CN116206285A

CN116206285A - Traffic weakness group collision risk assessment method and system applied to automatic driving

Info

Publication number: CN116206285A
Application number: CN202310227793.XA
Authority: CN
Inventors: 廖文龙; 王帝; 何弢; 张润玺
Original assignee: Shanghai Kuyi Robot Co ltd; Kuwa Technology Co ltd
Current assignee: Shanghai Kuyi Robot Co ltd; Kuwa Technology Co ltd
Priority date: 2023-03-06
Filing date: 2023-03-06
Publication date: 2023-06-02

Abstract

The invention provides a traffic weakness group collision risk assessment method and a system applied to automatic driving, wherein the method comprises the following steps: identifying and acquiring information of traffic weakness groups around the vehicle, acquiring position and speed information of the vehicle, and acquiring information of the surrounding traffic environment of the traffic weakness groups; classifying and identifying the driving intention of the traffic weakness group by using the acquired traffic weakness group information; based on the historical multi-frame information of the traffic weakness group, predicting the running track of the traffic weakness group by combining the running intention of the traffic weakness group; generating an alternative track according to the driving target, performing collision detection on the track and a predicted track of a weak group, expanding an influence boundary by the weak group according to the driving intention, and calculating the position of collision and the time of collision; and calculating collision risk of the vulnerable group and the automatic driving vehicle. The invention can effectively avoid unnecessary deceleration of the vehicle caused by unrelated pedestrians or non-motor vehicles.

Description

Traffic weakness group collision risk assessment method and system applied to automatic driving

Technical Field

The invention relates to the technical field of automatic driving, in particular to a traffic weakness group collision risk assessment method and system applied to automatic driving.

Background

With the continuous development of automatic driving technology, an automatic driving vehicle has realized stable and safe driving under specific scenes such as expressways, overhead, etc., and for general urban roads, due to the existence of traffic weakness groups such as pedestrians, bicycles, tricycles, etc., the driving scene of the automatic driving vehicle becomes extremely complex, and therefore, the automatic driving vehicle is difficult to correctly evaluate the collision risk with the traffic weakness groups.

The traffic weakness group has the characteristics of strong randomness and high flexibility in behavior, and meanwhile, due to the fact that the constraint force of traffic regulations on the traffic weakness group is weak, the individuals can have sporadic behavior which violates the traffic regulations, and normal running of the automatic driving vehicle can be interfered. The traditional collision detection method based on the collision point and the safety distance can keep the distance between the self-vehicle and the traffic weakness group, so that collision is avoided, but due to the too conservative characteristic, when the traffic flow of people is large, people can be too cautious to cause difficult smooth traffic.

Patent document CN111746527B discloses a method for predicting collision between a vehicle and a pedestrian, which is characterized in that according to related information, a vehicle movement time budget model predicts the driving time of the vehicle and the pedestrian, and a pedestrian movement time budget model predicts the movement time of the pedestrian, which is likely to collide with the vehicle, wherein the related information comprises acquired related data of the vehicle, the pedestrian and a road, and a collision early warning classification model compares the calculated time difference between the vehicle and the pedestrian by a remote server, and alarms are carried out at a pedestrian mobile phone application end and a vehicle-mounted application end through radio signals, so that the risk of collision between the pedestrian and the vehicle is reduced. The method predicts the collision from the angle of collision time and feeds back the predicted result to the vehicle, but does not consider the behavior intention of pedestrians, and also ignores the characteristic that the behavior of the pedestrians can change at any time, so that the problem is that the pedestrian is too stiff to treat, the pedestrians possibly at risk are not considered, the pedestrians possibly at risk in the future are not careful to treat, if the predicted pedestrian is taken as the prediction of the collision risk of automatic driving, the behavior such as sudden braking and the like is likely to happen frequently due to misidentification of the collision risk of the vehicle, and adverse effects are generated on the driving safety and the traffic efficiency.

