Environment parameter self-learning system, identification method and intelligent auxiliary driving method

Abstract

The present application relates to the technical field of intelligent driving, and provides an environment parameter self-learning system, an identification method and an intelligent auxiliary driving method to solve the technical problem that the prior art relies on offline images for training and needs to rely on manual marking, comprising an image acquisition unit, a vehicle state acquisition unit, a marking unit and a scene library; the image acquisition unit is used for real-time acquisition of vehicle body surface images; the vehicle state acquisition unit is used for real-time acquisition of vehicle states caused by reactions to the current environment by driving behaviors; the marking unit is used for synchronous marking of corresponding vehicle states for the vehicle body surface images; and the scene library is used for matching of corresponding environment parameters for the vehicle body surface images according to a mapping rule of environment parameters and vehicle states. The self-learning system of the present application opens up a brand-new training path, does not need to pre-acquire a large number of offline images, does not need manual marking, can more accurately match driving scenes and driving habits, and realizes personalized auxiliary driving.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of intelligent driving, in particular to an environment parameter self-learning system, an identification method and an intelligent auxiliary driving method. BACKGROUND

[0002] At present, the identification of light rain intensity of a vehicle mainly depends on a light rain sensor, and the control of a vehicle headlight switch and a wiper gear position is designed based on a fixed threshold of light rain intensity, which cannot be optimized later.

[0003] The identification of light rain intensity based on a camera is currently in the research stage and has not been widely applied to vehicles. Figure 1 As shown in the figure, the camera scene recognition in the prior art mainly relies on deep learning of an artificially post-labeled image scene library (image library), but in the image light rain determination, the artificially post-labeled image light rain information is easily inconsistent. SUMMARY

[0004] The present application provides an environment parameter self-learning system, which solves the technical problem of the prior art that relies on offline images for training and needs to rely on artificial marking, and can more accurately match the driving scene and driving habits, laying a foundation for realizing personalized auxiliary control strategies.

[0005] The present application is implemented through the following technical solutions: an environment parameter self-learning system, comprising an image acquisition unit, a vehicle state acquisition unit, a marking unit and a scene library.

[0006] The image acquisition unit is configured to acquire real-time vehicle body surface images.

[0007] The vehicle state acquisition unit is configured to acquire real-time vehicle states caused by reactions of driving behaviors to the current environment.

[0008] The marking unit is configured to mark the corresponding vehicle states for the vehicle body surface images synchronously.

[0009] The scene library is configured to match the corresponding environment parameters for the vehicle body surface images according to the mapping rules of the environment parameters and the vehicle states.

[0010] Further, the image processing unit is used for preliminarily judging the environmental parameter according to the vehicle body outer surface image; and the validity judging unit is used for judging whether the marked vehicle body outer surface image is valid according to whether the reaction of the driving behavior to the environmental parameter is logical, storing the valid vehicle window image to the scene library and intercepting the invalid vehicle body outer surface image.

[0011] Further, the marking unit marks the vehicle window image according to a trigger rule.

[0012] Further, the vehicle body outer surface image is a vehicle window image or a roof image.

[0013] Further, the light intensity level is identified according to the vehicle body outer surface image; the vehicle state includes the vehicle speed and the headlight switch state; the trigger rule is that the vehicle speed is greater than or equal to a set value and the headlight switch state changes; and the validity judging unit judges whether the vehicle body outer surface image is valid according to whether the headlight switch state caused by the driving behavior is logical.

[0014] Further, the rain intensity level is identified according to the vehicle body outer surface image; the vehicle state includes the vehicle speed and the wiper gear position; the trigger rule is that the vehicle speed is greater than or equal to a set value and the wiper gear position changes; and the validity judging unit judges whether the vehicle body outer surface image is valid according to whether the wiper gear position caused by the driving behavior is logical.

[0015] The application further provides an image-based environmental parameter identification method, which collects a current vehicle body outer surface image and sends the image to a scene library in an image-based environmental parameter self-learning system as claimed in claim 1, searches whether there is a similar image of the current vehicle body outer surface image in the scene library, and outputs an environmental parameter matched with the similar image as a current environmental parameter if the similar image meets a threshold.

[0016] Further, the light intensity level and the rain intensity level are identified; and the vehicle state includes the vehicle speed, the headlight switch state and the wiper gear position.

[0017] The application further provides an intelligent auxiliary driving method, which collects a current vehicle body outer surface image and sends the image to a scene library in an image-based environmental parameter self-learning system, searches whether there is a similar image of the current vehicle body outer surface image in the scene library, and automatically controls a vehicle to switch to a corresponding state according to a vehicle state marked on the similar image if the similar image meets a threshold.

[0018] Further, if there is no similar image of the current vehicle body outer surface image, a driver is informed, and a vehicle state caused by a reaction of the driving behavior to a current environment is marked for the current vehicle body outer surface image, and the marked vehicle body outer surface image is stored in the scene library.

[0019] Compared with the prior art, the beneficial effects of the present invention include:

[0020] 1. Existing technologies use a large number of offline images to train artificial neural networks, see reference. Figure 1 As shown, each image requires manual labeling to create training samples, which not only results in high sample acquisition costs and low training efficiency, but also suffers from strong subjectivity, failing to accurately quantify environmental parameters based on the images. However, the self-learning system of this invention opens up a completely new training path, eliminating the need for both pre-collection of large numbers of offline images and manual labeling: this invention associates images with specific vehicle states, and then automatically identifies environmental parameters based on the mapping rules between environmental parameters and vehicle states.

[0021] 2. Current intelligent assisted driving systems rely on the magnitude of environmental parameters and corresponding thresholds for auxiliary control. Inaccurate magnitude judgments can lead to inappropriate auxiliary control strategies. However, the intelligent assisted driving method of this invention does not depend on the accurate identification of environmental parameter magnitudes. Instead, it outputs corresponding control strategies based on image similarity. More specifically, it outputs control strategies based on the vehicle states marked on similar images.

[0022] 3. This invention utilizes the vehicle state caused by the driver's reaction to the current environment. For example, when it rains and the windshield wipers are turned on, the higher the rainfall level, the higher the wiper setting. Different drivers may have different driving behaviors in the same environment. This invention realizes personalized training for a single vehicle. By identifying the vehicle state, it effectively learns the driving habits of different drivers and realizes personalized driving assistance strategies, which is something that existing technologies cannot achieve.

[0023] 4. The system performs self-learning for light and rain recognition in the background of the actual vehicle without being noticed, without affecting the driver's normal use of the vehicle. As the driver continues to use the vehicle, the system will automatically update. Attached Figure Description

[0024] Figure 1 A flowchart for identifying light rain using existing technologies;

[0025] Figure 2 A flowchart for the self-learning process of illumination levels;

[0026] Figure 3 This is a flowchart of the self-learning process for rainfall levels. Detailed Implementation

[0027] An environmental parameter self-learning system comprises an image acquisition unit, a vehicle state acquisition unit, a marking unit and a scene library; the image acquisition unit is used for acquiring a vehicle body surface image in real time; the vehicle state acquisition unit is used for acquiring a vehicle state caused by a reaction of a driving behavior to a current environment in real time; the marking unit is used for marking the vehicle body surface image with a corresponding vehicle state synchronously; and the scene library is used for matching the vehicle body surface image with a corresponding environmental parameter according to a mapping rule of the environmental parameter and the vehicle state.

[0028] In order to exclude error samples, an image processing unit and a validity judging unit are further included; the image processing unit is used for preliminarily judging an environmental parameter according to the vehicle body surface image; and the validity judging unit is used for judging whether the marked vehicle body surface image is valid according to whether a reaction of the driving behavior to the environmental parameter is logical, storing a valid vehicle window image to the scene library and intercepting an invalid vehicle body surface image.

[0029] In order to reduce data storage pressure, the marking unit marks according to a triggering rule, and marks the vehicle window image when the vehicle state satisfies the triggering rule.

[0030] The vehicle body surface image is a vehicle window image, a vehicle body image or a roof image, such as a front windshield image, a rear door image and the like, and a camera is installed at a corresponding position.

[0031] Based on the environmental parameter self-learning system of the present application, an environmental parameter identification method is provided, a current vehicle body surface image is acquired and sent to a scene library, whether a similar image of the current vehicle body surface image exists in the scene library is searched, an image with a similarity satisfying a threshold value is a similar image, and if so, an environmental parameter matched with the similar image is output as a current environmental parameter.

[0032] The present application can be used for identifying light quantity, rainfall or light rainfall, and realizing intelligent auxiliary driving according to the light quantity, the rainfall or the light rainfall, and the following will provide respective real-time examples and make a further detailed description of the present application in combination with the accompanying drawings.

[0033] Embodiment 1

[0034] The environmental parameter self-learning system of the present embodiment is used for identifying an illumination level according to a vehicle body surface image.

[0035] The self-learning process of the present embodiment is referred to Figure 2 as shown in the following figure:

[0036] S1: A camera acquires a vehicle body surface image.

[0037] S2: A vehicle state acquisition unit acquires a vehicle state caused by a reaction of a driving behavior to a current environment in real time, and the vehicle state comprises a vehicle speed and a headlight switch state.

