CN117864075A

CN117864075A - Active brake control method, device and storage medium

Info

Publication number: CN117864075A
Application number: CN202410153663.0A
Authority: CN
Inventors: 刘新; 刘子元
Original assignee: Shenzhen Yixin Yiyi Software Development Co ltd
Current assignee: Shenzhen Yixin Yiyi Software Development Co ltd
Priority date: 2024-02-02
Filing date: 2024-02-02
Publication date: 2024-04-12

Abstract

The invention discloses an active braking control method, active braking control equipment and a storage medium, and belongs to the technical field of active braking of vehicles. The invention collects the environmental information around the vehicle; identifying and tracking an obstacle in front of the vehicle by using an obstacle identification model according to the surrounding environment information of the vehicle to obtain obstacle information; predicting collision risk of the vehicle and the obstacle based on the obstacle information to obtain a collision risk prediction result; and if the collision risk prediction result meets the brake triggering condition, sending a brake instruction to a brake control unit, and braking by the brake control unit. According to the scheme, the obstacle recognition model is utilized to recognize and track the obstacle in the environment information, so that the accuracy of the obstacle recognition and tracking is improved, the reliability of the obstacle information is improved, the accuracy of collision risk prediction is further improved, and finally the judgment accuracy of the vehicle active braking system is improved.

Description

Active brake control method, device and storage medium

Technical Field

The present invention relates to the field of active braking technologies for vehicles, and in particular, to an active braking control method, device and storage medium.

Background

With the rapid development of the transportation industry, road traffic accidents frequently occur, and once the traffic accidents occur, serious personal safety injury and property loss can be caused. Therefore, automobile manufacturers have carried out various safety designs aiming at vehicles, and besides passive safety systems such as safety belts, safety airbags, bumpers and the like, active braking systems are also designed, so that the automobiles can actively take measures to avoid accidents.

At present, an active braking system of a vehicle, such as an AEBS (emergency braking assistance system) and the like, mainly utilizes sensors such as a laser radar, a millimeter wave radar and the like to measure distance of an obstacle, and controls alarming and braking according to the distance between the vehicle and the obstacle. However, the existing active braking system for vehicles has a problem that the judgment accuracy of whether collision with an obstacle occurs is not high enough.

Therefore, how to improve the accuracy of the judgment of the active braking system of the vehicle is a technical problem that needs to be solved by those skilled in the art.

Disclosure of Invention

The invention mainly aims to provide an active brake control method, active brake control equipment and a storage medium, and aims to solve the technical problem of how to improve the judgment accuracy of an active brake system of a vehicle.

In order to achieve the above object, the present invention provides an active brake control method, which includes the following steps:

collecting environmental information around a vehicle;

identifying and tracking an obstacle in front of the vehicle by using an obstacle identification model according to the surrounding environment information of the vehicle to obtain obstacle information;

predicting collision risk of the vehicle and the obstacle based on the obstacle information to obtain a collision risk prediction result;

and if the collision risk prediction result meets the brake triggering condition, sending a brake instruction to a brake control unit, and braking by the brake control unit.

Optionally, the step of collecting environmental information around the vehicle includes:

collecting ultrasonic reflection information around the vehicle by using an ultrasonic sensor;

collecting laser reflection information around the vehicle by using a laser radar;

acquiring video image information around the vehicle by using a visual sensor;

and forming the ultrasonic reflection information, the laser reflection information and/or the video image information into environment information around the vehicle.

Optionally, the step of identifying and tracking the obstacle in front of the vehicle by using the obstacle identification model according to the environmental information around the vehicle to obtain the obstacle information includes:

Extracting characteristics in the environment information by using a pre-trained obstacle recognition model, and recognizing the obstacle in the environment information after the multi-layer neural network layer of the pre-trained obstacle recognition model is processed;

tracking the obstacle according to an obstacle recognition model to obtain the size of the obstacle, the relative distance between the obstacle and the vehicle and the relative speed;

the obstacle information is composed of a size of the obstacle, a relative distance of the obstacle from the vehicle, and the relative speed.

Optionally, the step of using the pre-trained obstacle recognition model to extract the features in the environmental information, after processing the multi-layer neural network layer of the pre-trained obstacle recognition model, before the step of recognizing the obstacle in the environmental information, further includes:

making an obstacle data set by utilizing the pre-collected vehicle environment information, and dividing the obstacle data set into an obstacle training set and an obstacle testing set;

training the obstacle recognition model by using the obstacle training set to obtain a trained obstacle recognition model;

testing the trained obstacle recognition model by using an obstacle test set to obtain a first test result of obstacle recognition and tracking;

And repeating training and testing on the trained obstacle recognition model according to the first test result until a final obstacle recognition model is output.

Optionally, the step of repeating training and testing the trained obstacle recognition model according to the first test result until a final obstacle recognition model is output includes:

a parameter adjustment step, according to the first test result, adjusting parameters of the trained obstacle recognition model according to characteristics of different road conditions and different obstacles;

training and testing, namely repeatedly training the adjusted obstacle recognition model by using an obstacle training set with different road conditions and different obstacle types to obtain a new obstacle recognition model, and testing the new obstacle recognition model by using an obstacle testing set with different road conditions and different obstacle types to obtain a second test result of obstacle recognition and tracking;

outputting a final obstacle recognition model if the second test result meets the expected requirement; otherwise, repeating the parameter adjustment step and the training test step until the second test result reaches the expected requirement, and outputting a final obstacle recognition model.

