A forward collision warning optimization method and system based on a convolutional neural network

Abstract

The application relates to the technical field of vehicle auxiliary driving, and discloses a forward collision warning optimization method and system based on a convolutional neural network, which comprises the following steps: acquiring relevant information of a host vehicle and sending the information to other vehicles; screening effective vehicles around and acquiring basic information of the effective vehicles; judging the driving directions of the effective vehicles around; judging whether a preceding vehicle meets a condition of being overtaken; when the preceding vehicle meets the condition of being overtaken, screening effective vehicles with opposite driving directions to the host vehicle and acquiring screening basic information, building a convolutional neural network to acquire positioning information of the effective vehicles with opposite driving directions to the host vehicle; judging the effective vehicle with the most collision danger according to the screening basic information and the positioning information, and warning the host vehicle. The forward collision warning of the vehicle is realized by adopting the inter-vehicle communication and the convolutional neural network, and the accuracy of the forward collision warning is improved.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of vehicle auxiliary driving, in particular to a forward collision warning optimization method and system based on a convolutional neural network. BACKGROUND

[0002] With the increase in the number of cars, car safety becomes an important factor, and at present collision is the biggest factor affecting car safety. Especially when overtaking is needed, it is more likely to cause collision between vehicles.

[0003] CN109835251A discloses a vehicle forward collision warning system, which mainly acquires and judges external obstacles through a camera and a sensor, acquires information of the vehicle, and then judges whether there is a collision danger and issues a warning.

[0004] The above-mentioned technology acquires information of obstacles, but since forward collision warning has high real-time requirements, it is difficult to accurately make collision warning for dynamic vehicles using only the existing method. SUMMARY

[0005] The present application provides a forward collision warning optimization method and system based on a convolutional neural network, which uses vehicle-to-vehicle communication and convolutional neural network to make forward collision warning for vehicles, thereby improving the accuracy of forward collision warning.

[0006] The above-mentioned application object of the present application is realized by the following technical scheme:

[0007] A forward collision warning optimization method based on a convolutional neural network, comprising the following steps:

[0008] Acquiring relevant information of the host vehicle and sending it to other vehicles;

[0009] Screening effective vehicles around and acquiring basic information of the effective vehicles;

[0010] Judging the driving direction of the effective vehicles around;

[0011] Judging whether the preceding vehicle meets the overtaking condition;

[0012] When the preceding vehicle meets the overtaking condition, screening effective vehicles with opposite driving direction to the host vehicle and acquiring screening basic information, and building a convolutional neural network to acquire positioning information of the effective vehicles with opposite driving direction to the host vehicle;

[0013] According to the screening basic information and the positioning information, the effective vehicle with the most collision danger is judged, and the host vehicle is warned.

[0014] By adopting the technical scheme, the information is interacted through the inter-vehicle communication, the host vehicle can obtain the basic information of other vehicles around, and then the driving direction of the other vehicles is judged according to the basic information of the vehicles, when overtaking, the effective vehicle opposite to the driving direction of the host vehicle is judged as the vehicle that may collide, through the information of the vehicle opposite to the driving direction of the host vehicle and the vehicle positioning information obtained by the convolutional neural network, the driving tendency of the vehicle can be judged, then the effective vehicle most dangerous to collide is judged, and the host vehicle is warned, which greatly improves the accuracy of the collision warning judgment.

[0015] Optionally, the step of judging the driving direction of the effective vehicle around includes:

[0016] judging whether the heading angle of the effective vehicle around is same as the host vehicle;

[0017] judging whether the lateral distance difference between the two vehicles is within a threshold range;

[0018] When the heading angle of the effective vehicle around is same as the host vehicle and the lateral distance difference between the two vehicles is within the threshold range, the effective vehicle and the host vehicle are considered to be on the same lane.

[0019] By adopting the technical scheme, the threshold range is set according to the range of the required warning, and then the driving direction of the effective vehicle is judged, and some situations with wide range and not easy to exist collision risk are excluded.

[0020] Optionally, the step of judging whether the front vehicle meets the overtaking condition includes:

[0021] judging whether the front vehicle is in a straight running state;

[0022] judging whether the left and right turn signals of the vehicle are in an off state.

[0023] By adopting the technical scheme, the overtaking is ensured to be in a safe state through the judgment of the front vehicle.