[0038] S3: The marking unit marks according to a triggering rule, and marks the vehicle window image when the vehicle state meets the triggering rule. The triggering rule is that the vehicle speed is greater than or equal to a set value and the headlight switch state changes.

[0039] S4: The image processing unit preliminarily judges the environmental parameter according to the vehicle body outer surface image; specifically, the brightness of the image is used to judge the light intensity, for example, if the brightness is lower than a set threshold, it is considered that the light intensity is lower than a certain value.

[0040] S5: The effectiveness judging unit judges whether the marked vehicle body outer surface image is effective according to whether the reaction of the driving behavior to the environmental parameter is logical, stores the effective vehicle window image to the scene library, and intercepts the invalid vehicle body outer surface image.

[0041] Specifically, the effectiveness judging unit judges whether the vehicle body outer surface image is effective according to whether the headlight switch state caused by the driving behavior is logical. For example, when the vehicle speed is 100 kph, the headlight switch is off, and the preliminarily judged light intensity is lower than 10 lx (the national standard requires less than 1000 lx to automatically turn on the headlight), at this time, the camera may be blocked, causing the recognized image light intensity to be very low, so the image is discarded.

[0042] S6: The scene library matches the corresponding environmental parameter for the vehicle body outer surface image according to the mapping rule of the environmental parameter and the vehicle state.

[0043] For example, the headlight is in the on state corresponding to the low intensity, and the headlight is in the off state corresponding to the high intensity, and the low intensity can be set to be lower than 1000 lx and the high intensity can be set to be 1000 lx or more, so as to output the quantized light intensity.

[0044] When the environmental parameter self-learning system learns to a certain extent, it can be used to realize the intelligent auxiliary driving method, collect the current vehicle body outer surface image and send it to the scene library, search whether there is a similar image of the current vehicle body outer surface image in the scene library, and the image with a similarity degree satisfying a threshold value is a similar image, if so, automatically control the vehicle to switch to the corresponding state according to the vehicle state marked on the similar image.

[0045] For example, the similarity degree is greater than 0.7, and the vehicle state marked on the similar image is that the headlight is turned on, then the headlight is automatically controlled to be turned on.

[0046] If there is no similar image with a similarity degree greater than 0.7, the driver is notified to control the state of the headlight, and the vehicle state caused by the reaction of the driving behavior to the current environment is marked for the current vehicle body outer surface image, and after the marking is completed, it is stored in the scene library. In this way, the scene library can be continuously expanded, and the system can be updated.

[0047] Embodiment 2

[0048] The environmental parameter self-learning system of the embodiment is used to identify the rain grade according to the vehicle body outer surface image.

[0049] The self-learning process of the embodiment refers to Figure 3 as shown:

[0050] S1: The camera collects the vehicle body outer surface image.

[0051] S2: The vehicle state collection unit collects the vehicle state caused by the reaction of the driving behavior to the current environment in real time, and the vehicle state includes the vehicle speed and the wiper gear position, off gear, intermittent gear, gear 1, gear 2, and the like.

[0052] S3: The marking unit marks according to the triggering rule, and marks the window image when the vehicle state meets the triggering rule. The triggering rule is that the vehicle speed is greater than or equal to the set value and the wiper gear position changes.

[0053] S4: The image processing unit preliminarily judges the environmental parameter according to the vehicle body outer surface image; specifically, the rain grade is judged according to the number of raindrops in the image, such as the number of raindrops being lower than the set threshold, and it is considered that the rain grade is lower than a certain value.

[0054] S5: The effectiveness judging unit judges whether the marked vehicle body outer surface image is effective according to whether the reaction of the driving behavior to the environmental parameter is logical, stores the effective window image to the scene library, and intercepts the invalid vehicle body outer surface image.

[0055] Specifically, the effectiveness judging unit judges whether the vehicle body outer surface image is effective according to whether the wiper gear position caused by the driving behavior is logical. For example, in the high-speed state, the wiper switch is closed, there are water droplets on the windshield (where the camera is installed), and it is considered that it is heavy rain at this time, but after judging the whole vehicle state (high speed and wiper closed), it is considered that the camera misjudges at this time, and the image is discarded at this time.

[0056] S6: The scene library matches the corresponding environmental parameter for the vehicle body outer surface image according to the mapping rule of the environmental parameter and the vehicle state.

[0057] For example, the wiper gear off gear corresponds to no rain, the intermittent gear corresponds to light rain, the gear 1 corresponds to moderate rain, and the gear 2 corresponds to heavy rain.

[0058] When the environmental parameter self-learning system learns to a certain extent, it can be used to realize the intelligent auxiliary driving method, collect the current vehicle body outer surface image and send it to the scene library, search whether there is a similar image of the current vehicle body outer surface image in the scene library, and the image with a similarity meeting the threshold is the similar image, if so, automatically control the vehicle to switch to the corresponding state according to the vehicle state marked on the similar image.

