A highway automatic driving risk identification system based on a multi-modal AI model

Abstract

The application discloses a kind of highway automatic driving risk identification systems based on multi-modal AI model, it is related to risk identification technical field.The information acquisition module: with own vehicle as determination midpoint, set determination range, based on determination range, the speed information of target vehicle in determination range is acquired, first data set is obtained, while the swing information of target vehicle in determination range is acquired, second data set is obtained.The application realizes systematic risk assessment by the dynamic determination range of information acquisition module and multi-source data fusion, determination range is dynamically adjusted based on initial value in combination with real-time speed and road width, ensure to cover the key area around vehicle, at the same time, information acquisition module utilizes vehicle-mounted radar cluster and multi-view vision system, the speed information and swing information of target vehicle are collected in fixed window value, first data set and second data set are generated, provide multi-dimensional environmental data.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of risk identification, in particular to a highway automatic driving risk identification system based on a multi-modal AI model. BACKGROUND

[0002] In an automatic driving system, risk identification is a process of systematic, continuous identification, evaluation and monitoring of risk factors that may cause system performance degradation, loss of control, accidents or harm to passengers and pedestrians, the goal is to discover potential risks in advance and take measures to reduce or control these risks, by analyzing environmental, system itself, algorithm decision, sensor, communication, human-computer interaction and other multi-dimensional factors, finding out factors that may cause incorrect judgment, operation out of control, slow reaction, etc., and developing appropriate mitigation strategies, it is a key link in the safety engineering system, often closely linked with risk assessment, fault tree analysis, event tree analysis, safety requirement engineering and operation safety monitoring, etc., in the process of highway automatic driving, risk identification is also very important.

[0003] The patent with publication number CN117719524A discloses a driving safety risk identification and early warning method, device, terminal and storage medium, which obtains a risk level identification result according to collected vehicle driving information and road condition information in a current time window based on a pre-trained conditional random field model; issues a driving risk warning according to the risk level identification result; wherein the conditional random field model includes a first conditional random field model and / or a second conditional random field model, the first conditional random field model is trained by a pre-constructed lateral driving training data set, and the second conditional random field model is trained by a pre-constructed longitudinal driving training data set, the model is divided into two models of lateral driving and longitudinal driving, which can analyze driving behavior specifically, and distinguish different risks produced by driving behavior under different road conditions, thereby improving the identification accuracy of driving behavior and the accuracy of early warning.

[0004] The above and similar technical solutions have a large blind area in the process of automatic driving, because conventional risk analysis is mostly based on single vehicle perception of the preceding vehicle, for example, detecting the distance from the preceding vehicle and the speed of the preceding vehicle, and taking appropriate braking measures, but lacking systematic evaluation of surrounding vehicles, unable to dynamically adapt to complex environments, and thus unable to make systematic risk judgments according to the specific driving information of surrounding vehicles, resulting in unsatisfactory risk identification effect. SUMMARY

[0005] The present application aims to provide a highway automatic driving risk identification system based on a multi-modal AI model to solve the problems raised in the background.

[0006] To achieve the above object, the application provides the following technical scheme: a highway automatic driving risk identification system based on a multi-modal AI model, comprising:

[0007] An information acquisition module: taking the own vehicle as a judgment midpoint, setting a judgment range, acquiring speed information of a target vehicle in the judgment range based on the judgment range to obtain a first data set, and simultaneously acquiring swing information of the target vehicle in the judgment range to obtain a second data set;

[0008] A risk calculation module: setting a first risk assessment formula based on the first data set to calculate a first risk value of the target vehicle, setting a second risk assessment formula based on the second data set to calculate a second risk value of the target vehicle, and evaluating a comprehensive risk index of the target vehicle based on a combination result of the first risk value and the second risk value to obtain a risk index of the target vehicle in the judgment range and obtain a comprehensive risk item;

[0009] A position adjustment module: based on the comprehensive risk item, obtaining an adjustment point through an adjustment method, taking the adjustment point as a target point, obtaining an adjustment path, and adjusting the position of the own vehicle based on the adjustment path;

[0010] A risk updating module: in the process of adjusting the position of the own vehicle, repeatedly calculating the first risk value and the second risk value of the target vehicle in the judgment range based on the AI model to update the adjustment point in real time, thereby realizing systematic risk evaluation of surrounding vehicles and real-time change of the position of the own vehicle according to the risk evaluation result.