Patent document CN114299607a discloses a method for analyzing the risk of collision of a person and a vehicle based on an automatically driven vehicle, comprising: according to the street crossing characteristics in the historical data set of the pedestrian crossing, the street crossing behaviors are divided into different habit types through Gaussian clustering, the automatic driving vehicle is helped to acquire a future possible state set according to the current motion state of the pedestrian by utilizing a joint probability distribution function, the future possible track set of the pedestrian is obtained by considering the dynamic space-time relationship of the human vehicle through a GCN graph convolution neural network, finally, the track collision probability and the minimum meeting distance are considered, the characteristic dimension reduction is carried out by utilizing a matter extension theory, a risk function is established, and the real-time judgment of the collision risk of the human vehicle is realized. However, the invention does not expand the vulnerable group to affect the boundary, and can not enable the self-vehicle to provide more safety buffering for high-risk obstacles.

Patent document CN112258841B discloses an intelligent vehicle risk assessment method based on its vehicle trajectory prediction, comprising a trajectory prediction step and a threat assessment step; the track prediction step adopts a deep learning algorithm, and training and evaluating a behavior prediction model by utilizing an NGSIM data set to realize driving intention classification and obtain track probability distribution of an autonomous vehicle; the threat assessment step quantitatively and objectively assesses the threat in each state by a vehicle risk assessment function including a time to collision TTC, a headway TH, and an enhanced time to collision ETTC. But the invention cannot recognize the driving intention of the vulnerable group.

Disclosure of Invention

Aiming at the defects in the prior art, the invention aims to provide a traffic weakness group collision risk assessment method and system applied to automatic driving.

The invention provides a traffic weakness group collision risk assessment method applied to automatic driving, which comprises the following steps:

step S1: identifying and acquiring type, position, speed and orientation information of a preset group, acquiring vehicle position and speed information, and acquiring traffic environment information of the preset group;

step S2: classifying and identifying the driving intention of the preset group by using the acquired speed, position and orientation information of the preset group;

step S3: based on the historical multi-frame information of the preset group, predicting the running track of the preset group by combining the running intention of the preset group;

step S4: generating an alternative track according to the driving target, performing collision detection on the alternative track and a predicted track of a preset group, expanding an influence boundary by the preset group according to the driving intention, and calculating the position of the collision and the time of the collision;

step S5: and calculating the collision risk of the preset group and the vehicle.

Preferably, in said step S1:

identifying and acquiring type, position, speed and orientation information of a preset group around the vehicle according to the laser radar, the camera and the radar; and acquiring position and speed information of the vehicle according to the map and the positioning information, and acquiring surrounding traffic environment information of a preset group.

Preferably, in said step S2:

classifying and identifying the driving intention of the preset group by using the acquired speed, position and orientation information of the preset group, wherein the identification process is as follows:

step S2.1: dividing the preset group into non-motor vehicles and pedestrians according to the types of the preset group;

step S2.2: identifying the color based on the position of the preset group;

for a non-motor vehicle group, if the non-motor vehicle is in a non-intersection state, checking whether the non-motor vehicle is in a bicycle lane, if so, the non-motor vehicle group is in a normal running state, and the attention level is lower; if it is in the motor vehicle lane, the class of interest is higher; if the non-motor vehicle is in the crossing state, checking whether the non-motor vehicle is at one side of the pedestrian crossing, if the non-motor vehicle is at one side of the pedestrian crossing, waiting for the green light to pass, if the non-motor vehicle is in the pedestrian crossing, passing, lifting the attention level, and if the non-motor vehicle is not in the pedestrian crossing, in the motor vehicle running area, lifting the attention level to the highest;

for a pedestrian group, if the pedestrian group is in a non-intersection state, checking whether the pedestrian group walks on the sidewalk, if the pedestrian walks on the sidewalk, the attention level is lower, if the pedestrian is in a non-motor vehicle lane, the attention level is improved, and if the pedestrian is in a motor vehicle lane, the attention level is improved to the highest; if the pedestrian is in the crossing state, checking whether the pedestrian is in the crossing, if the pedestrian is out of the crossing, reducing the attention level, if the pedestrian is on two sides of the crosswalk, improving the attention level, and if the pedestrian is in the crosswalk, improving the attention level to the highest;

