is provided, wherein the storage medium stores a computer program, and the computer program is configured to implement any method of the embodiment of the present invention when executed.

[0172] Storage media in embodiments of the present invention include permanent and non-permanent, removable and non-removable items that can be used to store information using any method or technology. Examples of storage media include, but are not limited to, phase change memory (PRAM), static random access memory (SRAM), dynamic random access memory (DRAM), other types of random access memory (RAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), flash memory or other memory technologies, compact disc read-only memory (CD-ROM), digital versatile disc (DVD) or other optical storage, magnetic cassettes, magnetic tape, magnetic disk storage or other magnetic storage devices, or any other non-transmission medium that can be used to store information that can be accessed by a computing device.

[0173] The methods, programs, systems, and apparatuses of the embodiments of the present invention may be executed or implemented in a single or multiple networked computers, or may be practiced in a distributed computing environment. In the embodiments of this specification, in these distributed computing environments, tasks may be performed by remote processing devices connected via a communication network.

[0174] Those skilled in the art will appreciate that the embodiments of this specification may be provided as methods, systems, or computer program products. Therefore, those skilled in the art will appreciate that the functional modules / units or controllers and related method steps described in the above embodiments may be implemented using software, hardware, or a combination of software / hardware.

[0175] Unless explicitly stated, the actions or steps of the methods, procedures, and methods described in accordance with the embodiments of the present invention do not have to be performed in a specific order and can still achieve the desired results. In some implementations, multitasking and parallel processing are also possible or may be advantageous.

[0176] In this document, multiple embodiments of the present invention are described, but for the sake of brevity, the description of each embodiment is not exhaustive, and the same or similar features or parts between the embodiments may be omitted. In this document, "one embodiment", "some embodiments", "example", "specific example", or "some examples" are intended to apply to at least one embodiment or example according to the present invention, but not all embodiments. The above terms do not necessarily mean to refer to the same embodiment or example. Those skilled in the art may combine and combine the different embodiments or examples described in this specification and the features of the different embodiments or examples, unless they are mutually contradictory.

[0177] While the exemplary systems and methods of the present invention have been specifically shown and described with reference to the foregoing embodiments, these are merely examples of the best modes for implementing the present systems and methods. Those skilled in the art will appreciate that various changes may be made to the embodiments of the systems and methods described herein when implementing the present systems and / or methods without departing from the spirit and scope of the present invention as defined in the appended claims.

Claims

1. A road driving sun glare warning model construction system, characterized in that: include: A vehicle information acquisition module is configured to acquire vehicle information, wherein the vehicle information includes synchronously acquired geographic location information, driving information, and corresponding synchronization time information of the vehicle; a road information acquisition module configured to acquire road geometric feature information corresponding to a road position based on the geographic location information; a solar geometric feature determination module configured to calculate the solar spatial position information at the corresponding synchronization time according to a preset solar position calculation algorithm based on the geographical location information and the synchronization time information; a perception acquisition module configured to acquire a graded perception label of the driver's perception of sun glare at a corresponding synchronization time; a multi-dimensional feature generation module configured to generate multi-dimensional features of glare samples based on the driving information, synchronization time information, road geometric feature information and sun spatial position information; The training module is configured to train a preset glare risk prediction decision tree model based on the multidimensional features of the glare samples and the corresponding hierarchical perception labels, and use the trained glare risk prediction decision tree model as a solar glare warning model.

2. A road driving sun glare warning model construction system, characterized in that: include: A vehicle information acquisition module is configured to acquire vehicle information, wherein the vehicle information includes synchronously acquired geographic location information, driving information, and corresponding synchronization time information of the vehicle; a road information acquisition module configured to acquire road geometric feature information corresponding to a road position based on the geographic location information; a perception acquisition module configured to acquire a graded perception label of the driver's perception of sun glare at a corresponding synchronization time; a multi-dimensional feature generation module configured to generate multi-dimensional features of the glare sample based on the geographic location information, driving information, synchronization time information and road geometric feature information; The end-to-end training module is configured to train an end-to-end integrated model based on the multidimensional features of the glare samples and the corresponding hierarchical perception labels, and use the trained end-to-end integrated model as a solar glare warning model. The end-to-end integrated model integrates a preset sun position calculation algorithm model and a glare risk prediction decision tree model.