[0011] Further, the setting method of the judgment range comprises:

[0012] Setting an initial judgment value, the initial judgment value being a fixed range value;

[0013] Acquiring speed information of the own vehicle to obtain a real-time speed item, and obtaining a preselected range based on a combination result of the real-time speed item and the initial judgment value;

[0014] Acquiring road width information to obtain a limiting width item, taking the limiting width item as a width limit of the preselected range, and further obtaining the judgment range.

[0015] Further, the speed information comprises instantaneous acceleration and instantaneous deceleration, and the first data set at least comprises a first data item composed of speed information of one target vehicle, and the acquisition method of the first data set comprises:

[0016] Setting an acquisition window value, the acquisition window value being a fixed time threshold, and acquiring acceleration information and deceleration information of the target vehicle within the acquisition window value through a vehicle-mounted radar cluster to obtain a first acquisition information item;

[0017] A first determination threshold is set, the first determination threshold is a fixed acceleration value and a fixed deceleration value, a first acquisition information item is determined, target information exceeding the first determination threshold in the first acquisition information item is acquired, and a first target index item is obtained;

[0018] A speed limit value including an acceleration limit value and a deceleration limit value is acquired based on the first target index item, a first target numerical value item is obtained, a first target time item is acquired based on an accumulated braking time of the first target index item, and a first data item is obtained by combining the first target numerical value item and the first target time item. The first data item is combined to obtain a first data set.

[0019] Further, the swing information includes a lateral swing distance, the second data includes at least a second data item composed of swing information of a target vehicle, and the second data set is acquired by the following method:

[0020] Based on the acquisition window value, a vehicle lateral swing distance of the target vehicle within the acquisition window value is acquired by a multi-view visual system to obtain a second acquisition information item;

[0021] A second determination threshold is set, the second determination threshold is a fixed lateral swing distance range, a second acquisition information item is determined, target information exceeding the second determination threshold in the second acquisition information item is acquired, and a second target index item is obtained;

[0022] Based on the second target index item, swing overshoot data of the target vehicle is acquired to obtain a second data item, and a second data set is obtained by combining the second data item.

[0023] Further, the first risk value acquisition method includes:

[0024] A first risk assessment formula is created:

[0025] ;

[0026] Wherein is the first risk value, is the instantaneous deceleration, is the dangerous deceleration limit value, is the instantaneous acceleration, is the dangerous acceleration limit value, , and is the initial weight coefficient, is the braking time.

[0027] Further, the second risk value acquisition method includes:

[0028] A first risk assessment formula is created:

[0029] ;

[0030] wherein is a second risk value, is a lateral swing distance, is a lateral swing reference value, is a vehicle speed, is a reference vehicle speed, is a sensitivity coefficient.

[0031] Further, the adjusting method comprises path adjustment, and the method for obtaining the adjustment point comprises:

[0032] a risk evaluation value is set, the risk evaluation value is a fixed value, the comprehensive risk term is judged based on the risk evaluation value, a high risk range and a low risk range are obtained;

[0033] a risk avoidance range is set, the risk avoidance range is a fixed range value, position information of the target vehicle in the high risk range is obtained, a target position term is obtained, a risk avoidance range set is obtained based on a combination result of the risk avoidance range and the target position term;

[0034] based on the elimination result of the risk avoidance range set by the determination range, a position closest to the determination midpoint is obtained as an adjustment point, and an adjustment path is generated.

