An unmanned dynamic obstacle prediction system based on a visual large model

Abstract

The application relates to the field of obstacle prediction of unmanned vehicles, and discloses an unmanned vehicle dynamic obstacle prediction system based on a visual large model, which comprises a road division module, a time division module, a road unit collection module, a road unit risk assessment module, a vehicle information collection module, a vehicle obstacle avoidance assessment module, a vehicle driving risk assessment module and an obstacle avoidance module, and the vehicle obstacle avoidance state index of the unmanned vehicle is obtained through the vehicle obstacle avoidance assessment module; the obstacle avoidance ability state of the vehicle itself can be timely and accurately mastered, so that the vehicle can make more reasonable decisions according to real-time conditions in the driving process, and then the current vehicle risk level is obtained through the vehicle driving risk assessment module; the comprehensive evaluation mode fully considers various factors such as the road environment, surrounding obstacles and the vehicle state, and can provide comprehensive and accurate risk evaluation results for the unmanned vehicle.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of obstacle prediction for unmanned vehicles, and specifically relates to an obstacle prediction system for unmanned vehicles based on a visual large model. BACKGROUND

[0002] With the rapid development of artificial intelligence, sensor technology, computer vision and other fields, the application of unmanned vehicle technology in the express delivery field has gradually moved from the experimental stage to large-scale commercial use, becoming the core force driving the upgrading of intelligent logistics.

[0003] In the prior art, unmanned vehicle technology mainly uses multi-sensor fusion technology, such as laser radar, camera, millimeter wave radar, etc., installed on unmanned vehicles to perceive and model the environment around the vehicle, and then realize path planning, decision control and other functions.

[0004] However, in the actual express delivery scenario, the road environment is complex and variable, and the appearance of dynamic obstacles (such as pedestrians, other vehicles, and suddenly appearing animals) has uncertainty and randomness. Although the existing multi-sensor fusion technology can obtain rich environmental information, it still has certain limitations in predicting the behavior of dynamic obstacles and assessing risks when faced with complex and variable dynamic obstacles, making it difficult to accurately judge the risk level of unmanned vehicles in different scenarios, and thus unable to timely and effectively develop reasonable obstacle avoidance strategies, which to some extent restricts the wider and safer application of unmanned vehicle technology in the express delivery field. SUMMARY

[0005] The purpose of the present application is to provide an obstacle prediction system for unmanned vehicles based on a visual large model, which solves the following technical problems:

[0006] In the scenario of unmanned vehicles, how to accurately assess the risk level currently faced by unmanned vehicles.

[0007] The purpose of the present application can be achieved by the following technical solutions:

[0008] An obstacle prediction system for unmanned vehicles based on a visual large model, the system comprising:

[0009] a road division module for dividing the roads in a target area into a plurality of road units;

[0010] a time division module for dividing a day into a plurality of time units;

[0011] a plurality of road unit acquisition modules corresponding to the road units one by one for obtaining road image information data and traffic signal information data of the corresponding road units;

[0012] a road unit risk assessment module, configured to analyze the road image information data and the traffic signal information data of each road unit, and obtain a comprehensive road risk index of the corresponding road unit in each time unit;

[0013] a vehicle information collection module, arranged on the unmanned vehicle, and configured to collect dynamic obstacle information data within a preset range of the unmanned vehicle;

[0014] a vehicle obstacle avoidance assessment module, configured to perform obstacle avoidance detection on the unmanned vehicle before a task, and obtain obstacle avoidance information data; and analyze the obstacle avoidance information data, and obtain an obstacle avoidance state index of the unmanned vehicle;

[0015] a vehicle driving risk assessment module, configured to analyze the comprehensive road risk index of the road unit where the unmanned vehicle is located in the current time unit, the dynamic obstacle information data, and the obstacle avoidance state index of the unmanned vehicle, and obtain a current vehicle risk level;

[0016] an obstacle avoidance module, configured to determine an obstacle avoidance strategy according to the current vehicle risk level.

[0017] As a further scheme of the present application, the road unit risk assessment module comprises an identification unit and an analysis unit; the identification unit is a trained convolutional neural network model, configured to identify the dynamic obstacle types according to the road image information data, and obtain the dynamic obstacle types of each dynamic obstacle on the road image information data; and the analysis unit is configured to analyze the movement rules of each dynamic obstacle in the road unit according to the corresponding dynamic obstacle type, and determine whether each dynamic obstacle is compliant in the road unit.

[0018] As a further scheme of the present application, the analysis unit further calculates the comprehensive road risk index of any road unit in any time unit by the following formula:

[0019] ;

[0020] calculates the comprehensive road risk index of any road unit in any time unit ;

[0021] wherein, is any road unit; is any time unit; is a past preset number of days, ; is a number of dynamic obstacle types, ; is a number of compliant dynamic obstacle types of the road unit in the time unit on the day of the past preset number of days; is the th dynamic obstacle type in the time unit on the day of the past preset number of days; a total number of dynamic obstacle types; a risk weight coefficient for the first dynamic obstacle type.

[0022] As a further scheme of the present application: the working process of the vehicle obstacle avoidance evaluation module is:

[0023] S1: obtaining an environmental influence coefficient according to environmental information data; the environmental information data includes environmental temperature, visibility, rainfall intensity and wind intensity;

[0024] S2: controlling the unmanned vehicle to move at a preset speed in a preset direction;

[0025] S3: controlling the preset detection dynamic obstacle to move at a random speed from the front of the unmanned vehicle to the unmanned vehicle at a constant speed, obtaining the distance between the unmanned vehicle and the dynamic obstacle at the beginning of this obstacle avoidance, the completion time of this obstacle avoidance of the unmanned vehicle and the minimum distance between the unmanned vehicle and the dynamic obstacle in this obstacle avoidance process;

[0026] S4: analyzing the distance between the unmanned vehicle and the dynamic obstacle at the beginning of this obstacle avoidance, the completion time of this obstacle avoidance of the unmanned vehicle and the minimum distance between the unmanned vehicle and the dynamic obstacle in this obstacle avoidance process to obtain an obstacle avoidance state index of the unmanned vehicle.