3. The solar glare warning model construction system according to claim 1 or 2, characterized in that: The vehicle information acquisition module is configured to acquire the vehicle information with different sample collection densities in a first collection period corresponding to a preset time condition and a second collection period not corresponding to the preset time condition based on a day cycle, wherein the sample collection density in the first collection period is higher than the sample collection density in the second collection period; The multi-dimensional feature generation module includes a sample balancing submodule configured to perform data augmentation processing on samples collected during the second collection period based on the difference in sample collection density between the first collection period and the second collection period, so as to compensate for the difference in sample collection density between the first collection period and the second collection period; The glare samples used for training are glare samples collected at different collection densities and subjected to compensation processing.

4. The solar glare warning model construction system according to claim 1 or 2, characterized in that: The vehicle information acquisition module is configured to acquire vehicle information based on a preset spatial sampling interval; the road information acquisition module and the perception acquisition module are configured to synchronously acquire the road geometric feature information and the driver's glare perception information based on the same spatial sampling interval.

5. The solar glare warning model construction system according to claim 1 or 2, characterized in that: Also includes: an eye movement detector configured to detect eye movement data of the driver at corresponding synchronous times, wherein the eye movement data includes pupil changes and / or gaze deviation; The system further includes a label generation module configured to construct a dual perception label from the eye movement data and the hierarchical perception label; Wherein, the training module is configured to train a preset glare risk prediction decision tree model based on the multidimensional features and the dual perception labels.

6. A sun glare warning system for road driving, characterized in that: include: a trajectory node determination module configured to determine future trajectory nodes of a user's vehicle; a solar glare warning model constructed by the solar glare warning model construction system according to any one of claims 1 to 5, wherein the solar glare warning model is configured to output a glare risk prediction result for the future trajectory node; The warning control module is configured to selectively generate a solar glare risk warning signal according to the glare risk prediction result.

7. The sun glare warning system according to claim 6, characterized in that: Also includes: A filter is configured to identify a glare-risk-free road section and filter the future trajectory node and / or the glare risk prediction result based on the determined glare-risk-free road section, wherein the glare-risk-free road section is determined based on a tunnel section of the road.

8. A method for constructing a sun glare warning model for road driving, characterized in that: include: Acquiring vehicle information, the vehicle information including synchronously acquired vehicle geographic location information, driving information, and corresponding synchronization time information; Based on the geographic location information, obtaining road geometric feature information corresponding to the road location; Based on the geographical location information and the synchronization time information, calculating the solar spatial position information at the corresponding synchronization time according to a preset solar position calculation algorithm; Obtaining the graded perception labels of the driver's sun glare perception at the corresponding synchronization time; generating a multi-dimensional feature of the glare sample based on the driving information, synchronization time information, road geometric feature information and sun spatial position information; A preset glare risk prediction decision tree model is trained based on the multidimensional features of the glare samples and the corresponding hierarchical perception labels, and the trained glare risk prediction decision tree model is used as a solar glare warning model.

9. A method for constructing a sun glare warning model for road driving, characterized in that: include: Acquiring vehicle information, the vehicle information including synchronously acquired vehicle geographic location information, driving information, and corresponding synchronization time information; Based on the geographic location information, obtaining road geometric feature information corresponding to the road location; Obtaining the graded perception labels of the driver's sun glare perception at the corresponding synchronization time; generating multi-dimensional features of the glare sample based on the geographic location information, driving information, synchronization time information and road geometric feature information; An end-to-end integrated model is trained based on the multidimensional features of the glare samples and the corresponding hierarchical perception labels, and the trained end-to-end integrated model is used as a solar glare warning model. The end-to-end integrated model integrates a preset sun position calculation algorithm model and a glare risk prediction decision tree model.

10. A sun glare warning method for road driving, characterized in that: The method includes: Determine the future trajectory nodes of the user's vehicle; Outputting a glare risk prediction result of the future trajectory node using the solar glare warning model constructed by the solar glare warning model construction method according to claim 8 or 9; A solar glare risk warning signal is selectively generated based on the glare risk prediction result.

Images