[0035] Further, the adjusting method further comprises speed adjustment, and the method for obtaining the adjustment point comprises:

[0036] based on the elimination result of the risk avoidance range set by the determination range, a position closest to the determination midpoint is obtained as a pre-adjustment point, and an adjustment path is generated;

[0037] speed information of the ego vehicle and the target vehicle in the high risk range is obtained, the speed information is judged, when the speed information of the ego vehicle exceeds the speed information of the target vehicle in the high risk range, the position of the ego vehicle is adjusted based on the adjustment path, the pre-adjustment point is taken as the adjustment point, when the speed information of the ego vehicle does not exceed the speed information of the target vehicle in the high risk range, the pre-adjustment point is taken as an eliminated adjustment point, a position closest to the determination midpoint is obtained again as a pre-adjustment point, an adjustment path is generated, and the step is repeated until the position of the ego vehicle is adjusted.

[0038] Compared with the prior art, the present application has the following beneficial effects:

[0039] The highway automatic driving risk identification system based on the multi-modal AI model effectively overcomes the blind area problem of traditional single vehicle perception through the dynamic determination range of the information acquisition module and multi-source data fusion, realizes systematic risk assessment, and the determination range is dynamically adjusted based on the initial value in combination with the real-time speed and road width, ensures to cover the key areas around the vehicle, avoids the limited field of view caused by focusing on the front vehicle only, at the same time, the information acquisition module uses the vehicle-mounted radar cluster and multi-view vision system to collect the speed information and swing information of the target vehicle within the fixed window value, generates the first data set and the second data set, provides multi-dimensional environmental data, and lays a foundation for risk calculation, and the risk calculation module integrates the first risk value and the second risk value, and comprehensively outputs the risk index, quantifies the risk of the surrounding vehicles, and makes the risk identification more comprehensive and real-time.

[0040] At the same time, the position adjustment module realizes intelligent position adjustment based on the comprehensive risk term, actively avoids high-risk targets, thereby greatly improving driving safety and efficiency, when the risk index exceeds the threshold value, the system positions the high-risk vehicle position through the risk avoidance range, generates an adjustment path and acquires the nearest adjustment node, reduces the collision probability, in addition, the risk updating module continuously updates the data, supports dynamic adaptation to environmental changes, this active control mechanism not only solves the deficiency of passive braking of the traditional system, but also realizes preventive risk avoidance through systematic risk determination, and is suitable for complex scenes such as highways. BRIEF DESCRIPTION OF DRAWINGS

[0041] Figure 1 It is a whole process schematic diagram of the application;

[0042] Figure 2 It is a determination range acquisition process schematic diagram of the application;

[0043] Figure 3 It is a preselected range and determination range relationship schematic diagram of the application;

[0044] Figure 4 It is a swing amplitude structure schematic diagram of the application;

[0045] Figure 5 It is an adjustment node position schematic diagram of the application. DETAILED DESCRIPTION