Step S2.3: identifying the behavior intention of the group according to the current position, speed and direction of the preset group and the position speed and direction of the past preset time; for a non-motor vehicle, if the non-motor vehicle runs on a non-motor vehicle lane, checking whether the current speed and the direction of the non-motor vehicle are consistent with the direction of the lane where the non-motor vehicle is located, and if the current speed and the direction of the non-motor vehicle are consistent with the directions of the lanes where the non-motor vehicle is located, the non-motor vehicle keeps straight running with high probability; checking whether the non-motor vehicle is blocked from running if a stationary vehicle or other obstacle exists on the non-motor vehicle lane, and if the non-motor vehicle is blocked from running, the non-motor vehicle has a preset probability of invading the motor vehicle lane; if the non-motor vehicle is in the reverse running of the lane, checking whether the adjacent lane is physically blocked by the non-motor vehicle, and if not, crossing the lane boundary to get into the motor vehicle lane with preset probability; if the non-motor vehicle is in the intersection, judging the future running direction of the non-motor vehicle according to the position of the sidewalk where the non-motor vehicle is located and the speed of the non-motor vehicle, if the non-motor vehicle is in the running range of the motor vehicle, checking the direction of the traffic light green light at the moment according to traffic light information, and judging the future running direction of the non-motor vehicle;

For pedestrians, if the pedestrians are in a non-motor vehicle lane or a motor vehicle lane, the traveling trend of the pedestrians is judged according to the direction and the speed of the pedestrians, and if the pedestrians are in the sidewalk, the pedestrians have lower influence probability on the running of the self-vehicle; if a pedestrian is at an intersection, the possibility of the pedestrian to walk is estimated by combining traffic light information, and if the pedestrian has already performed walking, the future destination is judged according to the speed and the direction of the pedestrian.

Preferably, in said step S3:

based on the historical multi-frame information of the preset group, the travel track of the preset group is predicted by using a deep learning method in combination with the travel intention of the preset group. The method comprises the following steps:

step S3.1: collecting preset group data comprising speed and orientation information;

step S3.2: classifying according to the types and intentions of the preset groups, establishing different neural network models, and respectively training preset group track prediction models in different scenes;

step S3.3: the track prediction model is deployed to an automatic driving vehicle, tracks of a preset group are predicted in real time according to the information of the preset group, and a plurality of possible paths and probabilities of all paths are given.

Preferably, in said step S4:

the automatic driving vehicle decision-making planning module generates an alternative track according to a driving target, carries out collision detection on the track and a predicted track of a preset group, expands an influence boundary according to the driving intention of the preset group, and calculates the position of the collision and the time of the collision, wherein the specific method comprises the following steps:

step S4.1: according to the position of the preset group and the driving intention, the influence of the preset group on the automatic driving vehicle is evaluated, and if the preset group is in a traversing and reversing state, the influence on the automatic driving vehicle is increased, and the influence boundary of the preset group is correspondingly expanded;

step S4.2: the influence boundary of the preset group exists in a polygonal form, and when the intention is evaluated as important attention, the influence boundary is extended to the periphery and extended to an action target point of the preset group;

step S4.3: when collision detection is carried out, forming an influence range of a preset group at each time point, forming a polygonal envelope of a vehicle at corresponding time, detecting whether the two are overlapped, if so, meaning that the automatic driving vehicle collides with the preset group at the time point in the future, if not, calculating the distance between the two, and calculating a risk coefficient according to the nearest distance.