[0027] As a further scheme of the present application: in step S1, in step S1, the environmental influence coefficient is calculated by the formula:

[0028] ;

[0029] ;

[0030] wherein, is a first judgment function, when , ; when , ; is the current environmental temperature; is a preset environmental temperature; is a preset temperature allowable error value; is a preset visibility; is a current visibility; is a current wind intensity; is a preset wind intensity; is a current rainfall intensity; is a preset rainfall intensity; is a first weight coefficient; is a second weight coefficient; is a third weight coefficient;​ This is the fourth weighting coefficient; This is the first preset constant; This is the second preset constant; This is the third preset constant; This is the fourth preset constant.

[0031] As a further aspect of the present invention: In step S4, the formula is used:

[0032] ;

[0033] Calculate the obstacle avoidance state index of autonomous vehicles ;

[0034] in, This represents the distance between the autonomous vehicle and the dynamic obstacle at the start of the obstacle avoidance maneuver. The preset distance between the autonomous vehicle and dynamic obstacles at the start of obstacle avoidance; Preset completion time for obstacle avoidance by autonomous vehicles; This refers to the time it took for the autonomous vehicle to complete this obstacle avoidance maneuver. This represents the minimum distance between the autonomous vehicle and the dynamic obstacle during this obstacle avoidance process. The preset minimum distance between the autonomous vehicle and dynamic obstacles during obstacle avoidance; This is the first weighting coefficient; This is the second weighting coefficient; This is the third weighting coefficient; This is the first preset constant; This is the second preset constant; This is the third preset constant; This is the first preset adjustment coefficient; The preset adjustment coefficient is for number two; This is the preset adjustment coefficient for number three.

[0035] As a further aspect of the present invention: the preset range is obtained as follows:

[0036] S10: Analyze the comprehensive road risk index of the road unit where the autonomous vehicle is located in the current time unit to obtain the current range scaling ratio of the road unit.

[0037] S20: Adjust the standard range according to the current range scaling ratio to obtain the current preset range of the autonomous vehicle.

[0038] As a further aspect of the present invention: In step S10, the formula is used:

[0039] ;

[0040] Calculating current range scaling ratio of unmanned vehicle ;

[0041] wherein, is a comprehensive road risk index corresponding to the current time unit of the road unit where the unmanned vehicle is located; is a preset comprehensive road risk index; is a proportional preset constant.

[0042] As a further scheme of the present application: the dynamic obstacle information data includes the number of dynamic obstacles and the type of each dynamic obstacle.

[0043] As a further scheme of the present application: by the formula:

[0044] ;

[0045] Calculating current vehicle risk index ;

[0046] wherein, is the number of dynamic obstacles of the th dynamic obstacle type in the preset range; is a preset dynamic obstacle number risk index in the preset range; is the number of dynamic obstacles in the preset range; is an obstacle number weight coefficient; is an obstacle risk weight coefficient; is an obstacle number preset constant; is an obstacle preset constant;

[0047] Then, the current vehicle risk index is compared with a preset threshold ;

[0048] When , the current vehicle risk level is low risk;

[0049] When , the current vehicle risk level is medium risk;

[0050] When , the current vehicle risk level is high risk.

[0051] The beneficial effects of the present application are:

[0052] (1) The present application refines the target area road into several road units through the road division module, and combines the time division module to accurately segment the time unit in a day, so that the road risk assessment can be in-depth to the micro level of the specific road area in the specific time period; the road unit collection module comprehensively obtains road image information data and traffic signal information data, providing rich and accurate basic data support for the road unit risk assessment module, and then the comprehensive road risk index of each road unit in each time unit can be accurately calculated; compared with the traditional extensive road risk assessment, this refined assessment method can more sensitively capture the changes of road conditions with time and space, provide more accurate road risk warning for the unmanned vehicle, and effectively reduce the probability of accidents caused by insufficient understanding of road risk;

[0053] (2) The vehicle information collection module arranged on the unmanned vehicle collects dynamic obstacle information data within the preset range of the unmanned vehicle; the vehicle obstacle avoidance evaluation module performs obstacle avoidance detection on the unmanned vehicle before the task, and obtains obstacle avoidance information data; then, the obstacle avoidance information data is analyzed to obtain the obstacle avoidance state index of the unmanned vehicle; the dynamic environment around the vehicle and the obstacle avoidance ability state of the vehicle itself can be timely and accurately mastered, so that the vehicle can make more reasonable decisions according to the real-time situation during driving, and avoid collision and other dangerous situations caused by incorrect judgment of the obstacle avoidance ability; then, the vehicle driving risk evaluation module analyzes the comprehensive road risk index of the road unit where the unmanned vehicle is located in the current time unit, the dynamic obstacle information data and the vehicle obstacle avoidance state index to obtain the current vehicle risk level; this comprehensive evaluation mode fully considers the road environment, surrounding obstacles and vehicle state and other factors, avoids the one-sidedness of single factor evaluation, and can provide comprehensive and accurate risk assessment results for the unmanned vehicle;

[0054] (3) The obstacle avoidance module determines the obstacle avoidance strategy according to the current vehicle risk level; the effective connection between risk assessment and obstacle avoidance action is realized, the corresponding obstacle avoidance strategy is formulated according to different risk levels, the unmanned vehicle can take the most appropriate measures when facing different risk situations, the safety of driving is ensured, the efficiency and fluency of driving are improved, and the driving safety and reliability of the unmanned vehicle in complex road environment are greatly improved. BRIEF DESCRIPTION OF DRAWINGS

[0055] The present application will be further described below in conjunction with the drawings.

[0056] Figure 1 The system module framework of an embodiment of the present application. DETAILED DESCRIPTION

[0057] With reference to the drawings of the embodiments of the present application, the technical solutions in the embodiments of the present application will be described clearly and completely. Obviously, the described embodiments are only a part of the embodiments of the present application, rather than all the embodiments of the present application. Based on the embodiments of the present application, all the other embodiments obtained by those skilled in the art without creative work fall within the scope of the present application.