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

[0047] The traditional automatic driving risk analysis method is often based on single vehicle perception, that is, mainly relying on the information obtained by the vehicle's own sensors such as laser radar and camera, especially detecting the distance and speed of the front vehicle, and taking corresponding braking or other avoidance measures accordingly. This method has obvious limitations, resulting in unsatisfactory risk identification effect. Firstly, single vehicle perception has a large blind area. In addition, the structure of the vehicle itself also causes a certain blind area. For example, the B-pillar of the vehicle may block the side rear view. These blind areas make it difficult for the automatic driving system to fully understand the surrounding environment, and thus potential risks may be missed. Secondly, single vehicle perception lacks systematic evaluation of surrounding vehicles. Traditional risk analysis mainly focuses on the front vehicle, while ignoring the behavior of surrounding vehicles. However, in a complex traffic environment, the driving information of surrounding vehicles such as speed, acceleration, and steering angle is crucial to the safety of the autonomous vehicle. For example, a vehicle located on the side of the autonomous vehicle suddenly changes lanes, or a rear vehicle approaches at high speed, which may pose a threat to the autonomous vehicle. If the autonomous driving system cannot identify and respond to these risks in a timely manner, it may lead to traffic accidents. The technical solution provided in the present application effectively overcomes the blind area problem of traditional single vehicle perception through dynamic determination of the range of the information acquisition module and multi-source data fusion, realizes systematic risk evaluation, and dynamically adjusts the determination range based on the initial value combined with the real-time speed and road width to ensure coverage of the key areas around the vehicle, avoiding the limited view caused by focusing only on the front vehicle. At the same time, the information acquisition module uses the vehicle-mounted radar cluster and multi-view vision system to collect the speed information and swing information of the target vehicle within a fixed window value, generating a first data set and a second data set, providing multi-dimensional environmental data to lay the foundation for risk calculation. The risk calculation module integrates the first risk value and the second risk value to comprehensively output the risk index, quantifying the risk of surrounding vehicles, making the risk identification more comprehensive and real-time, and all the calculation processes are automatically calculated by the AI model, making the result acquisition faster and more accurate, as shown in Figure 1 The technical solution provided in the present application effectively overcomes the blind area problem of traditional single vehicle perception through dynamic determination of the range of the information acquisition module and multi-source data fusion, realizes systematic risk evaluation, and dynamically adjusts the determination range based on the initial value combined with the real-time speed and road width to ensure coverage of the key areas around the vehicle, avoiding the limited view caused by focusing only on the front vehicle. At the same time, the information acquisition module uses the vehicle-mounted radar cluster and multi-view vision system to collect the speed information and swing information of the target vehicle within a fixed window value, generating a first data set and a second data set, providing multi-dimensional environmental data to lay the foundation for risk calculation. The risk calculation module integrates the first risk value and the second risk value to comprehensively output the risk index, quantifying the risk of surrounding vehicles, making the risk identification more comprehensive and real-time, and all the calculation processes are automatically calculated by the AI model, making the result acquisition faster and more accurate, as shown in

[0048] Information acquisition module: taking the own vehicle as the determination midpoint, setting the determination range, obtaining the speed information and swing information of the target vehicle within the determination range based on the determination range.

[0049] It should be noted that, as shown in Figure 2 The setting method of the determination range includes: setting an initial determination value, the initial determination value is a fixed range value, which is 2m; obtaining the speed information of the own vehicle to obtain the real-time speed item; based on the combination result of the real-time speed item and the initial determination value, a preselected range is obtained; obtaining the road width information to obtain the limit width item, taking the limit width item as the width limit of the preselected range, and then obtaining the determination range.

[0050] It should be noted that the initial determination value is set to 2m, and based on the combination of the real-time speed term and the initial determination value, a preselected range is obtained, and because the vehicle is driving on a highway, the single side is a single lane, so the width of the single side single lane can be used to limit the width of the preselected range, and then the determination range is obtained.

[0051] In the specific implementation process, for example Figure 3 , the existing vehicle is automatically driving on the highway, and the point is the determination midpoint, and at this time the vehicle speed is 100km / h, according to the set initial determination value, the preselected range obtained at this time is 200m, that is, the range is circled with a radius of 200m, and the width of the highway is 11.25m, at this time the 11.25m is used as the limiting width term to limit the width of the preselected range, and then the determination range is obtained.

[0052] It should be noted that the speed information of the target vehicle in the determination range is obtained, and a first data set is obtained, the speed information including instantaneous acceleration and instantaneous deceleration, the first data set including at least a first data item composed of the speed information of a target vehicle, and the method for obtaining the first data set includes: setting an acquisition window value, the acquisition window value being a fixed time threshold, based on the acquisition window value, obtaining the acceleration information and deceleration information of the target vehicle within the acquisition window value through the vehicle-mounted radar cluster to obtain a first acquisition information item; setting a first determination threshold, the first determination threshold being a fixed acceleration value and a fixed deceleration value, determining the first acquisition information item, obtaining the target information in the first acquisition information item that exceeds the first determination threshold to obtain a first target index item; based on the first target index item, obtaining a speed limit value including an acceleration limit value and a deceleration limit value to obtain a first target value item, and simultaneously obtaining a cumulative braking time of the first target index item to obtain a first target time item, the first target value item and the first target time item are combined to obtain the first data item, and the first data item is combined to obtain the first data set.

[0053] It should be noted that the acquisition window value is set to 5s, that is, within the window time of 5s, the acceleration information and deceleration information of the target vehicle are obtained through the vehicle-mounted radar cluster, the first determination threshold is set to ±3m / s², that is, the target information of the target vehicle within the window time of 5s whose acceleration or deceleration exceeds 3m / s² is obtained to obtain the first target index item, based on the first target index item, the speed limit value including the acceleration limit value and the deceleration limit value is obtained to obtain the first target value item, and when accelerating or decelerating, there is a braking time, and the cumulative braking time of the first target index item is obtained to obtain the first target time item, the first target value item and the first target time item are combined to obtain the first data item, and the first data item is combined to obtain the first data set.

[0054] In the specific implementation process, a target vehicle a is in automatic driving on a highway, at this time its speed is 100 km / h, the set determination range is 200 m, there are three target vehicles b, c and d within the range of 200 m, at this time the acceleration information and deceleration information of b, c and d are obtained through the vehicle-mounted radar cluster on a vehicle, and the information is shown in Table 1:

[0055] Table 1

[0056] Target vehicle b c d Acceleration / deceleration information 1 3 m / s2 4 m / s2 6.5 m / s2 Cumulative braking time 1 1s 1s 2s Acceleration / deceleration information 2 -2 m / s2 2.8 m / s2 5 m / s2 Cumulative braking time 2 2s 1s 2s Acceleration / deceleration information 3 1 m / s2 2.7 m / s2 2.8 m / s2 Cumulative braking time 3 3s 2s 5s Acceleration / deceleration information 4 2 m / s2 -2.5 m / s2 -3 m / s2 Cumulative braking time 5 3s 3s 5s

[0057] At this time, the first target index item obtained is that b is the acceleration / deceleration information 1, c is the acceleration / deceleration information 1, and d is the acceleration / deceleration information 1, the acceleration / deceleration information 2 and the acceleration / deceleration information 4, the speed limit value is obtained based on the first target index item, including the acceleration limit value and the deceleration limit value, the first target value item is obtained, and the corresponding time is obtained to obtain the first target time item, wherein b is 3 m / s2 and 1 s, c is 4 m / s2 and 1 s, and d is 6.5 m / s2 and 2.5 s.

[0058] It should be noted that the swing information of the target vehicle in the determination range is obtained to obtain a second data set, the swing information includes a lateral swing distance, the second data at least includes a second data item composed of the swing information of a target vehicle, and the second data set is obtained by the following method: based on an acquisition window value, the lateral swing distance of the target vehicle within the acquisition window value is obtained through a multi-view visual system to obtain a second acquisition information item; a second determination threshold is set, the second determination threshold is a fixed lateral swing distance range, the second acquisition information item is determined, the target information exceeding the second determination threshold in the second acquisition information item is obtained to obtain a second target index item; the swing overshoot data of the target vehicle is obtained based on the second target index item to obtain a second data item, and the second data item is combined to obtain the second data set.

[0059] It should be noted that the same acquisition window value is 5 s, the lateral swing distance of the target vehicle within the acquisition window value is obtained through the multi-view visual system, that is, the lateral displacement, the second determination threshold is set to 0.5 m, the target information exceeding the lateral swing distance of 0.5 m is set as the second index information, and the swing overshoot data in the second index information is determined as the second data item.

[0060] In the specific implementation process, as shown in Figure 4 a target vehicle aa is in automatic driving on a highway, at this time its speed is 100 km / h, the set determination range is 200 m, there are three target vehicles bb, cc and dd within the range of 200 m, at this time the lateral swing distance of bb, cc and dd is obtained through the multi-view visual system on aa vehicle, and the information is shown in Table 2:

[0061] Table 2

[0062] Target vehicle bb cc dd Lateral swing distance 1 0.3m 0.6m 0.8m Lateral swing distance 2 0.4m 0.7m 0.7m Lateral swing distance 3 0.3m 0.6m 0.3m Lateral swing distance 4 0.2m 0.4m 0.4m Lateral swing distance 5 0.7m 0.4m 0.8m

[0063] The target information exceeding 0.5m lateral swing distance is set as the second target information, that is, the second target information is: bb is lateral swing distance 5, cc is lateral swing distance 1, lateral swing distance 2 and lateral swing distance 3, dd is lateral swing distance 1, lateral swing distance 2 and lateral swing distance 5, the swing overshoot data of the target vehicle is obtained based on the second target index item, and the second data item is obtained, that is, the swing overshoot data of the target vehicle is: bb is 0.5m, cc is 0.4m, and dd is 0.8m.

