Safe driving system for chemical goods truck

Abstract

The invention provides a safe driving method for a chemical goods freight vehicle, which comprises the following steps: judging whether a dynamic motor vehicle, a non-motor vehicle and a person enter a dangerous area or not when the vehicle reaches a turning area to turn by predicting the direction of a moving object, and marking the predicted dangerous vehicle and person. Therefore, the influence of a driving blind area on safe driving of a driver is reduced.

Description

TECHNICAL FIELD

[0001] The present application relates to the safe driving of a freight vehicle, in particular, to a safe driving system for a chemical freight vehicle. BACKGROUND

[0002] Due to the height and body of the chemical freight vehicle, there are multiple visual blind spots for the driver during the turning process near the vehicle, which can easily lead to the involvement of nearby pedestrians and vehicles, causing serious safety hazards.

[0003] In the existing technology, researchers have been working to solve this problem.

[0004] For example: In the patent 2020102463751 of Dongfeng Company, a scheme based on rearview mirror information for collision prediction is proposed, but the prediction basis is to identify the relative position relationship between people and vehicles in real time, and then judge whether a collision will occur. In this scheme, the driver does not need to frequently observe the rearview mirror, but since it can only avoid vehicles within the rearview mirror, vehicles that suddenly slow down in the dead angle or to the side or front of the vehicle cannot be monitored, even if they can be monitored, it is already too late.

[0005] For example: In the patent 202110361067.8 of Tongji University, a scheme based on radar to predict collisions is proposed, which only differs from Dongfeng Company in the difference in monitoring equipment, i.e., it uses a radar monitoring scheme to monitor nearby vehicles.

[0006] Therefore, from the existing schemes, theoretical risk areas are mostly used, and rearview mirror or nearby monitoring methods are selected, which can avoid blind spots to some extent, but the effect is not comprehensive. On the one hand, the theoretical risk area is only a theoretical simulation result that cannot cover the real risk area. In the actual driving process, based on the differences in vehicle models, road conditions, and driving speeds, the theoretical value will be biased. On the other hand, rearview mirror monitoring means only monitoring vehicles behind, which ignores the problem of parallel driving vehicles or vehicles in front of the vehicle suddenly slowing down. When it enters the rearview mirror due to deceleration, it means that the risk vehicle may have approached or entered the danger zone. Similarly, the problem also exists in radar monitoring, which is a typical nearby monitoring scheme. Although some monitoring methods achieve the so-called God's view of full-body display, the vehicles that can be captured by the God's view are also nearby vehicles, which increases the risk of too close distance. Moreover, relying on the display screen to drive is extremely unsafe and does not comply with driving standards. SUMMARY

[0007] The present application aims to overcome the above-mentioned defects, and provides a safe driving method for chemical goods transport vehicles based on a more comprehensive vehicle prediction strategy.

[0008] The present application provides a safe driving method for chemical goods transport vehicles, comprising the following steps: 1. A safe driving method for chemical goods transport vehicles, characterized in that it comprises the following steps: S1. Based on vehicle navigation, obtain the corner information of the front turning intersection; S2. Obtain the vehicle volume information, and obtain the risk area at the corner under different vehicle speeds; S3. When the vehicle enters the turning area after T seconds based on GPS prediction, start the image device on the side to be turned; The image device has a shooting range of N1 range in front of the vehicle to N2 range behind the vehicle; S4. Analyze the road condition information captured by the image device, and identify all dynamic motor vehicles and persons within the range of the vehicle to the sidewalk; S5. Dynamically track the dynamic motor vehicles and persons, and predict the direction of the dynamic motor vehicles, non-motor vehicles and persons based on the indicator lights of the dynamic motor vehicles and the body movements of the persons; S6. Based on the predicted direction of S5, determine whether the dynamic motor vehicles, non-motor vehicles and persons will enter the dangerous area when the vehicle reaches the turning area for turning, and mark the predicted dangerous vehicles and persons.

[0009] Further, the present application provides a safe driving method for chemical goods transport vehicles, characterized in that: In step S1, the corner information includes: the real-time distance of the vehicle from the corner, the shape of the corner, and the surrounding information of the corner.