Optionally, the step of predicting a collision risk of the vehicle with the obstacle based on the obstacle information, and obtaining a collision risk prediction result includes:

screening abnormal data from the obstacle information to obtain risk prediction data;

and extracting data features in the risk prediction data by using a pre-trained collision risk prediction model, and predicting the collision risk of the vehicle and the obstacle by using a classifier of the pre-trained collision risk prediction model to obtain a collision risk prediction result.

Optionally, the step of extracting data features in the risk prediction data by using a pre-trained collision risk prediction model, predicting collision risk of the vehicle and the obstacle by using a classifier of the pre-trained collision risk prediction model, and obtaining a collision risk prediction result further includes:

creating a risk prediction data set by utilizing pre-collected risk prediction data, and dividing the risk prediction data set into a risk prediction training set and a risk prediction test set;

training the collision risk prediction model by using a risk prediction training set to obtain a trained collision risk prediction model;

Testing the trained collision risk prediction model by using a risk prediction test set to obtain a third test result of collision risk prediction;

and according to the third test result, repeating training and testing on the trained collision risk prediction model until a final collision risk prediction model is output.

Optionally, the step of repeatedly training and testing the trained collision risk prediction model according to the third test result until the final collision risk prediction model is output includes:

a parameter adjustment step, according to the third test result, adjusting parameters of the collision risk prediction model according to characteristics of different road conditions and different obstacles;

training and testing, namely repeatedly training the adjusted collision risk prediction model by using risk prediction training sets with different road conditions and different obstacle types to obtain a new collision risk prediction model, and testing the new collision risk prediction model by using risk prediction testing sets with different road conditions and different obstacle types to obtain a fourth test result of collision risk prediction;

if the fourth test result meets the expected requirement, outputting a final collision risk prediction model; otherwise, repeating the parameter adjustment step and the training test step until the fourth test result reaches the expected requirement, and outputting a final collision risk prediction model.

Optionally, if the collision risk prediction result meets a brake triggering condition, the step of sending a brake instruction to a brake control unit, where the step of braking by the brake control unit includes:

judging whether the collision risk prediction result exceeds a preset threshold value by utilizing a decision algorithm;

if the collision risk prediction result exceeds a preset threshold value, a braking instruction is sent;

and the brake command is transmitted to a brake control unit through a CAN bus, and the brake control unit executes braking.

The embodiment of the application also provides an active brake control device, which comprises:

the acquisition module is used for acquiring environmental information around the vehicle;

the identifying module is used for identifying and tracking the obstacle in front of the vehicle by utilizing the obstacle identifying model according to the surrounding environment information of the vehicle to obtain obstacle information;

the prediction module is used for predicting the collision risk of the vehicle and the obstacle based on the obstacle information to obtain a collision risk prediction result;

and the braking module is used for sending a braking instruction to the braking control unit and braking by the braking control unit if the collision risk prediction result meets the braking triggering condition.

The embodiment of the application also provides an active brake control device, which comprises: the system comprises a memory, a processor and an active brake control program stored on the memory and capable of running on the processor, wherein the active brake control program is configured to realize the steps of the active brake control method.

The embodiment of the application also provides a computer readable storage medium, wherein the computer readable storage medium stores an active brake control program, and the active brake control program realizes the steps of the active brake control method when being executed by a processor.

The invention provides an active brake control method, an active brake control device, active brake control equipment and a storage medium. The invention collects the environmental information around the vehicle; identifying and tracking an obstacle in front of the vehicle by using an obstacle identification model according to the surrounding environment information of the vehicle to obtain obstacle information; predicting collision risk of the vehicle and the obstacle based on the obstacle information to obtain a collision risk prediction result; and if the collision risk prediction result meets the brake triggering condition, sending a brake instruction to a brake control unit, and braking by the brake control unit. According to the scheme, the surrounding environment information of the vehicle is collected, and the obstacle recognition model is utilized to recognize and track the obstacle in the environment information, so that the accuracy of recognizing and tracking the obstacle is improved, the speed of recognizing and tracking the obstacle is accelerated, and the reliability of the obstacle information is improved; the collision risk prediction based on the high-reliability obstacle information is improved in accuracy, and the effect of improving the judgment accuracy of the active brake control device is finally achieved.

Drawings

FIG. 1 is a schematic diagram of an active brake control device of a hardware operating environment according to an embodiment of the present invention;

FIG. 2 is a flowchart of a first embodiment of an active brake control method according to the present invention.

The achievement of the objects, functional features and advantages of the present invention will be further described with reference to the accompanying drawings, in conjunction with the embodiments.

Detailed Description

It should be understood that the specific embodiments described herein are for purposes of illustration only and are not intended to limit the scope of the invention.

The main solutions of the embodiments of the present application are: collecting environmental information around the vehicle; identifying and tracking an obstacle in front of the vehicle by using an obstacle identification model according to the surrounding environment information of the vehicle to obtain obstacle information; predicting collision risk of the vehicle and the obstacle based on the obstacle information to obtain a collision risk prediction result; and if the collision risk prediction result meets the brake triggering condition, sending a brake instruction to a brake control unit, and braking by the brake control unit. According to the scheme, the surrounding environment information of the vehicle is collected, and the obstacle recognition model is utilized to recognize and track the obstacle in the environment information, so that the accuracy of recognizing and tracking the obstacle is improved, the speed of recognizing and tracking the obstacle is accelerated, and the reliability of the obstacle information is improved; the collision risk prediction based on the high-reliability obstacle information is improved in accuracy, and the effect of improving the judgment accuracy of the active brake control device is finally achieved.

At present, an active braking system of a vehicle, such as an AEBS (emergency braking assistance system) and the like, mainly utilizes sensors such as a laser radar, a millimeter wave radar and the like to measure distance of an obstacle, and controls alarming and braking according to the distance between the vehicle and the obstacle. However, the existing active braking system for vehicles has a problem that the judgment accuracy of whether collision with an obstacle occurs is not high enough. Therefore, how to improve the accuracy of the judgment of the active braking system of the vehicle is a technical problem that needs to be solved by those skilled in the art.

Based on the above, the embodiment of the application provides a solution, by collecting the environmental information around the vehicle and utilizing the obstacle recognition model to recognize and track the obstacle in the environmental information, the accuracy of recognizing and tracking the obstacle is improved, the speed of recognizing and tracking the obstacle is accelerated, and the reliability of the obstacle information is also improved; the collision risk prediction based on the high-reliability obstacle information is improved in accuracy, and the effect of improving the judgment accuracy of the active brake control device is finally achieved.

Specifically, referring to fig. 1, fig. 1 is a schematic functional block diagram of a terminal device to which an active brake control device of the present application belongs. The active brake control device may be a device independent of the terminal device and capable of performing data processing, and may be carried on the terminal device in a form of hardware or software. The terminal device can be an intelligent mobile terminal such as a vehicle-mounted terminal and a tablet personal computer, and the mobile phone is used for example in the embodiment.

In this embodiment, the terminal device to which the touch screen gesture operation control device belongs at least includes an output module 110, a processor 120, a memory 130, and a communication module 140.

The memory 130 stores an operating system and an active brake control program, and the active brake control device can store acquired environmental information, obtained obstacle information, collision risk prediction results and other information in the memory 130; the output module 110 may be a display screen, a speaker, etc. The communication module 140 may include a WIFI module, a mobile communication module, a bluetooth module, and the like, and communicates with an external device or a server through the communication module 140.

It will be appreciated by those skilled in the art that the configuration shown in FIG. 1 is not limiting and may include more or fewer components than shown, or may be combined with certain components, or a different arrangement of components.

Wherein the active brake control program in the memory 130 when executed by the processor performs the steps of:

collecting environmental information around a vehicle;

Further, the active brake control program in the memory 130 when executed by the processor also performs the steps of:

acquiring video image information around the vehicle by using a visual sensor;

and the ultrasonic reflection information, the laser reflection information and/or the video image information are combined to form the environment information around the vehicle.

manufacturing an obstacle data set by using the environment information, and dividing the obstacle data set into an obstacle training set and an obstacle testing set;

and extracting data features in the risk prediction data by using a pre-trained collision risk prediction model, predicting the collision risk of the vehicle and the obstacle by using a classifier of the pre-trained collision risk prediction model, and obtaining a collision risk prediction result.

creating a risk prediction data set by utilizing the pre-collected risk prediction data, and dividing the risk prediction data set into a risk prediction training set and a risk prediction test set;

According to the scheme, the method and the device specifically collect the environmental information around the vehicle; identifying and tracking an obstacle in front of the vehicle by using an obstacle identification model according to the surrounding environment information of the vehicle to obtain obstacle information; predicting collision risk of the vehicle and the obstacle based on the obstacle information to obtain a collision risk prediction result; and if the collision risk prediction result meets the brake triggering condition, sending a brake instruction to a brake control unit, and braking by the brake control unit. According to the scheme, the surrounding environment information of the vehicle is collected, and the obstacle recognition model is utilized to recognize and track the obstacle in the environment information, so that the accuracy of recognizing and tracking the obstacle is improved, the speed of recognizing and tracking the obstacle is accelerated, and the reliability of the obstacle information is improved; the collision risk prediction based on the high-reliability obstacle information is improved in accuracy, and the effect of improving the judgment accuracy of the active brake control device is finally achieved.

Based on the above terminal device architecture, but not limited to the above architecture, the method embodiments of the present application are presented.

Referring to fig. 2, fig. 2 is a flowchart illustrating a first embodiment of an active brake control method according to the present invention.

As shown in fig. 2, in this embodiment, the active brake control method includes:

step S10, collecting environmental information around the vehicle;

in this embodiment, how to obtain accurate obstacle information, and accurately pre-judging the risk of collision between the obstacle and the vehicle according to the obstacle information is an urgent problem to be solved by the active braking technology of the vehicle. With the increasing number of vehicles on the road surface, the number of vehicles running on the road surface is increased, the road conditions faced by people are also becoming more complex, and in many cases, it is difficult for people to accurately judge whether the vehicles collide. For example, when a vehicle is traveling on a dim road, if there is a dark obstacle on the road, the ability of the human eye to recognize the dark obstacle in the dark environment is poor, the driver may need to find the obstacle when the driver is closer to the obstacle, and braking may be too late after finding the obstacle; as another example, the driver is tired after driving for a long time, the reaction speed is slow, and it is difficult to react in time when encountering an emergency such as sudden braking of a front vehicle, crossing of a road by pedestrians on both sides of the road, etc., resulting in occurrence of traffic accidents. In short, due to limitations in various aspects such as eyesight and response speed of a person, traffic accidents are difficult to avoid by only actively braking the vehicle by the driver in many cases.

Therefore, there is a need for a vehicle active braking system to assist in pre-determining the risk of collision and braking. However, the existing active braking system of the vehicle has the problem of insufficient accuracy in judging whether collision occurs, which is mainly caused by insufficient accuracy in acquiring obstacle information and insufficient accuracy in predicting collision risk.

The method of the embodiment is applied to a vehicle. The active braking refers to an active braking system for automatically performing collision risk judgment on the vehicle, and if the judgment result shows that the collision risk exists, the vehicle automatically performs braking.

The present embodiment acquires various environmental information around the vehicle, thereby acquiring related information of the obstacle. The environmental information refers to physical information including, but not limited to, sound information of the surrounding environment, optical information, etc., wherein the sound information may include ultrasonic information, audible information, etc., and the optical information may include visible light band information, infrared information, ultraviolet information, laser information, etc.

In the embodiment, multiple types of data are acquired for the obstacle, so that the influence of single type data errors on the data accuracy is reduced, and the reliability of the data is improved.

Step S20, identifying and tracking an obstacle in front of the vehicle by using an obstacle identification model according to the surrounding environment information of the vehicle to obtain obstacle information;

in this embodiment, the obstacle recognition model refers to a pre-trained target recognition model, and a target recognition object of the target recognition model is an obstacle, and the target recognition model can recognize and track different types of obstacles. The object recognition model is used for finding out the obstacle by extracting the characteristics in the environment information, and labeling and tracking the obstacle.

The obstacle information refers to a multidimensional parameter combination obtained by processing environment information through an obstacle recognition model, and is a parameter set for describing characteristics of the obstacle in multiple dimensions, and the method comprises the following steps: the location of the obstacle, the relative distance of the obstacle from the vehicle, the relative speed of the obstacle from the vehicle, etc. For example, if a car is positioned 15 ° right ahead (the negative degree represents right ahead and the positive degree represents right ahead), the distance between the obstacle and the vehicle is 200m, the relative speed between the obstacle and the vehicle is-30 km/h (the negative relative speed represents that the obstacle is relatively close to the vehicle, and the positive relative speed represents that the obstacle is relatively far away from the vehicle), the obstacle information of the car may be represented as (15 °, 200' -30).

Step S30, predicting the collision risk of the vehicle and the obstacle based on the obstacle information, and obtaining a collision risk prediction result;

and S40, if the collision risk prediction result meets a brake triggering condition, a brake instruction is sent to a brake control unit, and the brake control unit brakes.

In this embodiment, the collision risk prediction result may be obtained by predicting the collision risk of the vehicle and the obstacle based on the obstacle information through the collision risk prediction model.

The collision risk prediction model refers to a pre-acquired classifier model or regression model, and is trained by utilizing pre-acquired collision data, so that the collision risk is predicted based on the obstacle information, and a collision risk prediction result is obtained.

The collision risk prediction result may be a collision risk level or a collision risk value of a collision of an obstacle with the vehicle. The brake triggering condition can be set according to the collision risk level, and can also be determined by setting a risk threshold.

In one embodiment, the operation of determining whether the collision risk prediction result meets the braking triggering condition may be implemented by a braking decision algorithm. After a collision risk prediction result is obtained through a braking decision algorithm, the collision risk prediction result is compared with a braking triggering condition, and if the collision risk prediction result meets the braking triggering condition, a braking instruction is sent.

As a specific embodiment, the collision risk prediction result may be set as a collision risk level, and then the collision risk level may be classified into three levels of high risk, medium risk and low risk; accordingly, the brake triggering condition may be set such that the triggering brake condition is reached when the collision risk prediction result is a high risk. After the collision risk level is obtained, the braking decision algorithm compares the collision risk level with a triggering braking condition, and if the collision risk level reaches the triggering braking condition with high risk, a braking instruction is sent to the braking control unit.

As another specific embodiment, the collision risk prediction result may be set to a collision risk value, and then the range of the collision risk value result may be set to [0, 100]; accordingly, the brake triggering condition may be set to exceed a certain pre-set collision risk threshold. Such as setting the collision risk threshold to 75. After the collision risk value is obtained, the braking decision algorithm compares the collision risk value with a triggering braking condition, and if the collision risk value is greater than or equal to 75, a braking instruction is sent to the braking control unit.

The brake control unit can be an executing mechanism such as an anti-lock brake system, a linear brake and the like. After the system sends a braking instruction to the braking control unit, an executing mechanism serving as the braking control unit performs braking treatment, generates braking force and slows down the vehicle.

In addition, in actual driving, the forced braking of the vehicle may increase the uncomfortable feeling of the person. Therefore, in addition to the implementation manner of step S40, the present embodiment provides another alternative implementation manner, and if the collision risk prediction result meets the alarm triggering condition, the alarm instruction is sent to the vehicle body controller, and the vehicle body controller performs alarm processing. The setting of the alarm triggering condition may refer to the setting mode of the brake triggering condition, and will not be described in detail herein.

The invention provides an active brake control method. The active brake control method is characterized by collecting environmental information around a vehicle; identifying and tracking an obstacle in front of the vehicle by using an obstacle identification model according to the surrounding environment information of the vehicle to obtain obstacle information; predicting collision risk of the vehicle and the obstacle based on the obstacle information to obtain a collision risk prediction result; and if the collision risk prediction result meets the brake triggering condition, sending a brake instruction to a brake control unit, and braking by the brake control unit. According to the scheme, the surrounding environment information of the vehicle is collected, the obstacle recognition model is utilized to recognize and track the obstacle in the environment information, so that the accuracy of recognizing and tracking the obstacle is improved, the speed of recognizing and tracking the obstacle is accelerated, and the reliability of the obstacle information is improved; the collision risk prediction based on the high-reliability obstacle information is improved in accuracy, and the effect of improving the judgment accuracy of the active brake control device is finally achieved.

Further, based on the first embodiment shown in fig. 2, a second embodiment of the active brake control method of the present invention is provided. In this embodiment, the step S10 includes the steps of:

step S101, collecting ultrasonic reflection information around the vehicle by using an ultrasonic sensor;

step S102, collecting laser reflection information around the vehicle by using a laser radar;

step S103, acquiring video image information around the vehicle by using a vision sensor;

and step S104, the ultrasonic reflection information, the laser reflection information and the video image information are combined into the environment information around the vehicle.

In this embodiment, the vehicle acquires obstacle information using various sensors to acquire the relative relationship of the obstacle and the vehicle for subsequent collision analysis. In the category of collection, three types of information, namely ultrasonic reflection information based on acoustics, laser reflection information based on optics and video image information based on optics, are collected. The acoustic and optical acquisition modes are independent of each other and do not interfere with each other, so that information acquisition of each other is not affected; in the optical-based acquisition mode, the laser wave bands commonly used by the laser radar are 1550nm and 905nm invisible light wave bands, and the visual sensor acquires video image information under the visible light wave bands, so that the acquisition of laser reflection information and the acquisition of video image information cannot interfere with each other. Three different types of obstacle information are acquired, and the acquired obstacle information of the three different sensors can be mutually verified due to different working principles of the different sensors and characteristics of acquired data, so that the reliability of the acquired obstacle data is improved.

The embodiment further adopts the scheme of collecting the obstacle information by adopting three different types of collecting modes, mutually verifies the accuracy of the obstacle information, and achieves the effect of improving the reliability and accuracy of the obstacle data.

Further, based on the first embodiment shown in fig. 2, a third embodiment of the active brake control method of the present invention is provided. In this embodiment, the step S20 identifies and tracks the obstacle in front of the vehicle by using the obstacle identification model according to the environmental information around the vehicle, and the obtaining of the obstacle information includes:

step S201, extracting characteristics in the environment information by using a pre-trained obstacle recognition model, and recognizing an obstacle in the environment information after processing a plurality of neural network layers of the pre-trained obstacle recognition model;

step S202, tracking the obstacle according to an obstacle recognition model to obtain the size of the obstacle, the relative distance between the obstacle and the vehicle and the relative speed;

and step S203, the size of the obstacle, the relative distance between the obstacle and the vehicle and the relative speed are combined into the obstacle information.

In this embodiment, the obstacle recognition model is a pre-trained neural network model, which is used to implement target recognition and tracking.

As a specific example, the obstacle recognition model may be configured as a pre-trained Convolutional Neural Network (CNN). Convolutional neural networks, which are a common model in the field of deep learning, have excellent feature extraction capabilities, and can automatically learn and extract features in video images without manually designing and selecting features. The video image data of the obstacle is used for training the convolutional neural network, and the convolutional neural network can automatically extract the characteristics of the obstacle in the image, so that the obstacle is identified. In addition, the multi-layer perceptron in the network architecture of the convolutional neural network can enable the multi-layer perceptron to extract more abstract and advanced features from the image, and is beneficial to capturing complex modes and structures in the image, so that the convolutional neural network can have good recognition effects on various types of obstacles or obstacles in complex environments.

As another specific example, the obstacle recognition model may be set as a pre-trained YOLO model (You Only Look Once). The YOLO model is a more compact and rapid target detection and tracking algorithm, which performs well in situations where real-time detection is required and accuracy is not required to be too high. YOLO adopts a single-stage detection method, which can directly predict the position and bounding box of a target on an image, and does not need to perform a multi-stage detection flow like CNN. Therefore, YOLO has an advantage in speed, is suitable for detecting obstacles in a scene requiring quick response, and can more quickly identify and track the obstacles. In addition, the YOLO model is relatively simple, has fewer parameters, is easier to realize and deploy, and is suitable for a vehicle-mounted operation platform with relatively limited computing resources.

After the model detects and tracks the obstacle in the environment information, the obstacle information can be obtained by adopting various methods such as image processing and analysis technology, a time difference ranging method and the like, and the calculation can be carried out by adopting the following method: the method for calculating the relative distance is a time difference distance measurement method, wherein the time difference distance measurement method is to transmit ultrasonic waves or laser to a target, the target transmits the ultrasonic waves or laser back, a sensor receives the reflected ultrasonic waves or laser, a timer measures the time from the transmission to the reception of the ultrasonic waves or the laser beam in the whole process, and then the distance from the sensor to the target is calculated according to the propagation speed of the ultrasonic waves or the laser beam. The method for calculating the relative speed is to continuously measure the distance between the target and the sensor by using a time difference ranging method to obtain the variation of the distance between the target and the sensor within a period of time, and calculate the relative speed. The method for obtaining the position of the obstacle is to obtain the distance and the angle between the obstacle and the vehicle by utilizing the point cloud measurement and the characteristic that the point cloud data contains abundant three-dimensional environment information when laser and ultrasonic waves are transmitted to the target surface and return a group of point cloud data, so that the position of the obstacle is obtained.

For video image information, the position of the obstacle in the image can be calculated through image processing and analysis technology, and then the actual position of the obstacle is determined. The actual distance between the obstacle and the vehicle can be calculated and known according to the size of the obstacle in the image, and the relative speed between the obstacle and the vehicle can be calculated and known according to the size change condition of the obstacle in the image within a period of time. After the parameters are obtained, the parameters are collected to form obstacle information, and the obstacle information is output.

Further, in the step S201, the feature in the environmental information is extracted by using a pre-trained obstacle recognition model, and before the obstacle in the environmental information is recognized after the processing of the multi-layer neural network layer of the pre-trained obstacle recognition model, the method further includes:

step S204, an obstacle data set is manufactured by utilizing the environment information, and the obstacle data set is divided into an obstacle training set and an obstacle testing set;

step S205, training the obstacle recognition model by using the obstacle training set to obtain a trained obstacle recognition model;

step S206, testing the trained obstacle recognition model by using an obstacle test set to obtain a first test result of obstacle recognition and tracking;

And step S207, repeating training and testing on the trained obstacle recognition model according to the first test result until a final obstacle recognition model is output.

In this embodiment, the training test is performed on the obstacle recognition model by using the environmental information including the obstacle, and the final obstacle recognition model is obtained after repeated adjustment.

Before training, the environmental information is made into a file with a corresponding format which can be identified by the obstacle identification model, and the file is an obstacle data set. After the obstacle data set is obtained, the obstacle data set needs to be divided into an obstacle training set and an obstacle testing set, and the dividing ratio of the obstacle training set and the obstacle testing set can be determined according to the size of the data set, for example, the obstacle training set is: obstacle test sets were divided into 7:3, the obstacle training set may be used for data sets with larger data set sizes: the proportion of obstacle test sets was further expanded, as with 9:1 so that the model fully extracts the characteristics of various obstacles from massive training data.

In the model test, the obstacle recognition model is utilized to recognize and track the obstacle in the obstacle test set, a test result and an evaluation index are output, the evaluation index can comprise evaluation indexes such as accuracy, precision, recall rate and the like, and the recognition capability of the obstacle recognition model is quantified by utilizing the evaluation index so as to optimize the model.

In the repeated training and testing process of the model, the hyper-parameters of the model are required to be adjusted according to the testing result, the obstacle recognition model is repeatedly trained and tested, the recognition capacity of the model is optimized, and the final obstacle recognition model is output until the evaluation index of the model meets the expected requirement.

Further, in step S207, training and testing the trained obstacle recognition model repeatedly according to the first test result until outputting a final obstacle recognition model includes:

step S2071, a parameter adjustment step, in which parameters of the trained obstacle recognition model are adjusted according to the first test result and aiming at the characteristics of different road conditions and different obstacles;

step S2072, a training test step, in which the adjusted obstacle recognition model is repeatedly trained by using the obstacle training sets with different road conditions and different obstacle types to obtain a new obstacle recognition model, and the new obstacle recognition model is tested by using the obstacle test sets with different road conditions and different obstacle types to obtain a second test result of obstacle recognition and tracking;

step S2073, if the second test result reaches the expected requirement, outputting a final obstacle recognition model; otherwise, repeating the parameter adjustment step and the training test step until the second test result reaches the expected requirement, and outputting a final obstacle recognition model.

In this embodiment, the recognition and tracking capabilities of the obstacle recognition model for different types of obstacles under different road conditions are mainly optimized. The different road conditions refer to various road traffic conditions and environmental conditions, for example, two automobiles A and B are arranged in front of the vehicle, and after the automobile A closest to the vehicle is identified to change the road, the model needs to be capable of quickly identifying and tracking the automobile B without changing the road in front; for another example, in rainy road conditions, the model still needs to have a good recognition effect on the front obstacle. Different obstacle types may include moving and immovable obstacles, for example, for moving obstacles, vehicles may be further subdivided into cars, vans, trucks, SUVs, etc., pedestrians, falling objects, etc. Because different road conditions can cause different influences on obstacle recognition, and each different obstacle has different appearance characteristics, the data set contains as many different road conditions and different types of obstacles as possible during training and testing, so that the generalization capability of the model on the characteristics of the different types of obstacles under different road conditions is improved, and the recognition and tracking effects of the obstacle recognition model on the obstacles are further improved.

The embodiment further achieves the purposes of automatically extracting and learning the image characteristics of the obstacle by utilizing the obstacle recognition model based on the neural network, and achieves good obstacle recognition and tracking effects; the obstacle information is obtained by adopting the obstacle recognition model, so that the reliability of the obstacle information is improved; repeated training and testing are carried out on the model by manufacturing an obstacle data set, so that the model achieves expected obstacle recognition and tracking capability; the model is trained and tested by using the data set containing different road conditions and different types of obstacles, so that the generalization capability of the model is improved, and the recognition and tracking effects of the model on the obstacles are improved.

Further, based on the first embodiment shown in fig. 2, a fourth embodiment of the active brake control method of the present invention is proposed. In this embodiment, the step S30 predicts a collision risk of the vehicle with the obstacle based on the obstacle information, and obtains a collision risk prediction result, including:

step S301, screening abnormal data from the obstacle information to obtain risk prediction data;

step S302, extracting data features in the risk prediction data by using a pre-trained collision risk prediction model, predicting the collision risk of the vehicle and the obstacle by using a classifier of the pre-trained collision risk prediction model, and obtaining a collision risk prediction result.

In this embodiment, the method of screening out abnormal data in the obstacle information may include data preprocessing, screening out abnormal values using a statistical-based method, screening out abnormal values using a machine learning-based method, and the like. The data preprocessing can adopt denoising, filtering, normalization and other operations, so that noise data generated by environmental interference or sensor faults are removed. Statistical-based methods may use a statistical indicator such as Z-score to measure the distance of a data point in a multi-dimensional parameter set from the average of the class of data, and if the value of Z-score exceeds a certain threshold, then the data point may be an outlier, so as to perform the screening. The machine learning-based method can learn the normal behavior of the data in the multidimensional parameter combination by adopting a clustering algorithm or a support vector machine and other models, and then identify the data points which deviate from the normal behavior greatly as abnormal values.

The collision risk prediction model refers to a neural network-based classifier model, which can automatically extract features from data and learn the features, and the collision risk prediction capability is obtained by learning historical vehicle collision data. The collision risk prediction model predicts the collision risk based on the multidimensional data of the obstacle information, and obtains a collision risk prediction result according to the position of the obstacle, the relative distance between the obstacle and the vehicle and the information of the relative speed. The collision risk prediction model is adopted to judge the collision risk, and the method has the advantages of high prediction speed, high prediction accuracy and the like.

Further, in the step S302, the data features in the risk prediction data are extracted by using a pre-trained collision risk prediction model, and before the classifier of the pre-trained collision risk prediction model predicts the collision risk between the vehicle and the obstacle and obtains the collision risk prediction result, the method further includes:

step S303, a risk prediction data set is manufactured by utilizing the pre-collected risk prediction data, and the risk prediction data set is divided into a risk prediction training set and a risk prediction test set;

step S304, training a collision risk prediction model by using a risk prediction training set to obtain a trained collision risk prediction model;

step S305, testing the trained collision risk prediction model by using a risk prediction test set to obtain a third test result of collision risk prediction;

and step S306, repeating training and testing on the trained collision risk prediction model according to the third test result until a final collision risk prediction model is output.

In this embodiment, the risk prediction data is used to perform a training test on the collision risk prediction model, and the final collision risk prediction model is obtained after repeated adjustment. Before training, the corresponding risk prediction data set is also required to be manufactured, and the risk prediction data set is divided into a training set and a testing set according to the size of the data set in a certain proportion. Such as training set of risk prediction: the risk prediction test set is divided into 7:3, the risk prediction training set may be used for data sets of larger data set size: the proportion of risk prediction test sets was further expanded, as with 9:1 so that the model fully extracts the characteristics of various obstacles from massive training data. During testing, the third test result needs to output evaluation indexes including accuracy, precision and the like, and whether the current collision risk prediction accuracy of the model reaches the expected requirement is judged according to the evaluation indexes.

Further, in step S306, according to the third test result, the trained collision risk prediction model is repeatedly trained and tested until a final collision risk prediction model is output, including:

step S3061, a parameter adjustment step, in which parameters of the collision risk prediction model are adjusted according to the third test result and aiming at the characteristics of different road conditions and different obstacles;

step S3062, a training test step, in which the risk prediction training set with different road conditions and different obstacle types is used for repeatedly training the adjusted collision risk prediction model to obtain a new collision risk prediction model, and the risk prediction testing set with different road conditions and different obstacle types is used for testing the new collision risk prediction model to obtain a fourth test result of collision risk prediction;

step S3063, if the fourth test result reaches the expected requirement, outputting a final collision risk prediction model; otherwise, repeating the parameter adjustment step and the training test step until the fourth test result reaches the expected requirement, and outputting a final collision risk prediction model.

In this embodiment, the prediction capability of the collision risk prediction model for collision risk between different types of obstacles and the vehicle under different road conditions is mainly optimized. In this embodiment, different road conditions may affect the braking distance of the vehicle at the same speed, for example, when the vehicle is traveling on a road condition where the road surface is slippery in a rainy day, the required braking distance at the same speed may be greater than the braking distance when the vehicle is traveling on a dry ground, and similarly, when the vehicle is traveling on a road where snow is present, the required braking distance may be greater than the braking distance when the vehicle is traveling on a ground where no snow is present. Therefore, even if the data such as the relative position, the relative distance, the relative speed and the like of the obstacle and the vehicle are the same, the risk of collision under different road conditions is different, so that the model needs to be trained and tested by using multidimensional risk prediction data containing different road conditions, the collision risk prediction capability of the model under different road conditions is improved, and a proper collision risk level is further output. Different obstacle types also have influence on the judgment of collision risk, for example, when the obstacle is a large truck and the obstacle is an automobile, the safety braking distance required by the automobile to avoid collision with the two types of obstacles is also different due to the difference between the running inertia and the braking capability of the large truck and the automobile, so that the model is required to be trained and tested by using multidimensional risk prediction data containing the different types of obstacles, the collision risk prediction capability of the model on the different types of obstacles is improved, and a proper collision risk level is further output. After the training, the effect of improving the collision risk prediction capability of the collision risk level model on various road conditions and various obstacle types can be achieved.

Further, in the step S40, if the collision risk prediction result meets the brake triggering condition, a brake command is sent to a brake control unit, and the brake control unit performs braking, including:

step S401, judging whether the collision risk prediction result exceeds a preset threshold value by utilizing a decision algorithm;

step S402, if the collision risk prediction result exceeds a preset threshold value, a braking instruction is sent;

step S403, the braking instruction is transmitted to the braking control unit through the CAN bus, and the braking control unit executes braking.

In this embodiment, the collision risk prediction result is set to be a collision risk value, the braking triggering condition is set such that the collision risk prediction result exceeds a preset collision risk threshold, and the operation of judging whether the collision risk prediction result meets the braking triggering condition may be implemented by a braking decision algorithm. After the collision risk prediction result is obtained by the braking decision algorithm, the collision risk prediction result is compared with a preset threshold value, and if the collision risk prediction result exceeds the preset threshold value, a braking instruction is sent.

As a specific example, the range of collision risk numerical results may be set to [0, 100]; accordingly, the collision risk threshold is set to 75. After the collision risk value is obtained, the braking decision algorithm compares the collision risk value with a triggering braking condition, and if the collision risk value is greater than or equal to 75, a braking instruction is sent to the braking control unit.

In this embodiment, the brake command is transmitted via the CAN bus. The CAN bus, i.e., the controller area network bus (CAN, controller Area Network), is a serial communication protocol bus for real-time applications, which CAN use twisted pair wires to transmit signals, and is one of the most widely used fieldbuses worldwide. The CAN bus is utilized to transmit the brake command, so that the data transmission speed is high, the brake command CAN be rapidly transmitted to the brake control unit, the rapid response of the brake is realized, and the personal safety and property safety of personnel on the vehicle are ensured. In addition, the CAN bus allows multiple nodes to be connected simultaneously, with a maximum of 110 nodes being connectable. The CAN bus CAN be simultaneously connected with a plurality of brake actuating mechanisms, and brake instructions are transmitted to the brake actuating mechanisms in parallel, so that the transmission efficiency and reliability are improved.

The embodiment further achieves the effects of improving the prediction accuracy of the collision risk and accelerating the prediction speed by screening abnormal data and predicting the collision risk by using a collision risk prediction model; the collision risk prediction model is repeatedly trained and tested by using the risk prediction data, so that the model achieves the expected prediction accuracy; training and testing the model by using a data set containing different road conditions and different types of obstacles, and improving the generalization capability of the model, so that the prediction effect of the collision risk prediction model on the collision risk is improved; by adopting a decision algorithm to analyze the collision risk prediction result, the decision of whether a braking instruction needs to be output or not is realized; by adopting the CAN bus to transmit the brake command, the response speed of the brake is accelerated, and the transmission efficiency and reliability are improved.

an acquisition module 10 for acquiring environmental information around the vehicle;

the identifying module 20 is configured to identify and track an obstacle in front of the vehicle by using an obstacle identifying model according to the environmental information around the vehicle, so as to obtain obstacle information;

a prediction module 30, configured to predict a collision risk of the vehicle and the obstacle based on the obstacle information, and obtain a collision risk prediction result;

and the braking module 40 is configured to send a braking instruction to a braking control unit if the collision risk prediction result meets a braking triggering condition, and the braking control unit performs braking.

The method executed by each program module may refer to each embodiment of the active brake control method of the present application, and will not be described herein.

The embodiment of the application also provides an active brake control device.

The active brake control device comprises a processor, a memory and an active brake control program which is stored in the memory and can run on the processor, wherein the active brake control program realizes the steps of the active brake control method when being executed by the processor.

Because the active brake control program is executed by the processor, all the technical schemes of all the embodiments are adopted, and therefore, the active brake control program at least has all the beneficial effects brought by all the technical schemes of all the embodiments, and the active brake control program is not described in detail herein.

The embodiment of the application also provides a storage medium.

The storage medium of the present invention stores an active brake control program which, when executed by a processor, implements the steps of the active brake control method described above.

It should be noted that, in this document, the terms "comprises," "comprising," or any other variation thereof, are intended to cover a non-exclusive inclusion, such that a process, method, article, or system that comprises a list of elements does not include only those elements but may include other elements not expressly listed or inherent to such process, method, article, or system. Without further limitation, an element defined by the phrase "comprising one … …" does not exclude the presence of other like elements in a process, method, article, or system that comprises the element.

The foregoing embodiment numbers of the present invention are merely for the purpose of description, and do not represent the advantages or disadvantages of the embodiments.

From the above description of the embodiments, it will be clear to those skilled in the art that the above-described embodiment method may be implemented by means of software plus a necessary general hardware platform, but of course may also be implemented by means of hardware, but in many cases the former is a preferred embodiment. Based on such understanding, the technical solution of the present invention may be embodied essentially or in a part contributing to the prior art in the form of a software product stored in a storage medium (e.g. ROM/RAM, magnetic disk, optical disk) as described above, comprising instructions for causing a terminal device (which may be a mobile phone, a computer, a server, or a network device, etc.) to perform the method according to the embodiments of the present invention.

The foregoing description is only of the preferred embodiments of the present invention, and is not intended to limit the scope of the invention, but rather is intended to cover any equivalents of the structures or equivalent processes disclosed herein or in the alternative, which may be employed directly or indirectly in other related arts.

Claims

1. An active brake control method is characterized by comprising the following steps:

collecting environmental information around a vehicle;

2. The active brake control method according to claim 1, wherein the step of identifying and tracking an obstacle in front of the vehicle using an obstacle identification model based on the environmental information around the vehicle, and obtaining the obstacle information includes:

The size of the obstacle, the relative distance of the obstacle from the vehicle, and the relative speed are taken as the obstacle information.

3. The active brake control method according to claim 2, wherein the step of using a pre-trained obstacle recognition model to extract the features in the environmental information, after the step of identifying the obstacle in the environmental information by processing the multi-layer neural network layer of the pre-trained obstacle recognition model, further comprises:

4. The active brake control method as claimed in claim 3, wherein the step of repeating training and testing the trained obstacle recognition model according to the first test result until a final obstacle recognition model is output comprises:

5. The active brake control method according to claim 1, wherein the step of predicting a collision risk of the vehicle with the obstacle based on the obstacle information, the collision risk prediction result including:

6. The active brake control method according to claim 5, wherein the step of extracting data features in the risk prediction data by using a pre-trained collision risk prediction model, predicting collision risk of the vehicle with the obstacle by using a classifier of the pre-trained collision risk prediction model, and obtaining a collision risk prediction result further comprises:

7. The method of active brake control according to claim 6, wherein the step of repeatedly training and testing the trained collision risk prediction model according to the third test result until a final collision risk prediction model is output comprises:

8. The active brake control method according to any one of claims 1 to 7, wherein the step of transmitting a brake instruction to a brake control unit if the collision risk prediction result satisfies a brake triggering condition, the brake control unit performing braking includes:

and if the collision risk prediction result exceeds a preset threshold value, a braking instruction is sent, the braking instruction is transmitted to a braking control unit through a CAN bus, and the braking control unit executes braking.

9. An active brake control apparatus, the apparatus comprising: memory, a processor and an active brake control program stored on the memory and executable on the processor, the active brake control program being configured to implement the steps of the active brake control method as claimed in any one of claims 1 to 8.

10. A computer readable storage medium, wherein an active brake control program is stored on the computer readable storage medium, which when executed by a processor, implements the steps of the active brake control method according to any one of claims 1 to 8.