[0064] The risk calculation module sets the first risk evaluation formula and the second risk evaluation formula to calculate the first risk value and the second risk value of the target vehicle, and evaluates the comprehensive risk index of the target vehicle.

[0065] It should be noted that the first risk value of the target vehicle is calculated based on the first data set by setting the first risk evaluation formula, the second risk value of the target vehicle is calculated based on the second data set by setting the second risk evaluation formula, and the comprehensive risk index of the target vehicle is evaluated based on the combination result of the first risk value and the second risk value, so as to obtain the risk index of the target vehicle in the judgment range and obtain the comprehensive risk item.

[0066] It should be noted that the first risk value is obtained by:

[0067] The first risk evaluation formula is created as:

[0068] ;

[0069] Wherein is the first risk value, is the instantaneous deceleration, is the dangerous deceleration limit value, and the set value is 8m / s², is the instantaneous acceleration, is the dangerous acceleration limit value, and the set value is 4m / s², , and are initial weight coefficients, and the numerical values are 0.7, 0.5 and 0.3 respectively, is the braking time, that is, the first target time item;

[0070] The second risk value is obtained by:

[0071] The first risk evaluation formula is created as:

[0072] ;

[0073] wherein is a second risk value, is a lateral swing distance, is a lateral swing reference value, and is 0.5 m, is a vehicle speed, is a reference vehicle speed, and a set value is 120 km / h, is a sensitivity coefficient.

[0074] In a specific implementation process, the existing vehicle is driving on a highway, at this time the vehicle speed is 100 km / h, the set determination range is 200 m, and there are two target vehicles in the 200 m range, which are a and b, at this time the vehicle speed of a is obtained through the vehicle-mounted radar cluster and the multi-view vision system, which is 90 km / h, the deceleration is 6.5 m / s², the time is 2 s, the lateral swing distance is 0.8 m, the vehicle speed of b is 80 km / h, the acceleration is 5.8 m / s², the time is 1.5 s, and the lateral swing distance is 1.2 m, at this time the first risk value and the second risk value of a and b are calculated according to the first risk evaluation formula and the second risk evaluation formula respectively:

[0075] ;

[0076] ;

[0077] ;

[0078] ;

[0079] Therefore, the comprehensive risk value of a is 1.93, and the comprehensive risk value of b is 1.4.

[0080] Position adjustment module: based on the comprehensive risk term, the adjustment point is obtained through the adjustment method, the adjustment point is taken as the target point, the adjustment path is obtained, and the position of the vehicle itself is adjusted based on the adjustment path.

[0081] It should be noted that the adjustment method includes path adjustment, and the adjustment point is obtained by the following method: setting a risk evaluation value, the risk evaluation value is a fixed value, the risk evaluation value is 1.5, the comprehensive risk term is determined based on the risk evaluation value, and a high risk range and a low risk range are obtained; set the risk avoidance range, the risk avoidance range is a fixed range value, which is 100 m, obtain the position information of the corresponding target vehicle in the high risk range to obtain the target position term, and obtain the risk avoidance range set based on the combination result of the risk avoidance range and the target position term; based on the elimination result of the risk avoidance range set in the determination range, the position closest to the determination midpoint is obtained as the adjustment point, and the adjustment path is generated.

[0082] In a specific implementation process, asFigure 5 As shown, the existing vehicle is driving on the highway, at this time its speed is 100 km / h, the set determination range is 200 m, there are two target vehicles in the 200 m range, which are c and d, respectively, c is 80 m in front, d is 80 m behind, and the comprehensive risk value of c is 1.8 and the comprehensive risk value of d is 2.1, both of which exceed the risk evaluation value, so c and d are both high-risk ranges, at this time, according to the set risk avoidance range, the risk avoidance range set in the determination range is determined, based on the elimination result of the risk avoidance range set in the determination range, the position closest to the determination midpoint is obtained as the adjustment node, and the adjustment path is generated.