[0024] Optionally, the step of building a convolutional neural network to obtain the positioning information of the effective vehicle opposite to the driving direction of the host vehicle includes:

[0025] image collection is performed on the effective vehicle opposite to the driving direction of the host vehicle;

[0026] the collected image data is input into a PoseNet convolutional neural network for calculation;

[0027] the motion characteristics of the effective vehicle opposite to the driving direction of the host vehicle are obtained through multiple iterations and training;

[0028] finally, the positioning information of the effective vehicle opposite to the driving direction of the host vehicle is obtained from the collected image data.

[0029] By adopting the technical scheme, the accuracy of obtaining the position of the vehicle is improved through the PoseNet convolutional neural network to obtain the positioning information, so that the host vehicle can better perform forward collision warning.

[0030] Optionally, the step of determining the effective vehicle with the greatest collision danger according to the screening of the basic information and the positioning information comprises:

[0031] determining the first driving tendency of the vehicle according to the screening of the basic information;

[0032] determining the second driving tendency of the vehicle according to the positioning information;

[0033] estimating the driving tendency of the vehicle according to the first driving tendency and the second driving tendency;

[0034] determining the effective vehicle with the greatest collision danger according to the driving tendency of the vehicle.

[0035] By adopting the technical scheme, the first driving tendency of the vehicle can be determined according to the information between the vehicles, the real-time positioning information of the vehicle is obtained through the convolutional neural network, the second driving tendency of the vehicle can be estimated, and then the driving tendency of the vehicle is estimated according to the first driving tendency and the second driving tendency, so that the host vehicle can more accurately determine the driving tendency of the vehicle and improve the accuracy of forward collision warning.

[0036] Optionally, when it is determined that there is an effective vehicle with the greatest collision danger, the collision time is calculated, and the steps comprise:

[0037] calculating the forward collision time of the vehicle according to the screening of the basic information and the determination of the first driving tendency;

[0038] calculating the forward collision time of the vehicle according to the positioning information of the PoseNet convolutional neural network and the second driving tendency;

[0039] taking the minimum value of the collision time calculated in the two ways as the collision time.

[0040] By adopting the technical scheme, the minimum value is taken as the collision time, so that the host vehicle can make a correct response to the forward collision warning as much as possible, and safety problems caused by unreasonable collision time determination are avoided.

[0041] The above invention purpose of the application is realized through the following technical scheme:

[0042] A forward collision warning optimization system based on a convolutional neural network comprises:

[0043] an information transceiving module configured to send relevant information of the host vehicle to other vehicles and obtain basic information of effective vehicles;

[0044] a positioning module, a convolutional neural network is built to obtain positioning information of the effective vehicle opposite to the driving direction of the host vehicle

[0045] a processing module, which judges the driving direction of the effective vehicle according to the basic information of the effective vehicle, judges whether the front vehicle meets the overtaking condition, and when the front vehicle meets the overtaking condition, screens the effective vehicle opposite to the driving direction of the host vehicle and obtains screening basic information, and judges the effective vehicle most dangerous for collision according to the screening basic information and the positioning information;

[0046] a host vehicle warning module, which warns the host vehicle when there is an effective vehicle dangerous for collision.

[0047] Optionally, the positioning module comprises:

[0048] an image acquisition unit, which is used to acquire images of the effective vehicle opposite to the driving direction of the host vehicle;

[0049] an image processing unit, which inputs the acquired image data into the PoseNet convolutional neural network for calculation, obtains the motion characteristics of the effective vehicle opposite to the driving direction of the host vehicle through multiple iterations and training, and finally obtains the positioning information of the effective vehicle opposite to the driving direction of the host vehicle from the acquired image data.

[0050] Optionally, the processing module comprises a first processing unit and a second processing unit,

[0051] the first processing unit judges the first driving tendency of the vehicle according to the screening basic information;

[0052] the second processing unit judges the second driving tendency of the vehicle according to the positioning information;

[0053] the processing module combines the first driving tendency and the second driving tendency to estimate the driving tendency of the vehicle;

[0054] the processing module determines the effective vehicle most dangerous for collision according to the driving tendency of the vehicle.

[0055] The above invention purpose of the present application is realized by the following technical scheme:

[0056] A computer storage medium, which stores a computer program, the program is executed by a processor to realize the forward collision warning optimization method of the convolutional neural network as above.

[0057] In summary, the present application includes at least one of the following beneficial technical effects:

[0058] The application obtains the information of other vehicles by the way of communication between vehicles, so as to judge the driving direction and overtaking condition of the vehicle, and more accurately understand the current dynamic of other vehicles; then the convolutional neural network is used to obtain the positioning information of other vehicles, so that the real-time positioning of the vehicle is more accurate, the judgment of the driving tendency of the vehicle is facilitated, and the accuracy and precision of the forward collision warning judgment are improved. BRIEF DESCRIPTION OF DRAWINGS

[0059] Figure 1 It is a flow chart of a forward collision warning optimization method based on a convolutional neural network according to an embodiment of the application.

[0060] Figure 2 It is a structure block diagram of a forward collision warning optimization system based on a convolutional neural network according to an embodiment of the application. DETAILED DESCRIPTION

[0061] The following will be described in detail in combination with the accompanying Figures 1-2 The application will be further described in detail.

[0062] Reference Figure 1 A forward collision warning optimization method based on a convolutional neural network, comprising the following steps:

[0063] Step S100, obtaining the relevant information of the host vehicle and sending it to other vehicles;

[0064] Based on V2X communication technology, in road traffic, vehicles interact with each other through this technology to send their basic information such as position information, longitude and latitude, speed information, heading angle information, and vehicle state information to other vehicles, and also receive the basic information of other vehicles and then perform algorithm processing operation to judge whether there is a collision danger with the oncoming vehicle in the process of overtaking by borrowing the road.

[0065] Step S200, screening the surrounding effective vehicles and obtaining the basic information of the effective vehicles;

[0066] In this embodiment, the effective vehicle refers to a vehicle driving within a certain range, excluding vehicles that are parking or waiting.

[0067] Step S300, judging the driving direction of the surrounding effective vehicles; the steps include:

[0068] Judging whether the heading angle of the surrounding effective vehicles is the same as that of the host vehicle;

[0069] Judging whether the lateral distance difference between the two vehicles is within the threshold range;

[0070] When the heading angle of the effective vehicle is the same as that of the host vehicle and the lateral distance difference between the two vehicles is within the threshold range, the effective vehicle and the host vehicle are considered to be in the same lane.

[0071] The threshold range is set according to the needs, and the lateral distance difference between the host vehicle and the vehicle that may cause a collision is ensured to be within the threshold range, where the lateral distance difference refers to the relative width of the distance between the two vehicles.

[0072] Step S400, determine whether the front vehicle meets the overtaking condition; the steps include:

[0073] Determine whether the front vehicle is in a straight running state;

[0074] Determine whether the left and right turn signals of the vehicle are in an off state.

[0075] It is known in daily driving that overtaking can only be performed when the current vehicle is in a straight running state. By this judgment, safety hazards caused by the current vehicle deviating and the host vehicle not being aware of it can also be avoided, and the host vehicle can timely avoid risks.

[0076] Step S500, when the front vehicle meets the overtaking condition, screen effective vehicles in the opposite direction of the host vehicle and obtain screening basic information, and build a convolutional neural network to obtain positioning information of the effective vehicles in the opposite direction of the host vehicle;

[0077] Step S600, determine the effective vehicle with the most collision danger according to the screening basic information and the positioning information, and give a warning to the host vehicle.

[0078] Through the communication between vehicles, the host vehicle can obtain the basic information of other vehicles around, and then determine the driving direction of other vehicles according to the basic information of the vehicles. When overtaking, the effective vehicles in the opposite direction of the host vehicle are determined as the vehicles that may cause a collision. Through the information of the vehicles driving in the opposite direction of the host vehicle and the positioning information of the vehicles obtained by the convolutional neural network, the driving tendency of the vehicles can be determined, and then the effective vehicle with the most collision danger can be determined, and a warning is given to the host vehicle. This method greatly improves the accuracy of collision warning and judgment.

[0079] The steps of building a convolutional neural network to obtain the positioning information of the effective vehicles in the opposite direction of the host vehicle in step S500 include:

[0080] Step S510, image collection is performed on the effective vehicles in the opposite direction of the host vehicle;

[0081] Step S520, input the collected image data into the PoseNet convolutional neural network for calculation;

[0082] Step S530, the motion characteristics of the effective vehicle opposite to the driving direction of the host vehicle are obtained through multiple iterations and training.

[0083] Step S540, the positioning information of the effective vehicle opposite to the driving direction of the host vehicle is finally obtained from the collected image data.

[0084] PoseNet is a convolutional neural network for real-time 6-DOF camera localization. A convolutional neural network trained by the system takes a single RGB image as input and performs depth regression, and the output is a 6-DOF pose matrix of the camera. This algorithm can run in real time indoors and outdoors, and each frame takes 5ms to calculate. For large outdoor scenes, the accuracy is about 2m and 3°, and the indoor accuracy is about 0.5m and 5°. At the same time, PoseNet uses two techniques to overcome the expensive cost limitations of training a convolutional neural network that relies on very large labeled image datasets: one technique is automatic labeling data, which uses motion structure to generate a large regression dataset of camera poses. Another technique is transfer learning: training a pose regressor on a huge image recognition dataset, pre-training it as a classifier. Compared with training from scratch, this method can converge to smaller errors in a shorter time, even with very sparse training sets.

[0085] Network model of PoseNet: The position vector p in the output 6-DOF pose matrix is given by the 3D camera coordinates x, and the direction is represented by the quaternion q. The above parameters can be represented by the formula: p=[x,q]. The reason for choosing quaternion as the direction representation is that any four-dimensional value can be easily mapped to a standard rotation by normalizing them to unit length.

[0086] In this embodiment, since PoseNet is used, multi-angle image acquisition of other vehicles can be performed, and then the images of the to-be-tested regions are respectively intercepted from the collected images. After processing the images of the to-be-tested regions respectively, multiple iterations and training are performed to ensure that the motion characteristics of the vehicle can be obtained. Then, the positioning information of the vehicle is obtained according to the image data. Since the vehicle is always driving, the process of obtaining the positioning information is uninterrupted to obtain real-time positioning information. The obtained information can be stored in the processing module of the vehicle, so that the results obtained by PoseNet can be used for collision warning judgment.

[0087] Step S600 includes the steps of judging the effective vehicle with the most collision danger according to the screened basic information and positioning information.

[0088] Step S610, judging the first driving tendency of the vehicle according to the screened basic information;

[0089] First, the vehicle is classified according to the driving direction of the vehicle, and then the first driving tendency of the vehicle is estimated according to the heading angle, speed and other information of the vehicle, combined with the distance information between the host vehicle and the effective vehicle. Under the condition of meeting the overtaking condition, the first driving tendency here mainly needs to pay attention to the tendency of the vehicle driving in reverse.

[0090] Step S620, judging the second driving tendency of the vehicle according to the positioning information;

[0091] Since the convolutional neural network obtains the real-time position information of the vehicle, the second driving tendency of the vehicle can be summarized and estimated according to a plurality of real-time position information. The driving tendency here is mainly obtained according to the law of position change.

[0092] Step S630, estimating the driving tendency of the vehicle according to the first driving tendency and the second driving tendency;

[0093] After combining the two driving tendencies, different real-time positions have corresponding heading angle, speed and steering information, so that a more accurate driving tendency can be obtained to better estimate and judge the forward collision warning.

[0094] Step S640, determining the effective vehicle with the most collision danger according to the driving tendency of the vehicle.

[0095] When determining the effective vehicle with the most collision danger, the vehicle with the most tendency to approach the host vehicle can be judged as the most dangerous, or other judgment criteria can be used.

[0096] When judging that there is an effective vehicle with the most collision danger, the collision time is calculated, and the steps include:

[0097] Calculating the forward collision time of the vehicle according to the judgment of the screening basic information and the first driving tendency;

[0098] Calculating the forward collision time of the vehicle according to the positioning information of PoseNet convolutional neural network and the second driving tendency;

[0099] Taking the minimum value of the collision time calculated by the two methods as the collision time.

[0100] The first driving tendency and the second driving tendency are obtained in the last step, so the collision time is calculated for the two driving tendencies respectively here. Although the total driving tendency of the vehicle is the superposition of the first driving tendency and the second driving tendency, the minimum value is taken in the collision time. The purpose is to better warn the host vehicle of the forward collision. If the average value is used, there may be a case where the actual collision time is the minimum value, so it is possible to avoid this situation for the host vehicle as much as possible.

[0101] Reference Figure 2The application discloses a forward collision warning optimization system based on a convolutional neural network, which comprises the following:

[0102] An information transceiving module is arranged to send relevant information of the host vehicle to other vehicles and acquire basic information of the effective vehicles;

[0103] A positioning module is arranged to acquire positioning information of the effective vehicles opposite to the driving direction of the host vehicle by building a convolutional neural network

[0104] A processing module is arranged to judge the driving direction of the effective vehicles according to the basic information of the effective vehicles, judge whether the preceding vehicle meets the overtaking condition, and when the preceding vehicle meets the overtaking condition, screen the effective vehicles opposite to the driving direction of the host vehicle and acquire screening basic information, and judge the effective vehicle most dangerous in collision according to the screening basic information and the positioning information;

[0105] A host vehicle warning module is arranged to warn the host vehicle when the effective vehicle most dangerous in collision exists.

[0106] The processing module not only receives and judges information, but also stores the collected information, so that the forward collision warning can refer to continuous information of the vehicles and make more accurate judgments.

[0107] The positioning module comprises:

[0108] An image collecting unit is arranged to collect images of the effective vehicles opposite to the driving direction of the host vehicle, and in the embodiment, the image collecting unit can be an information collecting component such as a camera and a sensor.

[0109] An image processing unit is arranged to input the collected image data into a PoseNet convolutional neural network for calculation, obtain the motion characteristics of the effective vehicles opposite to the driving direction of the host vehicle through multiple iterations and training, and finally acquire the positioning information of the effective vehicles opposite to the driving direction of the host vehicle from the collected image data. The image processing unit is arranged on the processing module and is mainly used after the convolutional neural network is built.

[0110] The processing module comprises a first processing unit and a second processing unit,

[0111] The first processing unit is arranged to judge the first driving tendency of the vehicle according to the screening basic information;

[0112] The second processing unit is arranged to judge the second driving tendency of the vehicle according to the positioning information;

[0113] The processing module is arranged to estimate the driving tendency of the vehicle according to the first driving tendency and the second driving tendency;

[0114] The processing module is arranged to determine the effective vehicle most dangerous in collision according to the driving tendency of the vehicle.

[0115] A computer storage medium, which stores a computer program, the program is executed by a processor to realize the forward collision warning optimization method of the convolutional neural network as described above.

[0116] The following application scenarios exist:

[0117] A, B, C and D are set as four vehicles, A is the front vehicle of the host vehicle, B and C are vehicles driving in the opposite direction, and D is a vehicle that is parking, when effective vehicle screening is performed, D is first excluded, then the driving directions of A, B and C are judged, and it is determined that A meets the overtaking condition, at this time, whether B and C have collision risks is judged, through the first driving tendency, the collision time of B is 2 minutes, and the collision time of C is 2 minutes and 30 seconds, through the second driving tendency, the collision time of B is 1 minute and 30 seconds, and the collision time of C is 2 minutes, at this time, it can be judged that B is the most dangerous effective vehicle with collision risk, and the predicted collision time is 1 minute and 30 seconds, through the loudspeaker or display screen in the vehicle and other devices, the driver is warned, so that the driver of the host vehicle can perform overtaking and other operations according to the warning information, and the safety of the host vehicle is ensured.

[0118] The application obtains the information of other vehicles in a communication manner between vehicles, so as to more accurately understand the current dynamics of other vehicles by judging the driving direction and overtaking condition of the vehicle, and then uses the convolutional neural network to obtain the positioning information of other vehicles, so that the real-time positioning of the vehicle is more accurate, the judgment of the driving tendency of the vehicle is facilitated, and the accuracy and precision of the forward collision warning judgment are improved.

[0119] The specific embodiments are only an explanation of the application, and are not a limitation of the application, and those skilled in the art can make non-creative modifications to the embodiments according to the needs after reading the specification, but as long as the modifications are within the scope of the claims of the application, they are protected by the patent law.

[0120] To make the purposes, technical solutions and advantages of the embodiments of the application more clear, 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 some of the embodiments of the application, but not all the embodiments of the application. Based on the embodiments in the application, all other embodiments obtained by those skilled in the art without creative labor are within the protection scope of the application.

[0121] In addition, the term "and / or" in this paper is only to describe the association relationship of the associated objects, which means that there may be three relationships, for example, A and / or B, which means that there are three cases of A alone, A and B together, and B alone. In addition, the character " / " in this paper generally represents an "or" relationship between the front and rear associated objects unless otherwise specified.