[0058] Please refer to Figure 1 As shown in the drawings, in one embodiment, a visual large model-based dynamic obstacle prediction system for unmanned vehicles is provided, which comprises:

[0059] a road division module, configured to divide the roads in the target area into a plurality of road units;

[0060] a time division module, configured to divide a day into a plurality of time units;

[0061] a plurality of road unit acquisition modules, each corresponding to a road unit, configured to acquire road image information data and traffic signal information data of the corresponding road unit;

[0062] a road unit risk assessment module, configured to analyze the road image information data and the traffic signal information data of each road unit to obtain a comprehensive road risk index of the corresponding road unit in each time unit;

[0063] a vehicle information acquisition module, arranged on an unmanned vehicle, configured to acquire dynamic obstacle information data within a preset range of the unmanned vehicle;

[0064] a vehicle obstacle avoidance assessment module, configured to perform obstacle avoidance detection on the unmanned vehicle before performing a task to obtain obstacle avoidance information data, and analyze the obstacle avoidance information data to obtain an obstacle avoidance state index of the unmanned vehicle;

[0065] a vehicle driving risk assessment module, configured to analyze the comprehensive road risk index of the road unit where the unmanned vehicle is located in the current time unit, the dynamic obstacle information data, and the obstacle avoidance state index of the unmanned vehicle to obtain a current vehicle risk level;

[0066] an obstacle avoidance module, configured to determine an obstacle avoidance strategy according to the current vehicle risk level;

[0067] Through the technical solution, the road of the target area is divided into several road units by the road division module; one day is divided into several time units by the time division module; the road image information data and the traffic signal information data of the corresponding road unit are obtained by the road unit collection module; the comprehensive road risk index of the corresponding road unit in each time unit is obtained by analyzing the road image information data and the traffic signal information data of each road unit by the road unit risk assessment module; the target area road is refined into several road units by the road division module, and the time unit of one day is accurately segmented in combination with the time division module, so that the road risk assessment can be deeply into the micro level of a specific road area in a specific time period; the road image information data and the traffic signal information data are comprehensively obtained by the road unit collection module, which provides rich and accurate basic data support for the road unit risk assessment module, and then the comprehensive road risk index of each road unit in each time unit can be accurately calculated; compared with the traditional extensive road risk assessment, this refined evaluation method can more sensitively capture the changes of road conditions with time and space, provide more accurate road risk warning for the unmanned vehicle, and effectively reduce the probability of accidents caused by insufficient understanding of road risk; the dynamic obstacle information data within the preset range of the unmanned vehicle is collected by the vehicle information collection module arranged on the unmanned vehicle; the obstacle avoidance detection of the unmanned vehicle is performed before the task is executed by the vehicle obstacle avoidance evaluation module, and the obstacle avoidance information data is obtained; then the obstacle avoidance state index of the unmanned vehicle is obtained by analyzing the obstacle avoidance information data; the dynamic environment around the vehicle and the obstacle avoidance ability state of the vehicle itself can be timely and accurately mastered, so that the vehicle can make more reasonable decisions according to the real-time situation during driving, and avoid collision and other dangerous situations caused by the failure of the obstacle avoidance ability judgment; then the current vehicle risk level is obtained by analyzing the comprehensive road risk index of the road unit where the unmanned vehicle is located in the current time unit, the dynamic obstacle information data and the vehicle obstacle avoidance state index by the vehicle driving risk assessment module; this comprehensive evaluation mode fully considers the road environment, surrounding obstacles and vehicle state and other factors, avoids the one-sidedness of single factor evaluation, and can provide comprehensive and accurate risk assessment results for the unmanned vehicle; finally, the obstacle avoidance strategy is determined according to the current vehicle risk level by the obstacle avoidance module; the effective connection of risk assessment and obstacle avoidance action is realized, the corresponding obstacle avoidance strategy is formulated according to different risk levels, which can make the unmanned vehicle take the most appropriate measures to deal with different risk situations, ensure the safety of driving, improve the efficiency and smoothness of driving, and greatly improve the driving safety and reliability of the unmanned vehicle in complex road environment.

[0068] As an embodiment of the present application, the road unit risk assessment module comprises an identification unit and an analysis unit; the identification unit is a trained convolutional neural network model, which is used for dynamic obstacle type identification according to road image information data, and obtains the dynamic obstacle type of each dynamic obstacle in the road image information data; the analysis unit analyzes the movement rule of each dynamic obstacle according to the corresponding dynamic obstacle type in the road unit, and judges whether each dynamic obstacle is compliant in the road unit;

[0069] Through the above technical solution, the embodiment identifies the dynamic obstacle in the road image information data through the trained convolutional neural network model, and identifies the dynamic obstacle type; the dynamic obstacle type includes cars, electric vehicles, bicycles and pedestrians, etc.; the movement rule of each dynamic obstacle in the road unit is obtained by analyzing the road layout (such as lane line, intersection shape) and traffic rules (such as speed limit, turning limit) of the road unit, which helps the system to reasonably predict the behavior of the dynamic obstacle, plan the driving path and speed of the unmanned vehicle in advance, avoid collision or conflict with the dynamic obstacle, and effectively improve the smoothness and safety of road traffic. The movement trajectory (position, speed, direction) of each dynamic obstacle and the traffic signal information data (traffic light, pedestrian crossing signal) in the road unit are analyzed to judge whether each dynamic obstacle is compliant in the road unit; through the above accurate dynamic obstacle identification and compliance judgment, not only the smoothness and safety of road traffic are improved, but also the reliability of the unmanned system is significantly enhanced, further promoting the wide application of unmanned technology.

[0070] It should be noted that the method of obtaining the movement rule of each dynamic obstacle in the road unit and the movement trajectory of each dynamic obstacle and the traffic signal information data in the road unit to judge whether each dynamic obstacle is compliant in the road unit according to the road layout and traffic rules of the road unit is a prior art, which is not described in detail here.