Preferably, in said step S5:

the collision risk of the preset group and the automatic driving vehicle is calculated, and the specific steps are as follows:

step S5.1: determining a collision basic risk according to the preset group traveling intention, wherein if the attention level is lower, the probability of collision between the preset group and the own vehicle is lower, and if the attention level is higher, the probability of collision between the preset group and the own vehicle is higher, and the basic risk is also higher;

step S5.2: determining an addition risk according to a collision detection result, wherein the addition risk is related to whether a vehicle collides with a preset group or not, if the vehicle does not collide, the addition risk is related to the minimum distance of the track, and if the collision or the minimum distance occurrence time is smaller than a preset value, the risk is higher, and measures are required to be taken immediately to avoid collision; if the collision or minimum distance occurrence time is between preset time, the risk is medium, and the self-vehicle running speed is limited according to the comprehensive risk; if the collision or minimum distance occurrence time is larger than a preset value, further considering uncertainty of the driving intention of the preset group, if the preset group is in a driving state with stability meeting a preset standard, taking speed limiting action by the self-vehicle, and if the preset group is in a reverse or crossing special state, further observing to determine subsequent action.

According to the invention, the traffic weakness group collision risk assessment system applied to automatic driving comprises:

module M1: identifying and acquiring type, position, speed and orientation information of a preset group, acquiring vehicle position and speed information, and acquiring traffic environment information of the preset group;

module M2: classifying and identifying the driving intention of the preset group by using the acquired speed, position and orientation information of the preset group;

module M3: based on the historical multi-frame information of the preset group, predicting the running track of the preset group by combining the running intention of the preset group;

module M4: generating an alternative track according to the driving target, performing collision detection on the alternative track and a predicted track of a preset group, expanding an influence boundary by the preset group according to the driving intention, and calculating the position of the collision and the time of the collision;

module M5: and calculating the collision risk of the preset group and the vehicle.

Preferably, in said module M1:

identifying and acquiring type, position, speed and orientation information of a preset group around the vehicle according to the laser radar, the camera and the radar; acquiring position and speed information of a vehicle according to the map and the positioning information, and acquiring surrounding traffic environment information of a preset group;

In the module M2:

module M2.1: dividing the preset group into non-motor vehicles and pedestrians according to the types of the preset group;

module M2.2: identifying the color based on the position of the preset group;

Module M2.3: identifying the behavior intention of the group according to the current position, speed and direction of the preset group and the position speed and direction of the past preset time; for a non-motor vehicle, if the non-motor vehicle runs on a non-motor vehicle lane, checking whether the current speed and the direction of the non-motor vehicle are consistent with the direction of the lane where the non-motor vehicle is located, and if the current speed and the direction of the non-motor vehicle are consistent with the directions of the lanes where the non-motor vehicle is located, the non-motor vehicle keeps straight running with high probability; checking whether the non-motor vehicle is blocked from running if a stationary vehicle or other obstacle exists on the non-motor vehicle lane, and if the non-motor vehicle is blocked from running, the non-motor vehicle has a preset probability of invading the motor vehicle lane; if the non-motor vehicle is in the reverse running of the lane, checking whether the adjacent lane is physically blocked by the non-motor vehicle, and if not, crossing the lane boundary to get into the motor vehicle lane with preset probability; if the non-motor vehicle is in the intersection, judging the future running direction of the non-motor vehicle according to the position of the sidewalk where the non-motor vehicle is located and the speed of the non-motor vehicle, if the non-motor vehicle is in the running range of the motor vehicle, checking the direction of the traffic light green light at the moment according to traffic light information, and judging the future running direction of the non-motor vehicle;

Preferably, in said module M3:

module M3.1: collecting preset group data comprising speed and orientation information;

module M3.2: classifying according to the types and intentions of the preset groups, establishing different neural network models, and respectively training preset group track prediction models in different scenes;

module M3.3: deploying a track prediction model to an automatic driving vehicle, predicting the track of a preset group in real time according to the perceived preset group information, and giving out a plurality of possible paths and the probability of each path;