Claims

1. A forward collision warning optimization method based on a convolutional neural network, characterized in that, The method comprises the following steps: acquiring relevant information of a host vehicle and sending the information to other vehicles; screening effective vehicles around the host vehicle and acquiring basic information of the effective vehicles; judging driving directions of the effective vehicles around the host vehicle; judging whether a front vehicle meets a condition of being overtaken; when the front vehicle meets the condition of being overtaken, screening effective vehicles in a driving direction opposite to that of the host vehicle and acquiring screening basic information, and building a convolutional neural network to acquire positioning information of the effective vehicles in the driving direction opposite to that of the host vehicle; judging an effective vehicle most dangerous in collision according to the screening basic information and the positioning information, and warning the host vehicle, which comprises the following steps: judging a first driving tendency of a vehicle according to the screening basic information; judging a second driving tendency of the vehicle according to the positioning information; combining the first driving tendency and the second driving tendency to estimate a driving tendency of the vehicle; determining the effective vehicle most dangerous in collision according to the driving tendency of the vehicle; when it is judged that there is the effective vehicle most dangerous in collision, calculating a collision time, which comprises the following steps: calculating a front collision time of the vehicle according to the screening basic information and the judgment of the first driving tendency; calculating the front collision time of the vehicle according to the positioning information and the second driving tendency of the PoseNet convolutional neural network; taking a minimum value of the collision times calculated in the two manners as the collision time.

2. The method of claim 1, wherein, The step of judging the driving directions of the effective vehicles around the host vehicle comprises the following steps: judging whether a heading angle of an effective vehicle around the host vehicle is the same as that of the host vehicle; judging whether a lateral distance difference between the effective vehicle and the host vehicle is within a threshold range; when the heading angle of the effective vehicle around the host vehicle is the same as that of the host vehicle and the lateral distance difference between the effective vehicle and the host vehicle is within the threshold range, the effective vehicle and the host vehicle are considered to be on the same lane.

3. The method of claim 1, wherein, The step of judging whether the front vehicle meets the condition of being overtaken comprises the following steps: judging whether the front vehicle is in a straight driving state; judging whether left and right turn signals of the vehicle are in an off state.

4. The method of claim 1, wherein, The step of building the convolutional neural network to acquire the positioning information of the effective vehicle in the driving direction opposite to that of the host vehicle comprises the following steps: collecting images of the effective vehicle in the driving direction opposite to that of the host vehicle; inputting the collected image data into the PoseNet convolutional neural network for calculation; obtaining motion characteristics of the effective vehicle in the driving direction opposite to that of the host vehicle through multiple iterations and training; finally acquiring the positioning information of the effective vehicle in the driving direction opposite to that of the host vehicle from the collected image data.

5. A forward collision warning optimization system based on a convolutional neural network, for implementing the forward collision warning optimization method according to any one of claims 1 to 4, characterized in that, The method comprises the following steps: an information transceiver module, which sends relevant information of a host vehicle to other vehicles and acquires basic information of effective vehicles; a positioning module, which builds a convolutional neural network to acquire positioning information of effective vehicles in a driving direction opposite to that of the host vehicle a processing module, which judges driving directions of the effective vehicles according to the basic information of the effective vehicles, judges whether a front vehicle meets a condition of being overtaken, screens the effective vehicles in the driving direction opposite to that of the host vehicle and acquires screening basic information when the front vehicle meets the condition of being overtaken, and judges an effective vehicle most dangerous in collision according to the screening basic information and the positioning information; a host vehicle warning module, which warns the host vehicle when there is the effective vehicle most dangerous in collision.

6. The system of claim 5, wherein, The positioning module comprises the following steps: an image collection unit, which is used for collecting images of the effective vehicle in the driving direction opposite to that of the host vehicle; An image processing unit inputs the collected image data into a PoseNet convolutional neural network for calculation, and obtains the motion characteristics of the effective vehicle opposite to the driving direction of the host vehicle through multiple iterations and training.

7. The system of claim 5, wherein, The processing module comprises a first processing unit and a second processing unit, The first processing unit determines the first driving tendency of the vehicle according to the screening basic information; The second processing unit determines the second driving tendency of the vehicle according to the positioning information; The processing module combines the first driving tendency and the second driving tendency to estimate the driving tendency of the vehicle; The processing module determines the effective vehicle with the most collision danger according to the driving tendency of the vehicle.

8. A computer storage medium having stored thereon a computer program, characterized in that The program is executed by the processor to implement the forward collision warning optimization method of the convolutional neural network according to any one of claims 1 to 4.

Images