[0071] As an embodiment of the present application, the analysis unit further calculates the comprehensive road risk index of any road unit in any time unit by the formula:

[0072] ;

[0073] The comprehensive road risk index of any road unit in any time unit ;

[0074] wherein, is any road unit; is any time unit; is the past preset number of days, ; is the number of dynamic obstacle types, ; compliance quantity of the i-th dynamic obstacle type in the time unit of the j-th day in the past preset number of days of the road unit; compliance quantity of the i-th dynamic obstacle type in the time unit of the j-th day in the past preset number of days of the road unit; total quantity of the i-th dynamic obstacle type in the time unit of the j-th day in the past preset number of days of the road unit; total quantity of the i-th dynamic obstacle type in the time unit of the j-th day in the past preset number of days of the road unit; risk weight coefficient of the i-th dynamic obstacle type;

[0075] Through the above technical solutions, the embodiment compliance rate of the i-th dynamic obstacle type in the time unit of the j-th day in the past preset number of days of the road unit, that is, the ratio of the compliance quantity to the total quantity of the i-th dynamic obstacle type in the time unit of the j-th day in the past preset number of days of the road unit; compliance rate of the i-th dynamic obstacle type in the time unit of the j-th day in the past preset number of days of the road unit; compliance rate of the i-th dynamic obstacle type in the time unit of the j-th day in the past preset number of days of the road unit; average violation rate of the i-th dynamic obstacle type in the time unit of the road unit in the past preset number of days; comprehensive road risk index of all dynamic obstacle types in the time unit of the road unit in the past preset number of days; the higher the average violation rate of each dynamic obstacle type in the time unit of the road unit in the past preset number of days, the higher the comprehensive road risk index of the road unit in the time unit;

[0076] It should be noted that the past preset number of days ​​​​​​​​​​The risk weight coefficient of each dynamic obstacle type is a preset value, which is obtained according to experience and will not be described here.

[0077] As an embodiment of the present application, the working process of the vehicle obstacle avoidance evaluation module is as follows:

[0078] S1: obtaining an environmental influence coefficient according to environmental information data; the environmental information data includes environmental temperature, visibility, rainfall intensity and wind intensity;

[0079] S2: controlling the unmanned vehicle to move at a preset speed in a preset direction;

[0080] S3: controlling the preset detection dynamic obstacle to move at a random speed from the front of the unmanned vehicle to the unmanned vehicle at a constant speed, obtaining the distance between the unmanned vehicle and the dynamic obstacle at the beginning of this obstacle avoidance, the completion time of this obstacle avoidance of the unmanned vehicle and the minimum distance between the unmanned vehicle and the dynamic obstacle in the process of this obstacle avoidance;

[0081] S4: analyzing the distance between the unmanned vehicle and the dynamic obstacle at the beginning of this obstacle avoidance, the completion time of this obstacle avoidance of the unmanned vehicle and the minimum distance between the unmanned vehicle and the dynamic obstacle in the process of this obstacle avoidance to obtain the obstacle avoidance state index of the unmanned vehicle.

[0082] Through the above technical solution, the embodiment first obtains an environmental influence coefficient according to environmental temperature, visibility, rainfall intensity and wind intensity; the obtaining method of environmental temperature, visibility, rainfall intensity and wind intensity is a prior art, which will not be described here; then controls the unmanned vehicle to move at a preset speed in a preset direction; then controls the preset detection dynamic obstacle to move at a random speed from the front of the unmanned vehicle to the unmanned vehicle at a constant speed, obtains the distance between the unmanned vehicle and the dynamic obstacle at the beginning of this obstacle avoidance, the completion time of this obstacle avoidance of the unmanned vehicle and the minimum distance between the unmanned vehicle and the dynamic obstacle in the process of this obstacle avoidance; finally, analyzes the distance between the unmanned vehicle and the dynamic obstacle at the beginning of this obstacle avoidance, the completion time of this obstacle avoidance of the unmanned vehicle and the minimum distance between the unmanned vehicle and the dynamic obstacle in the process of this obstacle avoidance to obtain the obstacle avoidance state index of the unmanned vehicle; the obstacle avoidance ability state of the unmanned vehicle itself can be grasped in time and accurately, and the collision and other dangerous situations caused by the judgment error of the obstacle avoidance ability of the unmanned vehicle itself during the execution of the task can be avoided.

[0083] It should be noted that the obtaining method of the distance between the unmanned vehicle and the dynamic obstacle at the beginning of this obstacle avoidance, the completion time of this obstacle avoidance of the unmanned vehicle and the minimum distance between the unmanned vehicle and the dynamic obstacle in the process of this obstacle avoidance is a prior art, which will not be described here.

[0084] As an embodiment of the present application, in step S1, the formula is as follows:

[0085] ;

[0086] Computing an environmental impact coefficient ;

[0087] wherein, is a first judging function, when , ; when , ; is a current environmental temperature; is a preset environmental temperature; is a preset temperature allowable error value; is a preset visibility; is a current visibility; is a current wind strength; is a preset wind strength; is a current rainfall intensity; is a preset rainfall intensity; is a first weight coefficient; is a second weight coefficient; is a third weight coefficient; is a fourth weight coefficient; is a first preset constant; is a second preset constant; is a third preset constant; is a fourth preset constant;