[0059] For example, if the similarity is greater than 0.7 and the vehicle state of the similar image is marked as rain, then the wiper gear is automatically controlled to switch to gear 1.

[0060] If there is no similar image with a similarity greater than 0.7, the driver is notified to control the state of the wiper gear, and the vehicle state caused by the reaction of the driver to the current environment is marked for the current vehicle surface image, and the marking is stored in the scene library. In this way, the scene library can be continuously expanded, and the system can be updated.

[0061] Embodiment 3

[0062] The environmental parameter self-learning system of this embodiment is used to identify the light rain level according to the vehicle surface image.

[0063] This embodiment can be understood as a combination of Embodiment 1 and Embodiment 2: the vehicle state includes vehicle speed, headlight switch state, and wiper gear. There are two ways to combine them:

[0064] 1) The vehicle speed, headlight switch state, and wiper gear can be marked on the vehicle surface image at the same time, and then the image similarity is used to comprehensively control the headlights or wipers of the vehicle.

[0065] 2) Copy the vehicle surface image, mark the vehicle speed and headlight switch state on one vehicle surface image, and mark the vehicle speed and wiper gear on another vehicle surface image, so that the headlights and wipers can be controlled independently.

[0066] The above technical solutions are only one embodiment of the present application. For those skilled in the art, based on the principles disclosed in the present application, various types of improvements or modifications can be easily made, and the technical solutions described in the above embodiments are not limited to the above technical solutions. Therefore, the above description is only preferred, and does not have the meaning of limitation.

Claims

1. An environmental parameter self-learning system, characterized by, The system comprises an image acquisition unit, a vehicle state acquisition unit, a marking unit and a scene library, an image processing unit and an effectiveness judging unit. The image acquisition unit is configured to acquire vehicle body surface images in real time. The vehicle state acquisition unit is configured to acquire vehicle states caused by reactions of driving behaviors to current environments in real time. The marking unit is configured to mark corresponding vehicle states for vehicle body surface images synchronously. The scene library is configured to match corresponding environmental parameters for vehicle body surface images according to mapping rules of environmental parameters and vehicle states. The image processing unit is configured to preliminarily judge environmental parameters according to vehicle body surface images, and the effectiveness judging unit is configured to judge whether the marked vehicle body surface images are valid according to whether reactions of driving behaviors to environmental parameters are logical, store valid vehicle window images to the scene library and intercept invalid vehicle body surface images.

2. The environmental parameter self-learning system according to claim 1, characterized in that The marking unit marks according to a triggering rule, and marks vehicle window images when vehicle states satisfy the triggering rule.

3. The environmental parameter self-learning system of claim 1, wherein The vehicle body surface images are vehicle window images, vehicle body images or vehicle roof images.

4. The environmental parameter self-learning system of claim 1, wherein The system is configured to identify illumination levels according to vehicle body surface images, the vehicle states include vehicle speeds and headlight switch states, the triggering rule is that the vehicle speed is greater than or equal to a set value and the headlight switch state changes, and the effectiveness judging unit judges whether the vehicle body surface images are valid according to whether the headlight switch states caused by driving behaviors are logical.

5. The environmental parameter self-learning system according to claim 1 or 4, characterized in that, The system is configured to identify rain levels according to vehicle body surface images, the vehicle states include vehicle speeds and wiper gear positions, the triggering rule is that the vehicle speed is greater than or equal to a set value and the wiper gear position changes, and the effectiveness judging unit judges whether the vehicle body surface images are valid according to whether the wiper gear positions caused by driving behaviors are logical.

6. A method of environmental parameter identification, characterized by: The system is configured to acquire current vehicle body surface images and send them to the scene library in the environmental parameter self-learning system, search whether similar images of the current vehicle body surface images exist in the scene library, and if yes, output environmental parameters matched with the similar images as current environmental parameters.

7. The environmental parameter identification method of claim 6, wherein: The system is configured to identify illumination and rain levels, and the vehicle states include vehicle speeds, headlight switch states and wiper gear positions.

8. An intelligent co-piloting method, characterized in that, The system is configured to acquire current vehicle body surface images and send them to the scene library in the environmental parameter self-learning system, search whether similar images of the current vehicle body surface images exist in the scene library, and if yes, automatically control the vehicle to switch to corresponding states according to vehicle states marked on the similar images. 9.The intelligent assisted driving method of claim 8, wherein, If no similar images of the current vehicle body surface images exist, the system is configured to notify the driver and mark vehicle states caused by reactions of driving behaviors to current environments for the current vehicle body surface images, and store the marked vehicle states to the scene library after the marking is completed.

Images