In the module M4:

module M4.1: according to the position of the preset group and the driving intention, the influence of the preset group on the automatic driving vehicle is evaluated, and if the preset group is in a traversing and reversing state, the influence on the automatic driving vehicle is increased, and the influence boundary of the preset group is correspondingly expanded;

module M4.2: the influence boundary of the preset group exists in a polygonal form, and when the intention is evaluated as important attention, the influence boundary is extended to the periphery and extended to an action target point of the preset group;

module M4.3: when collision detection is carried out, forming an influence range of a preset group at each time point, forming a polygonal envelope of a vehicle at corresponding time, detecting whether the two are overlapped, if so, meaning that the automatic driving vehicle collides with the preset group at the time point in the future, if not, calculating the distance between the two, and calculating a risk coefficient according to the nearest distance.

Preferably, in said module M5:

module M5.1: determining a collision basic risk according to the preset group traveling intention, wherein if the attention level is lower, the probability of collision between the preset group and the own vehicle is lower, and if the attention level is higher, the probability of collision between the preset group and the own vehicle is higher, and the basic risk is also higher;

module M5.2: determining an addition risk according to a collision detection result, wherein the addition risk is related to whether a vehicle collides with a preset group or not, if the vehicle does not collide, the addition risk is related to the minimum distance of the track, and if the collision or the minimum distance occurrence time is smaller than a preset value, the risk is higher, and measures are required to be taken immediately to avoid collision; if the collision or minimum distance occurrence time is between preset time, the risk is medium, and the self-vehicle running speed is limited according to the comprehensive risk; if the collision or minimum distance occurrence time is larger than a preset value, further considering uncertainty of the driving intention of the preset group, if the preset group is in a driving state with stability meeting a preset standard, taking speed limiting action by the self-vehicle, and if the preset group is in a reverse or crossing special state, further observing to determine subsequent action.

Compared with the prior art, the invention has the following beneficial effects:

1. the invention can identify the driving intention of the vulnerable group, determine the basic risk of collision according to the driving intention, and effectively avoid unnecessary deceleration of the vehicle caused by unrelated pedestrians or non-motor vehicles;

2. the invention expands the influence boundary of the vulnerable group, so that the own vehicle can provide more safety buffer for high-risk obstacles, and the coverage of collision detection is improved, thereby ensuring the driving safety;

3. the invention adopts the running track to carry out frame-by-frame collision detection, can accurately position the time and place of collision, and enables an automatic driving system to take effective actions in time.

Drawings

Other features, objects and advantages of the present invention will become more apparent upon reading of the detailed description of non-limiting embodiments, given with reference to the accompanying drawings in which:

FIG. 1 is a schematic flow chart of the present invention;

FIG. 2 is a schematic diagram of a traffic weakness group collision risk assessment method applied to automatic driving;

FIG. 3 is a schematic diagram of a process for classifying and identifying travel intents of a traffic offensive group;

FIG. 4 is a schematic diagram of a process for predicting travel trajectories of a traffic offending group using a deep learning method;

FIG. 5 is a schematic illustration of a time course of calculating the location of a collision and the collision;

FIG. 6 is a schematic diagram of a process for calculating risk of collision between a vulnerable group and an autonomous vehicle.

Detailed Description

The present invention will be described in detail with reference to specific examples. The following examples will assist those skilled in the art in further understanding the present invention, but are not intended to limit the invention in any way. It should be noted that variations and modifications could be made by those skilled in the art without departing from the inventive concept. These are all within the scope of the present invention.

Example 1:

on the basis of the existing method, the intention recognition and track prediction are carried out on various traffic weakness groups, the collision time and the collision position are accurately calculated according to the vehicle track information, the collision risk of the automatic driving vehicle and the traffic weakness groups is estimated, so that the vehicle behavior is guided, the collision risk of the vehicle and the traffic weakness groups is greatly reduced, and meanwhile, the driving efficiency of the vehicle is improved as much as possible.