[0088] Through the above technical solution, the embodiment is an absolute value of a temperature difference between the current environmental temperature and the preset environmental temperature; is a difference between the absolute value of the temperature difference between the current environmental temperature and the preset environmental temperature and the preset temperature allowable error value; in the formula , the first judging function in the formula refers to , used for judging whether the absolute value of the temperature difference between the current environmental temperature and the preset environmental temperature exceeds the preset temperature allowable error value; when , it is indicated that the absolute value of the temperature difference between the current environmental temperature and the preset environmental temperature exceeds the preset temperature allowable error value, which indicates that the current environmental temperature is too high or too low, which will cause the performance of the information acquisition unit (such as an image acquisition unit and a laser radar, etc.) on the unmanned vehicle to decline, therefore, the greater the difference between the absolute value of the temperature difference between the current environmental temperature and the preset environmental temperature and the preset temperature allowable error value, the greater the impact on the obstacle avoidance of the unmanned vehicle, and therefore the environmental impact coefficient is greater, ; when When the absolute value of the temperature difference between the current ambient temperature and the preset ambient temperature does not exceed the preset temperature allowable error value, it indicates that the current ambient temperature is within a suitable range and will not cause the performance of the information acquisition unit on the unmanned vehicle to decrease, thus having little effect on the obstacle avoidance of the unmanned vehicle, ; is the difference between the preset visibility and the current visibility; in the formula , the in the first judgment function refers to , which is used to determine whether the preset visibility exceeds the current visibility; when , it indicates that the preset visibility exceeds the current visibility, which means that the current environmental visibility is low. When the concentration of fog droplets and rain droplets is large, multiple scattering phenomena will occur, that is, the laser will be reflected and scattered multiple times between the fog droplets and rain droplets before reaching the receiver, which will increase the time delay of the return signal and cause errors in the measurement of the target distance by the laser radar. The fog droplets and rain droplets in the air will also cause the electromagnetic waves emitted by the millimeter wave radar to be reflected and refracted, causing part of the signal to propagate along a non-straight path to the receiver, interfering with the signal reflected directly, and forming multipath interference. This interference will distort the received signal waveform and affect the accurate measurement of the target distance, speed, and angle. Therefore, the greater the difference between the preset visibility and the current visibility, the greater the impact on the obstacle avoidance of the unmanned vehicle, and thus the greater the environmental influence coefficient ; ; when , it indicates that the preset visibility does not exceed the current visibility, which means that the current visibility is within a suitable range and will not cause the performance of the information acquisition unit on the unmanned vehicle to decrease, thus having little effect on the obstacle avoidance of the unmanned vehicle, ; is the difference between the current wind intensity and the preset wind intensity; in the formula , the in the first judgment function refers to , which is used to determine whether the current wind intensity exceeds the preset wind intensity; when , it indicates that the current wind intensity exceeds the preset wind intensity, which means that the current wind intensity is large. Dust, sand particles, and other small particles in the air will be lifted and accelerated by the wind. The laser radar relies on emitting a laser beam and receiving the reflected signal to perceive the surrounding environment. These fast-moving particles will cause the reflected signal to have a Doppler shift, resulting in a deviation between the received signal frequency and the actual transmitted frequency. This will make the environmental information obtained by the obstacle avoidance system inaccurate. Therefore, the greater the difference between the current wind intensity and the preset wind intensity, the greater the impact on the obstacle avoidance of the unmanned vehicle, and thus the greater the environmental influence coefficient ; ; when When the first judgment function is true, it indicates that the current wind intensity does not exceed the preset wind intensity, which means that the current wind intensity is within a suitable range and will not cause the performance of the information acquisition unit on the unmanned vehicle to decrease, so the influence on the obstacle avoidance of the unmanned vehicle is minimal, ; is the difference between the current rainfall intensity and the preset rainfall intensity; in the formula , the first judgment function in the formula refers to , which is used to determine whether the current wind intensity exceeds the preset wind intensity; when , it indicates that the current rainfall intensity exceeds the preset rainfall intensity, which means that the current rainfall intensity is large, the raindrops are dense and fall quickly, and the laser beams emitted by the laser radar will interact with a large number of raindrops during propagation. Raindrops cause scattering and absorption of laser, resulting in a significant attenuation of the received echo signal strength and a significant reduction in signal-to-noise ratio; this makes it difficult for the laser radar to accurately detect distant or small-sized obstacles, increasing the probability of missed detection and false detection; rainfall can change the dielectric constant of the atmospheric medium, and the presence of raindrops can cause scattering and attenuation of electromagnetic waves, so the detection performance of the millimeter wave radar will be significantly affected, with a shortened detection distance and reduced resolution capability for targets; especially in close-range detection, the strong signal reflected by the raindrops may mask the echoes of actual obstacles, causing the obstacle avoidance system to fail to correctly identify the position and characteristics of the obstacles; therefore, the greater the difference between the current rainfall intensity and the preset rainfall intensity, the greater the influence on the obstacle avoidance of the unmanned vehicle, and thus the greater the environmental influence coefficient ; ; when , it indicates that the current rainfall intensity does not exceed the preset rainfall intensity, which means that the current rainfall intensity is within a suitable range and will not cause the performance of the information acquisition unit on the unmanned vehicle to decrease, so the influence on the obstacle avoidance of the unmanned vehicle is minimal, ;

[0089] It should be noted that the preset environmental temperature , the preset temperature allowable error value , the preset visibility , the preset wind intensity , the preset rainfall intensity , the first weight coefficient , the second weight coefficient , the third weight coefficient , the fourth weight coefficient , the first preset constant , the second preset constant , the third preset constant , and the fourth preset constant The preset value is obtained empirically and is not described in detail herein.

[0090] As an embodiment of the present application, in step S4, the avoidance state index is calculated by the formula:

[0091] ;

[0092] Calculating an avoidance state index of an unmanned vehicle ;

[0093] wherein, is the distance between the unmanned vehicle and the dynamic obstacle at the beginning of the current avoidance; is the preset distance between the unmanned vehicle and the dynamic obstacle at the beginning of the avoidance; is the preset completion time of the avoidance of the unmanned vehicle; is the completion time of the current avoidance of the unmanned vehicle; is the minimum distance between the unmanned vehicle and the dynamic obstacle during the current avoidance; is the preset minimum distance between the unmanned vehicle and the dynamic obstacle during the avoidance; is a first weight coefficient; is a second weight coefficient; is a third weight coefficient; is a first preset constant; is a second preset constant; is a third preset constant; is a first preset adjustment coefficient; is a second preset adjustment coefficient; is a third preset adjustment coefficient;

[0094] Through the above technical solution, the present embodiment is the preset distance between the unmanned vehicle and the dynamic obstacle at the beginning of the avoidance under the current environment, and the environmental influence coefficient is greater, the greater the preset distance between the unmanned vehicle and the dynamic obstacle at the beginning of the avoidance under the current environment, is the difference between the distance between the unmanned vehicle and the dynamic obstacle at the beginning of the current avoidance and the preset distance between the unmanned vehicle and the dynamic obstacle at the beginning of the avoidance under the current environment, and when When the distance between the autonomous vehicle and the dynamic obstacle at the start of obstacle avoidance is greater than the preset distance for the same obstacle in the current environment, it indicates that the autonomous vehicle has more reaction time and space. The vehicle can more easily analyze the trajectory, speed, and other key information of the dynamic obstacle, accurately plan the obstacle avoidance path, and thus effectively reduce the risk of collision. This indicates that the autonomous vehicle's obstacle avoidance status is good. The greater the difference between the distance between the autonomous vehicle and the dynamic obstacle at the start of obstacle avoidance and the preset distance for the same obstacle in the current environment, the better the obstacle avoidance status index of the autonomous vehicle. The larger; when If the distance between the autonomous vehicle and the dynamic obstacle at the start of obstacle avoidance is less than the preset distance in the current environment, the vehicle's reaction time is significantly reduced, indicating a poor obstacle avoidance performance. The larger the absolute value of the difference between the initial distance and the preset distance in the current environment, the better the obstacle avoidance performance index. The smaller; similarly, The preset completion time for obstacle avoidance by autonomous vehicles under the current environment, and the environmental impact coefficient. The larger the value, the longer the preset obstacle avoidance completion time for autonomous vehicles in the current environment. The time difference between the preset completion time for obstacle avoidance by the autonomous vehicle in the current environment and the completion time of the obstacle avoidance in this instance is given. If the time difference is greater than the preset obstacle avoidance completion time under the current environment, it indicates that the obstacle avoidance completion time of the autonomous vehicle is relatively short, and the obstacle avoidance status of the autonomous vehicle is good. The greater the time difference between the preset obstacle avoidance completion time and the current obstacle avoidance completion time under the current environment, the better the obstacle avoidance status index of the autonomous vehicle. The larger; when If the time difference is greater than the preset obstacle avoidance completion time under the current environment, it indicates that the autonomous vehicle's obstacle avoidance status is poor. The larger the absolute value of the time difference between the preset obstacle avoidance completion time and the current obstacle avoidance completion time, the better the obstacle avoidance status index of the autonomous vehicle. The smaller; similarly, The environmental impact coefficient represents the preset minimum distance between the autonomous vehicle and dynamic obstacles during obstacle avoidance in the current environment. The larger the value, the greater the preset minimum distance between the autonomous vehicle and dynamic obstacles during obstacle avoidance in the current environment. When the difference between the minimum distance between the unmanned vehicle and the dynamic obstacle in the current obstacle avoidance process and the preset minimum distance between the unmanned vehicle and the dynamic obstacle in the current obstacle avoidance process is greater than 0, , it indicates that the minimum distance between the unmanned vehicle and the dynamic obstacle in the current obstacle avoidance process is greater than the preset minimum distance between the unmanned vehicle and the dynamic obstacle in the current obstacle avoidance process, which indicates that the obstacle avoidance state of the unmanned vehicle is better; the greater the difference between the minimum distance between the unmanned vehicle and the dynamic obstacle in the current obstacle avoidance process and the preset minimum distance between the unmanned vehicle and the dynamic obstacle in the current obstacle avoidance process, the greater the obstacle avoidance state index of the unmanned vehicle ; when , it indicates that the minimum distance between the unmanned vehicle and the dynamic obstacle in the current obstacle avoidance process is less than the preset minimum distance between the unmanned vehicle and the dynamic obstacle in the current obstacle avoidance process, which indicates that the obstacle avoidance state of the unmanned vehicle is poor; the greater the absolute value of the difference between the minimum distance between the unmanned vehicle and the dynamic obstacle in the current obstacle avoidance process and the preset minimum distance between the unmanned vehicle and the dynamic obstacle in the current obstacle avoidance process, the smaller the obstacle avoidance state index of the unmanned vehicle ;

[0095] It should be noted that the preset distance between the unmanned vehicle and the dynamic obstacle at the beginning of obstacle avoidance , the preset completion time of the unmanned vehicle in obstacle avoidance , the preset minimum distance between the unmanned vehicle and the dynamic obstacle in the obstacle avoidance process , the first weight coefficient , the second weight coefficient , the third weight coefficient , the first preset constant , the second preset constant , the third preset constant , the first preset adjustment coefficient , the second preset adjustment coefficient , and the third preset adjustment coefficient are preset values obtained according to experience, which will not be described in detail here.

[0096] As an embodiment of the present application, the preset range is obtained in the following manner:

[0097] S10: Analyze the comprehensive road risk index of the road unit where the unmanned vehicle is located to obtain the range scaling ratio of the road unit;

[0098] S20: Adjust the standard range according to the range scaling ratio to obtain the preset range.

[0099] Through the technical solution, the embodiment can make the vehicle information acquisition module more accurately acquire dynamic obstacle information data related to the current road risk according to the road unit risk adjustment preset range; in the road unit with high risk, appropriately expanding the preset range can discover potential obstacles in advance and increase the response time; in the unit with low risk, reducing the range can reduce unnecessary data processing and improve the pertinence and effectiveness of information acquisition.

[0100] It should be noted that the standard range is a preset value obtained according to experience, which is not described here.

[0101] As an embodiment of the application, in step S10, the current range scaling ratio of the unmanned vehicle is calculated by the formula:

[0102] ;

[0103] The current range scaling ratio of the unmanned vehicle ;

[0104] wherein, is the comprehensive road risk index corresponding to the current time unit of the road unit where the unmanned vehicle is located; is the preset comprehensive road risk index; is a proportional preset constant;

[0105] Through the technical solution, the embodiment is the difference between the comprehensive road risk index corresponding to the current time unit of the road unit where the unmanned vehicle is located and the preset comprehensive road risk index, in the formula , the in the first judgment function refers to , which is used to judge whether the comprehensive road risk index corresponding to the current time unit of the road unit where the unmanned vehicle is located exceeds the preset comprehensive road risk index; when , it means that the comprehensive road risk index corresponding to the current time unit of the road unit where the unmanned vehicle is located exceeds the preset comprehensive road risk index; it means that the risk of the road unit where the unmanned vehicle is located is large at the current time unit, so it is necessary to monitor a larger area and leave enough reaction time, so as to increase the current range scaling ratio of the unmanned vehicle , ; when , it means that the comprehensive road risk index corresponding to the current time unit of the road unit where the unmanned vehicle is located does not exceed the preset comprehensive road risk index, which means that the risk of the road unit where the unmanned vehicle is located is within a reasonable range, so there is no need to increase the current range scaling ratio of the unmanned vehicle , ;

[0106] It should be noted that the preset comprehensive road risk index and proportional preset constant These are preset values, obtained based on experience, and will not be detailed here.

[0107] As one embodiment of the present invention, the dynamic obstacle information data includes the number of dynamic obstacles and the types of each dynamic obstacle;

[0108] As one embodiment of the present invention, the process for determining the current vehicle risk level is as follows: First, using the formula:

[0109] ;

[0110] Calculate the current vehicle risk index ;

[0111] in, Among the dynamic obstacles within the preset range, the first one is... The number of dynamic obstacle types; This is the preset dynamic obstacle quantity risk index within the preset range; The number of dynamic obstacles within a preset range; This is the weighting coefficient for the number of obstacles; This refers to the obstacle risk weighting coefficient. A preset constant is set for the number of obstacles; Preset constants for obstacles;

[0112] Then, the current vehicle risk index With preset threshold Compare;

[0113] when At that time, the current risk level of the vehicle was low.

[0114] when At that time, the current risk level of the vehicle was medium risk;

[0115] when At that time, the current risk level of the vehicle was high.

[0116] Through the above technical solution, this embodiment The risk index for the number of dynamic obstacles within the current preset range; This is the difference between the current dynamic obstacle quantity risk index within the preset range and the preset dynamic obstacle quantity risk index within that preset range; when If the current vehicle obstacle quantity risk index is high, it indicates a higher risk level within the current preset range. Therefore, the greater the difference between the current preset range's dynamic obstacle quantity risk index and the preset dynamic obstacle quantity risk index within that range, the higher the current vehicle risk index. The larger; conversely, when If the current vehicle obstacle quantity risk index is low, it indicates that the risk within the current preset range is relatively small. Therefore, the larger the absolute value of the difference between the current preset range's dynamic obstacle quantity risk index and the preset dynamic obstacle quantity risk index within that preset range, the higher the current vehicle risk index. The smaller; Among the dynamic obstacles within the preset range, the first one is... The ratio of the number of dynamic obstacle types to the number of dynamic obstacles within a preset range; Among the dynamic obstacles within the preset range, the first one is... The probability of violation for each type of dynamic obstacle; The probability of a violation involving dynamic obstacles within a preset range; the higher the probability of a violation, the greater the risk, therefore the current vehicle risk index. The higher the value, the lower the probability of violation and the lower the risk; therefore, the current vehicle risk index... The smaller the value, the better; Autonomous vehicle obstacle avoidance index The higher the value, the better the obstacle avoidance performance of the autonomous vehicle; therefore, the current vehicle risk index... The smaller the value, the better; Autonomous vehicle obstacle avoidance index The smaller the value, the worse the obstacle avoidance performance of the autonomous vehicle; therefore, the current vehicle risk index... The larger the value, the greater the system's intelligence level is achieved by considering the probability of violations for different types of dynamic obstacles. Different types of dynamic obstacles have different driving characteristics and violation tendencies. For example, pedestrians may suddenly cross the road, while bicycles may change lanes arbitrarily. By accurately calculating the probability of violations for each type of dynamic obstacle and incorporating it into the risk assessment system, the autonomous driving system can more effectively respond to various complex situations, predict potential dangers in advance, and thus better ensure driving safety. At this time, the current vehicle risk level is low; this indicates that the surrounding environment is relatively safe, dynamic obstacles have little impact on vehicle operation, and the vehicle's obstacle avoidance is good. In this case, the obstacle avoidance strategy can be to maintain the current speed and direction, simply by continuously monitoring changes in the surrounding environment to ensure smooth driving without sudden changes in the risk level. For example, on an open suburban road with few and orderly surrounding vehicles, the vehicle can drive normally along a preset route; when... At this time, the current vehicle risk level is medium risk; this means there are certain potential dangers around the vehicle, such as dynamic obstacles that may affect its driving, or the vehicle's obstacle avoidance configuration, while capable of handling general situations, is not optimal. In this case, the obstacle avoidance strategy could be to appropriately reduce speed, increase the safe distance from vehicles in front and behind, as well as surrounding obstacles, and simultaneously increase the frequency of monitoring the surrounding environment. When the current vehicle risk level is high risk, it indicates that the vehicle is in an extremely dangerous driving environment, and may face serious collision risks, or the vehicle is difficult to deal with the current complex situation due to poor obstacle avoidance state; at this time, the obstacle avoidance strategy should immediately take emergency braking measures, and turn on the danger warning light to remind the surrounding vehicles and pedestrians to pay attention. If the vehicle has the function of automatic steering obstacle avoidance, it can try to avoid the obstacle by automatic steering under the premise of ensuring safety;

[0117] It should be noted that the preset dynamic obstacle quantity risk index in the preset range , the obstacle quantity weight coefficient , the obstacle risk weight coefficient , the obstacle quantity preset constant , the obstacle preset constant and the preset threshold are preset values obtained according to experience, which are not described here in detail.

[0118] The above describes one embodiment of the present application in detail, but the content described is only a preferred embodiment of the present application, and cannot be considered as limiting the scope of the present application. Any equivalent changes and improvements made according to the scope of the present application should still belong to the scope of the present application.

Claims

1. A visual large model based autonomous dynamic obstacle prediction system, characterized in that, The system comprises: a road division module for dividing roads in a target area into a plurality of road units; a time division module for dividing a day into a plurality of time units; a plurality of road unit acquisition modules corresponding to the road units respectively for acquiring road image information data and traffic signal information data of the corresponding road units; a road unit risk assessment module for analyzing the road image information data and the traffic signal information data of each road unit to obtain a comprehensive road risk index of the corresponding road unit in each time unit; a vehicle information acquisition module arranged on an unmanned vehicle for acquiring dynamic obstacle information data within a preset range of the unmanned vehicle; a vehicle obstacle avoidance assessment module for performing obstacle avoidance detection on the unmanned vehicle before a task to obtain obstacle avoidance information data, and analyzing the obstacle avoidance information data to obtain an obstacle avoidance state index of the unmanned vehicle; a vehicle driving risk assessment module for analyzing the comprehensive road risk index of the road unit where the unmanned vehicle is located in the current time unit, the dynamic obstacle information data, and the obstacle avoidance state index of the unmanned vehicle to obtain a current vehicle risk level; an obstacle avoidance module for determining an obstacle avoidance strategy according to the current vehicle risk level; The road unit risk assessment module comprises an identification unit and an analysis unit; the identification unit is a trained convolutional neural network model for dynamic obstacle type identification according to the road image information data to obtain the dynamic obstacle type of each dynamic obstacle on the road image information data; the analysis unit analyzes the movement rules of each dynamic obstacle in the road unit according to the corresponding dynamic obstacle type to determine whether each dynamic obstacle is compliant in the road unit; The analysis unit further analyzes the dynamic obstacle type of each dynamic obstacle in the road unit according to the corresponding dynamic obstacle type to determine whether each dynamic obstacle is compliant in the road unit; ； calculating a composite road risk index for any one road element in any one time element ; in, For any road unit; For any unit of time; Preset the number of days in the past. ; The number of dynamic obstacle types. ; For this road unit, the number of days in the past preset period The first day within that time unit Number of compliant dynamic obstacle types; For this road unit, the number of days in the past preset period The first day within that time unit The total number of dynamic obstacle types; For the first Risk weighting coefficients for each dynamic obstacle type.

2. The visual large model based dynamic obstacle prediction system for autonomous vehicles of claim 1, wherein, The working process of the vehicle obstacle avoidance assessment module is as follows: S1: obtaining an environmental impact coefficient according to environmental information data; the environmental information data includes environmental temperature, visibility, rainfall intensity, and wind intensity; S2: controlling the unmanned vehicle to move at a preset speed in a preset direction; S3: controlling the preset detection dynamic obstacle to move at a random speed from the front of the unmanned vehicle to the unmanned vehicle at a constant speed, obtaining the distance between the unmanned vehicle and the dynamic obstacle at the beginning of this obstacle avoidance, the completion time of this obstacle avoidance of the unmanned vehicle, and the minimum distance between the unmanned vehicle and the dynamic obstacle in this obstacle avoidance process; S4: analyzing the distance between the unmanned vehicle and the dynamic obstacle at the beginning of this obstacle avoidance, the completion time of this obstacle avoidance of the unmanned vehicle, and the minimum distance between the unmanned vehicle and the dynamic obstacle in this obstacle avoidance process to obtain the obstacle avoidance state index of the unmanned vehicle.

3. The visual large model based dynamic obstacle prediction system for autonomous vehicles of claim 2, wherein, In step S1, the environmental impact coefficient is obtained by the formula: ； Computing environmental impact coefficient ; wherein, is a first judgment function when , ; when , ; is a current ambient temperature; is a preset ambient temperature; is a preset temperature allowable error value; is a preset visibility; is a current visibility; is a current wind strength; is a preset wind strength; is a current rainfall intensity; is a preset rainfall intensity; is a first weight coefficient; is a second weight coefficient; is a third weight coefficient; is a fourth weight coefficient; is a first preset constant; is a second preset constant; is a third preset constant; is a fourth preset constant.

4. The visual large model based dynamic obstacle prediction system for autonomous vehicles of claim 3, wherein, In step S4, the obstacle avoidance state index of the unmanned vehicle is obtained by the formula: ； Computing an obstacle avoidance state index for an autonomous vehicle ; wherein, is the distance between the autonomous vehicle and the dynamic obstacle at the beginning of the current obstacle avoidance; is the preset distance between the autonomous vehicle and the dynamic obstacle at the beginning of the obstacle avoidance; is the preset completion time of the obstacle avoidance of the autonomous vehicle; is the completion time of the current obstacle avoidance of the autonomous vehicle; is the minimum distance between the autonomous vehicle and the dynamic obstacle during the current obstacle avoidance; is the preset minimum distance between the autonomous vehicle and the dynamic obstacle during the obstacle avoidance; is the first weight coefficient; is the second weight coefficient; is the third weight coefficient; is the first preset constant; is the second preset constant; is the third preset constant; is the first preset adjustment coefficient; is the second preset adjustment coefficient; is the third preset adjustment coefficient.

5. The visual large model based dynamic obstacle prediction system for autonomous vehicles of claim 4, wherein, The preset range is obtained in the following way: S10: analyzing the comprehensive road risk index of the road unit where the unmanned vehicle is located in the current time unit to obtain the current range scaling ratio of the road unit; S20: adjusting the standard range according to the current range scaling ratio to obtain the current preset range of the unmanned vehicle.

6. The visual large model based dynamic obstacle prediction system for autonomous vehicles of claim 5, wherein, In step S10, the current range scaling ratio is obtained by the formula: ； Computing a current range scaling for an unmanned vehicle ; wherein, is a comprehensive road risk index corresponding to a current time unit of a road unit where the unmanned vehicle is located; is a preset comprehensive road risk index; is a preset constant of proportionality.

7. The visual large model based dynamic obstacle prediction system for autonomous vehicles of claim 6, wherein, The dynamic obstacle information data includes a dynamic obstacle number and a dynamic obstacle type.

8. The visual large model based dynamic obstacle prediction system for autonomous vehicles of claim 7, wherein, Through the formula: ； Calculating a current vehicle risk index ; wherein, is the number of dynamic obstacles of the type of the first dynamic obstacle in the preset range; is the preset dynamic obstacle number risk index for the preset range; is the number of dynamic obstacles in the preset range; is the obstacle number weight coefficient; is the obstacle risk weight coefficient; is the obstacle number preset constant; is the obstacle preset constant; comparing the current vehicle risk index with a preset threshold value; When the current vehicle risk level is low risk; When the current vehicle risk level is medium risk; When the current vehicle risk level is high risk.

Images