[0010] Further, the present application provides a safe driving method for chemical goods transport vehicles, characterized in that: In step S2, the risk area at the corner under different vehicle speeds is obtained through a vehicle turning model; The construction method of the vehicle turning model is as follows: Sc1. Collect videos of different vehicles passing through corners at different speeds and different steering wheel rotation angles at different corners, forming a video set S; Sc2. Obtain the image of each frame of each video to form an image set Ti, label the risk area in the image to form a risk area subset Ti, and superimpose the risk areas of all frames in the image set Ti to form the simulated risk area of the current video, and form a risk area set Mi; i is a label represented by {c, n, s, t}, c is a vehicle type, n is a corner type, s is a speed, and t is a steering wheel rotation angle; Sc4. Complete the missing data by machine learning to form a vehicle turning model under different vehicle types, at different corners, at different speeds, and at different steering wheel rotation angles.

[0011] Further, the safety driving method for the chemical goods transport vehicle provided by the present application is characterized in that: In step Sc2, different drivers are also labeled, i.e., the i label is represented by {c, n, s, t, p}, and p is a person.

[0012] Further, the safety driving method for the chemical goods transport vehicle provided by the present application is characterized in that: In step S5, the direction of the motor vehicle is identified by the turn signal.

[0013] Further, the safety driving method for the chemical goods transport vehicle provided by the present application is characterized in that: In step S5, the direction of the non-motor vehicle and the person is predicted by the prediction model I; The construction method of the prediction model is as follows: Se1. Collect image data of non-motor vehicles and persons in the movement process to form a data set X; Se2. Classify the data set X and label it as {d, m}, d is a target object, and m is a behavior mode of d; Se3. Complete the missing data by machine learning to form a prediction model under different target objects and different behavior modes.

[0014] Further, the safety driving method for the chemical goods transport vehicle provided by the present application is characterized in that: In step Se2, data cleaning is also performed; The data cleaning includes the following steps: Se2.1. Each image data in the data set X is divided into a plurality of segments or frames in different ways, and labeled as {d, m, q}, q is a sequence number; Se2.2. For segments derived from the same image data, repetitive screening is performed to clean the duplicate data; Se2.3. Comparing the data of different image data after cleaning in Se2.2, finding the repeated data between different image data, marking it as suspicious data, and marking the distinguished data as identification data.

[0015] Further, the application provides a safe driving method for a chemical goods transport vehicle, and the method further has the characteristics that: In step S6, the method for judging whether the dynamic motor vehicle, non-motor vehicle and person will enter the dangerous area when the host vehicle reaches the turning area for turning is: S6.1. distinguishing the dynamic motor vehicle, non-motor vehicle and person, When the target object is a motor vehicle, S6.2A is performed; When the target object is a non-motor vehicle, S6.2B is performed; When the target object is a person, S6.2C is performed; S6.2A. identifying the speed of the motor vehicle, simulating the driving route based on the direction of the motor vehicle obtained in S5, and judging whether the motor vehicle will meet the host vehicle in the turning area and enter the turning dangerous area of the host vehicle; S6.2B. identifying the speed of the non-motor vehicle, simulating the driving route based on the direction of the non-motor vehicle obtained in S5, and judging whether the non-motor vehicle will meet the host vehicle in the turning area and enter the turning dangerous area of the host vehicle; S6.2C. identifying the speed of the person, simulating the moving route of the person based on the direction of the person obtained in S5, and judging whether the person will meet the host vehicle in the turning area and enter the turning dangerous area of the host vehicle. BRIEF DESCRIPTION OF DRAWINGS

[0016] Figure 1 The embodiment provides a flowchart of a safe driving method for a chemical goods transport vehicle. DETAILED DESCRIPTION

[0017] The application can be implemented in various ways and can have various embodiments, so specific embodiments are illustrated in the drawings and described. However, this is not intended to limit the application to specific embodiments, but should be understood to include all modifications, equivalents and even alternatives within the scope of the idea and technology of the application.

[0018] As shown in Figure 1 A safe driving method for a chemical goods transport vehicle includes the following steps: S1. Based on vehicle navigation, obtain the corner information of the front turning intersection; the corner information includes: the real-time distance of the vehicle and the corner, the shape of the corner (such as: the angle size of the corner, the number of roads on both sides of the corner, and the driving corner size when the vehicle turns into different channels, and other information related to the corner), and the surrounding information of the corner (such as: whether there is a separation belt, whether there is a pedestrian zebra crossing).

[0019] In the traditional scheme, the difference of risk area caused by the shape of the corner is not considered, and the theoretical value between the side roll shafts of the vehicle during real-time turning is often used for risk area calculation. This method relies on real-time calculation during field operation, on the one hand, if cloud processing is used, it is easy to cause information delay due to poor network environment, on the other hand, if a vehicle-mounted processing system is used, the calculation starts only after the vehicle moves, which has caused a delay and cannot truly predict. Moreover, the correctness of the theoretical value can only reach 70-80%, and it cannot cover the differences caused by the driving habits of the driver.

[0020] S2. Obtain the vehicle volume information, and obtain the risk area of the corner at different vehicle speeds; In step S2, the vehicle turning model is used to obtain the risk area of the corner at different vehicle speeds; The construction method of the vehicle turning model is as follows: Sc1. Collect videos of different vehicles passing through different corners at different speeds and different steering wheel turning angles to form a video set S; Sc2. Obtain the image of each frame of each video to form an image set Ti, label the risk area in the image to form a risk area subset Ti, and superimpose the risk area of all frames in the image set Ti to form the simulated risk area of the current video to form a risk area set Mi; i is a label represented by {c, n, s, t, p}, c is the vehicle type (i.e., different vehicle models have their own independent code, so as to mark the data as data of a specific vehicle, which is used for calling in the prediction stage), n is the corner type (i.e., the code for independent marking of different corner radian sizes), s is the speed (i.e., the code for different driving speeds, because it is found in this test example that even if the corner angle is consistent, different drivers will use different speeds to turn according to their habits or on-site environment, and such differences will cause the risk area to change), t is the steering wheel rotation angle (also, it is found in the process of this research that due to the difference in personal habits, when facing different corners, the driver's turning method is not the same, and for different turning methods, the risk area generated is not completely consistent with the theoretical value, and there will be some changes), and p is the person (the intervention of this factor is to better realize the simulation of individualized risk area, we find that different drivers have their unique driving style, so due to the difference in style, it will affect the risk area of the vehicle, therefore, in this research, the human factor is also independently labeled, of course, this part of the data is a progressive perfect data, which will be recorded in the process of driving, and then based on the way of machine learning, it will be constantly added and improved). Sc4. Complete the missing data by machine learning to form a vehicle turning model under different vehicle conditions, different corner angles, different speeds and different steering wheel rotation angles. This machine learning mode can use various mainstream algorithms such as KNN and decision tree, etc., the purpose of which is to cover the data parameters that the initial data cannot cover, and with the running of the system, through the backup of the driving data of the driver, the data will gradually be enriched, and then the real data will be used to replace the completed data.

[0021] S3. When the vehicle enters the turning area based on the GPS prediction after T seconds, start the image device on the side to be turned; The above image device, the shooting range is the image in the vehicle front N1 range to the vehicle rear N2 range; N1 and N2 are set according to the actual needs of the vehicle type, the purpose is to realize the monitoring of the motor vehicle, non-motor vehicle or person in front of the vehicle, behind the vehicle and parallel driving, which is not the current full vehicle surround view image system, the traditional surround view image system is aimed at the surrounding monitoring of the vehicle, but due to its special image processing process, the image will be distorted, and it cannot identify the image at a farther distance, so it cannot monitor the image on the other side of the isolation belt, the front dynamic situation at a farther position, and the rear dynamic situation. The purpose of the embodiment is to realize the prediction and monitoring of all dynamic objects that may enter the danger area, that is, the objects in front of the vehicle may meet at the turning place because of deceleration or speed lower than the vehicle, the objects behind the vehicle may meet at the turning place because of sudden acceleration or speed higher than the vehicle, and the objects in parallel motion may meet at the turning place because they always move at the same speed as the vehicle. Therefore, the significance of this prediction is to pre-evaluate objects with different driving possibilities without missing any moving body.

[0022] S4. Analyzing the road condition information shot by the image device, identifying all dynamic motor vehicles and persons within the range of the vehicle to the sidewalk; S5. Dynamically tracking the dynamic motor vehicles and persons, predicting the direction of the dynamic motor vehicles, non-motor vehicles and persons based on the indicator lights of the dynamic motor vehicles and the body movements of the persons; In step S5, when the identified object is a motor vehicle, the direction of the motor vehicle is identified by recognizing the turn signal, and the possibility of meeting is judged according to the vehicle speed.

[0023] In step S5, when the identified object is a non-motor vehicle and a pedestrian, the direction of the non-motor vehicle and the person is predicted by the prediction model I; The purpose of the prediction model I is to extract the characteristic behavior or suggestive behavior of the person (including the person driving the non-motor vehicle) in the process of preparing to cross the road, preparing to turn, preparing to stop, preparing to go straight, preparing to accelerate, preparing to decelerate, suddenly rushing out, and accidentally falling down, etc. when encountering a corner, which is a tool for predicting the behavior of non-motor vehicles or pedestrians. Its construction method is as follows: Se1. Collect image data of various non-motor vehicles and persons in the process of movement to form a data set X; Se2. Classify the data set X and label it as {d, m}, d is the target object (such as people, bicycles, mobility scooters, etc.), and m is the behavior mode of d (preparing to cross the road, preparing to turn, preparing to stop, preparing to go straight, preparing to accelerate, preparing to decelerate, suddenly rushing out, accidentally falling down, etc.); In order to avoid the problem of too large redundant data and too slow calculation speed, In step Se2, data cleaning is also performed, and the specific steps are as follows: Se2.1. Each image data in the data set X is divided into multiple segments or frames in different ways and labeled as {d, m, q}, q being a sequence number; Se2.2. For segments derived from the same image data, repetitive screening is performed to clean and remove duplicate data; here, the duplicate data can be adjusted according to actual needs to be considered as duplicate when 80-90% similar.

[0024] Se2.3. The data of different image data cleaned by Se2.2 is compared to find the duplicate data between different image data, which is marked as suspicious data, and the distinguished data is marked as identification data. That is, if the video of the prediction object contains similar data of the identification data (the similarity can be set according to actual needs), it is considered as the behavior mode marked as m. When there is only suspicious data, it is marked as m behavior suspicious, and the behavior marking is realized by strengthening the subsequent tracking.

[0025] Se3. The missing data is completed by machine learning to form a prediction model of different target objects and different behavior modes. As mentioned above, due to insufficient data at the initial stage of the system, only traditional missing data filling models such as decision tree can be used to fill the missing data, but when the system memory data becomes more and more abundant, the model can be gradually corrected.

[0026] S6. Based on the prediction trend of S5, it is judged whether the dynamic motor vehicle, non-motor vehicle and person will enter the dangerous area when the vehicle reaches the turning area to turn, and the predicted dangerous vehicles and persons are marked.

[0027] The specific process is as follows: S6.1. Distinguish dynamic motor vehicles, non-motor vehicles and persons, When the target object is a motor vehicle, S6.2A is performed; When the target object is a non-motor vehicle, S6.2B is performed; When the target object is a person, S6.2C is performed; S6.2A. Identify the speed of the motor vehicle, simulate the driving route based on the trend of the motor vehicle obtained by S5, and judge whether it will meet the vehicle at the turning area and enter the turning danger area of the vehicle; S6.2B. Identify the speed of the non-motor vehicle, simulate the driving route based on the trend of the non-motor vehicle obtained by S5, and judge whether it will meet the vehicle at the turning area and enter the turning danger area of the vehicle; S6.2C. Identify the speed of the person, based on the direction of the person obtained in S5, simulate the person's action route, judge whether it will meet the vehicle in the turning area, and enter the turning danger area of the vehicle.

[0028] Based on the above prediction process, the prediction of dynamic objects around the vehicle is realized, which can effectively predict the possibility of surrounding objects entering the blind area, and achieve the effect of taking the first step.

[0029] Although the above is described centering on the embodiment, this is only an example and does not limit the present application, and those skilled in the art will clearly understand that various modifications and applications not exemplified above can be made within the scope of the essential characteristics of the embodiment. For example, each of the constituent elements specifically shown in the embodiment can be modified and implemented. Moreover, various differences related to such modifications and applications should be interpreted as being included in the scope of the present application defined in the appended claims.