The invention relates to the technical field of automatic driving, in particular to a method for evaluating collision risks of traffic handicapped participants to adjust the self-vehicle behavior and ensure the driving safety.

1. According to the laser radar, the camera and the radar identify and acquire the type, position, speed and orientation information of traffic weakness groups around the vehicle. And acquiring position and speed information of the vehicle according to the high-definition map and the positioning information, and acquiring surrounding traffic environment information of the traffic weakness group.

2. Classifying and identifying the driving intention of the traffic weakness group by using the acquired speed, position and orientation information of the traffic weakness group

3. Based on historical multi-frame information of the traffic weakness group, the travel track of the traffic weakness group is predicted by a deep learning method in combination with the travel intention of the traffic weakness group

4. The automatic driving vehicle decision planning module generates an alternative track according to a driving target, carries out collision detection on the track and a predicted track of a weak group, expands an influence boundary according to the driving intention of the weak group, and then accurately calculates the position of the collision and the time of the collision.

5. And calculating collision risk of the vulnerable group and the automatic driving vehicle. And determining the basic risk of collision according to the driving intention of the vulnerable group. And determining the addition risk according to the collision detection result.

According to the invention, as shown in fig. 1-6, the traffic weakness group collision risk assessment method applied to automatic driving comprises the following steps:

specifically, in the step S1:

specifically, in the step S2:

step S2.2: identifying the color based on the position of the preset group;

specifically, in the step S3:

specifically, in the step S4:

Specifically, in the step S5:

Example 2:

example 2 is a preferable example of example 1 to more specifically explain the present invention.

The invention also provides a traffic weakness group collision risk assessment system applied to automatic driving, which can be realized by executing the flow steps of the traffic weakness group collision risk assessment method applied to automatic driving, namely, a person skilled in the art can understand the traffic weakness group collision risk assessment method applied to automatic driving as a preferred implementation mode of the traffic weakness group collision risk assessment system applied to automatic driving.

specifically, in the module M1:

in the module M2:

module M2.2: identifying the color based on the position of the preset group;

specifically, in the module M3:

in the module M4:

Specifically, in the module M5:

Example 3:

example 3 is a preferable example of example 1 to more specifically explain the present invention.

The invention carries out intention recognition and track prediction on various traffic weakness groups, expands the influence boundary of the traffic weakness groups according to intention information, accurately calculates collision time and collision position according to vehicle track information, evaluates the collision risk of the automatic driving vehicle and the traffic weakness groups, and guides the vehicle behavior, and comprises the following specific steps:

2. Classifying and identifying the driving intention of the traffic weakness group by using the acquired speed, position and orientation information of the traffic weakness group, wherein the identification process is as follows:

(1) According to the types of the traffic vulnerable groups, the traffic vulnerable groups are divided into two major categories of non-motor vehicles and pedestrians

(2) Based on the position of the traffic weakness group, the participation roles of the traffic weakness group in the whole traffic environment are identified.

For a non-motor vehicle group, if the non-motor vehicle group is in a non-intersection state, whether the non-motor vehicle group is in a bicycle lane or not is checked, if the non-motor vehicle group is in the bicycle lane, the non-motor vehicle group is considered to be in a normal running state, the concerned grade is low, if the non-motor vehicle group is in the bicycle lane, the probability of unsafe behavior such as crossing and the like is greatly increased, and the concerned grade is high. If the non-motor vehicle is in the crossing state, checking whether the non-motor vehicle is on one side of the crosswalk, if the non-motor vehicle is on one side of the crosswalk, the non-motor vehicle is considered to be in a state waiting for green light traffic, if the non-motor vehicle is in the crosswalk, the non-motor vehicle is considered to be in a state of traveling, the attention level needs to be raised, and if the non-motor vehicle is not in the crosswalk but is in a motor vehicle traveling area, the attention is required.