[0083] It should be noted that the adjustment method also includes speed adjustment, and the method for obtaining the adjustment node includes: based on the elimination result of the risk avoidance range set in the determination range, the position closest to the determination midpoint is obtained as the pre-adjustment node, and the adjustment path is generated; obtain the speed information of the target vehicle in the high-risk range and the target vehicle corresponding to the high-risk range, and determine the speed information, when the speed information of the vehicle exceeds the speed information of the target vehicle in the high-risk range, adjust the position of the vehicle based on the adjustment path, and when the speed information of the vehicle does not exceed the speed information of the target vehicle in the high-risk range, the pre-adjustment node is used as the elimination adjustment node, and the position closest to the determination midpoint is obtained again as the pre-adjustment node. The adjustment path is generated, and this step is repeated until the position adjustment of the vehicle is completed.

[0084] Specifically, when the speed information of the vehicle does not exceed the speed information of the target vehicle in the high-risk range, it means that the position adjustment cannot be performed in the risk avoidance range set at this time, so the pre-adjustment node is used as the elimination adjustment node, and the position closest to the determination midpoint is obtained again as the pre-adjustment node. The adjustment path is generated, and the speed information of the vehicle and the speed information of the target vehicle in the high-risk range are judged again until the position adjustment of the vehicle is completed.

[0085] Risk updating module: during the position adjustment of the vehicle, the first risk value and the second risk value of the target vehicle in the determination range are repeatedly calculated based on the AI model, and the adjustment node is updated in real time.

[0086] It should be noted that the adjustment node is updated in real time, and the systematic risk assessment of the surrounding vehicles is realized, and the position of the vehicle is changed in real time according to the risk assessment result. At the same time, the information acquisition module, the risk calculation module, the position adjustment module and the risk updating module are integrated in the system, and the data is acquired and calculated by the AI model, so the result is obtained quickly and accurately.

[0087] While embodiments of the application have been shown and described, it is to be understood that the embodiments described are merely exemplary of the principles and application of the present application. Numerous modifications and adaptions can be effected without departing from the spirit and scope of the present application, which is not limited to the embodiments disclosed except insofar as recited in the claims.

Claims

1. A highway automatic driving risk identification system based on a multi-modal AI model, comprising: an information acquisition module: taking the vehicle itself as the judgment midpoint, setting the judgment range, based on the judgment range, acquiring the speed information of the target vehicle in the judgment range to obtain a first data set, and simultaneously acquiring the swing information of the target vehicle in the judgment range to obtain a second data set; characterized in that: a risk calculation module: based on the first data set, a first risk assessment formula is set to calculate the first risk value of the target vehicle, based on the second data set, a second risk assessment formula is set to calculate the second risk value of the target vehicle, based on the combination of the first risk value and the second risk value, the comprehensive risk index of the target vehicle is evaluated, and then the risk index of the target vehicle in the judgment range is obtained, and the comprehensive risk item is obtained; a position adjustment module: based on the comprehensive risk item, an adjustment point is obtained through an adjustment method, the adjustment point is taken as a target point, an adjustment path is obtained, and the position of the vehicle itself is adjusted based on the adjustment path; a risk updating module: in the process of adjusting the position of the vehicle itself, based on the AI model, the first risk value and the second risk value of the target vehicle in the judgment range are repeatedly calculated, the adjustment point is updated in real time, and then the systematic risk assessment of the surrounding vehicles is realized and the position of the vehicle itself is changed in real time according to the risk assessment result; the speed information includes instantaneous acceleration and instantaneous deceleration, the first data set at least includes a first data item composed of the speed information of a target vehicle, and the acquisition method of the first data set includes: setting an acquisition window value, the acquisition window value is a fixed time threshold, based on the acquisition window value, the acceleration information and the deceleration information of the target vehicle within the acquisition window value are obtained through a vehicle-mounted radar cluster to obtain a first acquisition information item; setting a first judgment threshold, the first judgment threshold is a fixed acceleration value and a fixed deceleration value, the first acquisition information item is judged, the target information exceeding the first judgment threshold in the first acquisition information item is obtained, and a first target index item is obtained; based on the first target index item, the speed limit value including the acceleration limit value and the deceleration limit value is obtained, a first target numerical item is obtained, the cumulative braking time of the first target index item is obtained, a first target time item is obtained, the first target numerical item and the first target time item are combined to obtain the first data item, and the first data item is combined to obtain the first data set; the swing information includes lateral swing distance, the second data at least includes a second data item composed of the swing information of a target vehicle, and the acquisition method of the second data set includes: based on the acquisition window value, the vehicle lateral swing distance of the target vehicle within the acquisition window value is obtained through a multi-view visual system to obtain a second acquisition information item; setting a second judgment threshold, the second judgment threshold is a fixed lateral swing distance range, the second acquisition information item is judged, the target information exceeding the second judgment threshold in the second acquisition information item is obtained, and a second target index item is obtained; based on the second target index item, the swing amplitude data of the target vehicle is obtained to obtain a second data item, and the second data item is combined to obtain a second data set.

2. The expressway automatic driving risk identification system based on a multi-modal AI model according to claim 1, characterized in that: the setting method of the judgment range includes: An initial determination value is set, and the initial determination value is a fixed range value; Speed information of the ego vehicle is obtained to obtain a real-time speed item, and a preselected range is obtained based on a combination result of the real-time speed item and the initial determination value; Road width information is obtained to obtain a limit width item, the limit width item is used as a width limit of the preselected range, and a determination range is obtained.

3. The expressway automatic driving risk identification system based on a multi-modal AI model according to claim 1, characterized in that: The first risk value acquisition method includes: A first risk assessment formula is created: ； wherein is a first risk value, is an instantaneous deceleration, is a dangerous deceleration limit value, is an instantaneous acceleration, is a dangerous acceleration limit value, , and is an initial weight coefficient, is a braking time.

4. The expressway automatic driving risk identification system based on a multi-modal AI model according to claim 1, characterized in that: The second risk value acquisition method includes: A second risk assessment formula is created: ； wherein is a second risk value, is a lateral swing distance, is a lateral swing reference value, is a vehicle speed, is a reference vehicle speed, is a sensitivity coefficient.

5. The expressway automatic driving risk identification system based on a multi-modal AI model according to claim 1, characterized in that: The adjustment method includes path adjustment, and the adjustment point acquisition method includes: A risk judgment value is set, and the risk judgment value is a fixed numerical value; a comprehensive risk item is judged based on the risk judgment value to obtain a high-risk range and a low-risk range; An avoidance range is set, and the avoidance range is a fixed range value; position information of a target vehicle in the high-risk range is obtained to obtain a target position item; and an avoidance range set is obtained based on a combination result of the avoidance range and the target position item; Based on a removal result of the avoidance range set from the determination range, a position closest to the determination midpoint is obtained as an adjustment point, and an adjustment path is generated.

6. The highway automatic driving risk identification system based on a multi-modal AI model according to claim 5, characterized in that: The adjustment method also includes speed adjustment, and the adjustment point acquisition method includes: Based on a removal result of the avoidance range set from the determination range, a position closest to the determination midpoint is obtained as a pre-adjustment point, and an adjustment path is generated; Speed information of the ego vehicle and the target vehicle in the high-risk range is obtained, and the speed information is judged; when the speed information of the ego vehicle exceeds the speed information of the target vehicle in the high-risk range, the position of the ego vehicle is adjusted based on the adjustment path, and the pre-adjustment point is used as the adjustment point; when the speed information of the ego vehicle does not exceed the speed information of the target vehicle in the high-risk range, the pre-adjustment point is used as a removal adjustment point, a position closest to the determination midpoint is obtained again as the pre-adjustment point, an adjustment path is generated, and the step is repeated until the position of the ego vehicle is adjusted.

Images