Claims

1. A safe driving method for a chemical product truck, characterized in that: The following steps are included: S1. Based on vehicle navigation, obtain the corner information of the intersection ahead; S2. Obtain vehicle volume information and determine the risk areas at corners at different vehicle speeds. S3. When the vehicle enters the turning area after T seconds based on GPS prediction, the imaging device on the side to be turned is activated; The imaging device captures images within a range from N1 in front of the vehicle to N2 behind the vehicle; S4. Analyze the road conditions captured by the imaging device and identify all dynamic motor vehicles and people within the range of the vehicle to the sidewalk; S5. Dynamic tracking of dynamic motor vehicles and people, based on the dynamic motor vehicle indicator lights and the body performance of the characters to predict the direction of dynamic motor vehicles, non-motor vehicles and people; S6. Based on the predicted direction in S5, determine whether the dynamic motor vehicles, non-motor vehicles and people will enter the danger zone when the vehicle reaches the turning area and turns, and mark the predicted dangerous vehicles and people.

2. The safe driving method for a chemical product truck according to claim 1, characterized in that: In step S1 , the corner information includes: the real-time distance between the vehicle and the corner, the shape of the corner, and the surrounding information of the corner.

3. The safe driving method for a chemical product truck according to claim 1, characterized in that: In step S2, the risk area at the corner of the vehicle at different speeds is obtained through the vehicle turning model; The vehicle turning model is constructed as follows: Sc1. Collect videos of different vehicles at different corners, at different speeds and with different steering wheel rotation angles to form a video set S; Sc2. Obtain the image of each frame of each video to form an image set T i , mark the risk areas in the image and form a risk area subset T i , nested image set T i The risk areas of all frames in the video are formed to simulate the risk areas of the current video, forming the risk area set M i ; i is a label, represented by {c,n,s,t}, where c is the type of vehicle, n is the type of turning angle, s is the speed, and t is the steering wheel rotation angle; Sc4. Use machine learning to complete missing data and generate vehicle turning models for different vehicles at different corners, at different speeds, and with different steering wheel angles.

4. The safe driving method for a chemical product truck according to claim 3, characterized in that: In step Sc2 , different drivers are also labeled, that is, the i label is represented by {c, n, s, t, p}, where p is a person.

5. The safe driving method for a chemical product truck according to claim 1, characterized in that: In step S5 , the direction of the motor vehicle is detected by the turn signal.

6. The safe driving method for a chemical product truck according to claim 1, characterized in that: In step S5, the direction of non-motorized vehicles and people is predicted by prediction model 1; The prediction model is constructed as follows: Se1. Collect image data of non-motorized vehicles and people in motion to form a data set X; Se2. Classify the dataset X and label it as {d,m}, where d is the target object and m is the behavior pattern of d; Se3. Use machine learning to complete missing data and form prediction models for different target objects and different behavior patterns.

7. The safe driving method for a chemical product truck according to claim 6, characterized in that: In step Se2, data cleaning is also performed; The data cleaning comprises the following steps: Se2.

1. Split each image in dataset X into multiple segments or frames in different ways and label them as {d,m,q}, where q is a sequence number. Se2.

2. Perform duplication screening on segments from the same image data and clean and remove duplicate data; Se2.

3. Compare the data of different image materials after cleaning by Se2.2, find the duplicate data between different image materials, mark them as suspicious data, and mark the distinguished data as identified data.

8. The safe driving method for a chemical product truck according to claim 1, characterized in that: In step S6, the method for determining whether the dynamic motor vehicles, non-motor vehicles and people will enter the danger zone when the vehicle reaches the turning area and turns is as follows: S6.

1. Distinguish between dynamic motor vehicles, non-motor vehicles, and people. When the target object is a motor vehicle, proceed to S6.2A; When the target object is a non-motorized vehicle, proceed to S6.2B; When the target object is a person, proceed to S6.2C; S6.2A. Identify the speed of the motor vehicle and, based on the direction of the motor vehicle obtained in S5, simulate its route to determine whether it will meet the vehicle in the turning area and enter the vehicle's turning danger zone; S6.2B. Identify the speed of the non-motorized vehicle and, based on the direction of the non-motorized vehicle obtained in S5, simulate its route to determine whether it will meet the vehicle in the turning area and enter the vehicle's turning danger zone. S6.2C. Identify the person's speed and, based on the person's direction obtained in S5, simulate the person's route to determine whether the person will meet the vehicle in the turning area and enter the vehicle's turning danger zone.

Images