For a pedestrian group, if the pedestrian is in a non-intersection state, whether the pedestrian walks on a sidewalk or not is checked, if the pedestrian walks on the sidewalk, the influence on the vehicle running is low, the attention level is low, if the pedestrian is in a non-motor vehicle lane, which means that the pedestrian possibly generates an intention of walking, the attention level needs to be raised, and if the pedestrian is in a motor vehicle lane, which means that the pedestrian often already performs walking, and important attention needs to be paid. If the pedestrian is in the crossing state, whether the pedestrian is in the crossing is checked, if the pedestrian is out of the crossing, the pedestrian is not participated in the traffic scene of the crossing immediately in a short time, the attention level of the pedestrian can be properly reduced, if the pedestrian is on both sides of the crosswalk, the pedestrian is likely to walk, and when the pedestrian waits for proper crossing time, the attention level is required to be improved, and if the pedestrian is in the crosswalk, the pedestrian is in a crossing state, and important attention is required.

(3) The behavioral intent of the population is identified based on the current location, speed, and orientation of the traffic-impaired population and the location speed and orientation of the past period of time. For a non-motor vehicle, such as it is traveling on a non-motor vehicle lane, it is checked whether its current speed and heading are consistent with the heading of the lane in which it is located, and if so, the probability that the non-motor vehicle remains straight is high. If there is a stationary vehicle or other obstacle on the non-motor vehicle lane, it is checked whether it blocks the travel of the non-motor vehicle, and if it blocks the travel of the non-motor vehicle, the non-motor vehicle is likely to have an intrusion behavior into the motor vehicle lane. If the non-motor vehicle is traveling in the reverse direction, it is checked whether the adjacent lane has a significant physical barrier to the non-motor vehicle lane, and if not, the non-motor vehicle is likely to cross the inter-lane boundary to come into the motor vehicle lane. If the non-motor vehicle is in the intersection, the future running direction of the non-motor vehicle can be judged according to the position of the sidewalk where the non-motor vehicle is located and the speed of the sidewalk, and if the non-motor vehicle is in the running range of the motor vehicle, the direction of the traffic light green light at the moment is checked by combining traffic light information, and the future running direction of the non-motor vehicle is judged.

For pedestrians, if the pedestrians are in a non-motor vehicle lane or a motor vehicle lane, the traveling trend of the pedestrians needs to be judged according to the direction and the speed of the pedestrians, and if the pedestrians are in the sidewalk, the pedestrians have low influence probability on the running of the self-vehicle. If the pedestrian is at the intersection, the possibility of the pedestrian to walk is estimated by combining information such as traffic lights, and if the pedestrian has already walked, the future destination is judged according to the speed and the direction of the pedestrian.

3. Based on the historical multi-frame information of the traffic weakness groups, the travel track of the traffic weakness groups is predicted by using a deep learning method in combination with the travel intention of the traffic weakness groups. The method comprises the following steps:

(1) A large amount of traffic weakness group data is collected, including information on its speed, orientation, etc.

(2) Classifying according to the categories and intentions of the traffic weakness groups, establishing different neural network models, and respectively training traffic weakness group track prediction models under different scenes.

(3) And deploying the track prediction model to an automatic driving vehicle, predicting the track of the traffic weakness group in real time according to the perceived traffic weakness group information, and giving out a plurality of possible paths and the probability of each path.

4. The automatic driving vehicle decision planning module generates an alternative track according to a driving target, carries out collision detection on the track and a predicted track of a weak group, expands an influence boundary according to the driving intention of the weak group, and then accurately calculates the position of the collision and the time of the collision. The specific method comprises the following steps:

(1) According to the position and the driving intention of the vulnerable group, the influence of the vulnerable group on the automatic driving vehicle is evaluated, and particularly if the vulnerable group is in a state of crossing, reversing and the like, the influence on the automatic driving vehicle is increased, and the influence boundary of the vulnerable group is correspondingly expanded.

(2) The influence boundary of the vulnerable group exists in a polygonal form, and when the intention is evaluated as important attention is required, the influence boundary is extended to the periphery and the action target point of the vulnerable group is extended.

(3) When collision detection is carried out, the influence range of the weak group is formed at each time point, the polygonal envelope of the own vehicle is formed at the corresponding time, and whether the two are overlapped is detected. If overlap occurs, this means that the autonomous vehicle collides with the vulnerable group at this point in time in the future. If no overlap occurs, the distance between the two is calculated and the risk factor is calculated from the nearest distance.

5. The method for calculating the collision risk of the vulnerable group and the automatic driving vehicle comprises the following specific steps:

(1) And determining the basic risk of collision according to the driving intention of the vulnerable group, wherein if the attention level is considered to be low, the probability of collision between the vulnerable group and the self-vehicle is low, and if the attention level is considered to be high, the probability of collision between the vulnerable group and the self-vehicle is high, so that the basic risk is high.

(2) And determining the addition risk according to the collision detection result. The risk of addition is related to whether the vehicle collides with the vulnerable group, and if not, to the minimum distance of the trajectory. If the collision or minimum distance occurrence time is <2s, then the risk is considered to be high and immediate measures need to be taken to avoid collision. If collision or minimum distance occurs between 2s and 5s, the risk is considered to be medium, and the self-vehicle running speed needs to be limited according to the comprehensive risk, so that the overall situation is controllable. If the collision or minimum distance occurrence time is more than 5s, the uncertainty of the driving intention of the vulnerable group needs to be further considered, if the vulnerable group is in a stable driving state, the self-vehicle needs to take proper speed limiting action, and if the vulnerable group is in a special state such as retrograde or traversing, further observation can be performed to determine the subsequent action.

Those skilled in the art will appreciate that the systems, apparatus, and their respective modules provided herein may be implemented entirely by logic programming of method steps such that the systems, apparatus, and their respective modules are implemented as logic gates, switches, application specific integrated circuits, programmable logic controllers, embedded microcontrollers, etc., in addition to the systems, apparatus, and their respective modules being implemented as pure computer readable program code. Therefore, the system, the apparatus, and the respective modules thereof provided by the present invention may be regarded as one hardware component, and the modules included therein for implementing various programs may also be regarded as structures within the hardware component; modules for implementing various functions may also be regarded as being either software programs for implementing the methods or structures within hardware components.

The foregoing describes specific embodiments of the present invention. It is to be understood that the invention is not limited to the particular embodiments described above, and that various changes or modifications may be made by those skilled in the art within the scope of the appended claims without affecting the spirit of the invention. The embodiments of the present application and features in the embodiments may be combined with each other arbitrarily without conflict.

Claims

1. A traffic weakness group collision risk assessment method applied to automatic driving, comprising:

2. The traffic weakness group collision risk assessment method applied to automatic driving according to claim 1, wherein in the step S1:

3. The traffic weakness group collision risk assessment method applied to automatic driving according to claim 1, wherein in the step S2:

step S2.2: identifying the color based on the position of the preset group;

4. The traffic-weakening group collision risk assessment method applied to automatic driving according to claim 1, wherein in said step S3:

5. The traffic weakness group collision risk assessment method applied to automatic driving according to claim 1, wherein in the step S4:

6. The traffic weakness group collision risk assessment method applied to automatic driving according to claim 1, wherein in the step S5:

7. A traffic weakness group collision risk assessment system for automatic driving, comprising:

8. The traffic weakness group collision risk assessment system for automatic driving according to claim 7, wherein:

in the module M1:

In the module M2:

module M2.2: identifying the color based on the position of the preset group;

9. The traffic weakness group collision risk assessment system for automatic driving according to claim 7, wherein:

in the module M3:

In the module M4:

10. The traffic-weakening group collision risk assessment system applied to automatic driving according to claim 7, wherein in said module M5: