A convolutional neural network-based car parking space recognition system and method

Abstract

The application relates to a kind of automobile parking space recognition system and method based on convolutional neural network, especially to the field of automobile technology, including acquisition module, to the picture of vehicle periphery is collected in real time by camera;Acquisition module is used to acquire the picture of vehicle periphery collected in real time;Identification module is used to identify the parking space state of vehicle periphery;Push module is used to push the parking state of target parking space to automobile central screen, and interact with user through automobile central screen;Control module is used to control vehicle automatic parking to target parking space;Feedback module, the specified parking space is fed back to control module and push module, until the parking state of again specified parking space meets the automatic parking demand again.The application provides an automobile parking space recognition system and method based on convolutional neural network, and the accuracy of automobile parking space recognition and the safety of automatic parking are improved by recognizing parking space through convolutional neural network.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of automobiles, in particular to a vehicle parking space recognition system and method based on a convolutional neural network. BACKGROUND

[0002] With the continuous improvement of urbanization level and the rapid growth of economic level, people's demand for automobiles is increasingly booming, and the number of automobiles in possession is continuously increasing. Consequently, there are problems such as traffic congestion, shortage of urban parking space resources, and small parking space. Parking in such an environment is easy to cause local traffic congestion and scratching accidents. Parking in a large parking lot is easy to cause difficulties in finding a parking space and finding a car, which brings trouble to drivers. On the other hand, due to the "visual blind area" of vehicles and people caused by their own conditions, parking often consumes a lot of time and effort, which has always been a driving pain point for new and old drivers. Automatic parking is of great significance to the safety of people and vehicles and user experience. Intelligent and automatic recognition of parking spaces is the most important thing to bring good experience to users.

[0003] Chinese Patent Publication No. CN110689761A discloses an automatic parking method, which includes acquiring the position of an empty parking space in a parking area, and controlling a vehicle to park in the parking space according to a calculated parking path. In the method, a plurality of cameras located in the parking area are used to acquire parking space images of the parking area, and whether the parking space is occupied is determined by recognizing the parking space images, so as to acquire the distribution of parking spaces in the parking lot. Then, the empty parking spaces are sent to a vehicle that needs to automatically park in the parking area. After the vehicle parks in the parking space according to the calculated path, the camera is used again to recognize and confirm the parking space mark and license plate number. In the scheme, a neural network model is not used to recognize the parking space, and the parking space cannot be accurately recognized. SUMMARY

[0004] Therefore, the present application provides a vehicle parking space recognition system and method based on a convolutional neural network, to overcome the problems of low recognition efficiency and low safety of automatic parking in the prior art.

[0005] To achieve the above-mentioned purpose, on one hand, the present application provides a vehicle parking space recognition system based on a convolutional neural network, which includes,

[0006] The acquisition module is used to acquire the real-time image of the surrounding of the vehicle through the camera;

[0007] The acquisition module is used to acquire the real-time image of the surrounding of the vehicle through the camera;

[0008] The recognition module is configured to train a convolutional neural network according to the vehicle surrounding pictures in the database, generate a convolutional neural network model, and input the real-time acquired vehicle surrounding pictures into the convolutional neural network model to recognize the parking space state of the vehicle surrounding, obtain a parking space, and select the closest parking space to the vehicle as a target parking space.

[0009] The pushing module is configured to determine the parking state of the target parking space and push the parking state of the target parking space to the car central screen to interact with the user through the car central screen.

[0010] The control module is configured to control the vehicle to automatically park into the target parking space when the parking state of the target parking space is in automatic parking.

[0011] The feedback module is configured to determine the parking state of the specified parking space interacted by the user, and feed back the specified parking space to the control module when the parking state of the specified parking space is in automatic parking, and control the vehicle to automatically park into the specified parking space, and feed back the specified parking space to the pushing module when the parking state of the specified parking space is not in automatic parking, and send the parking state of the specified parking space to the car central screen to continuously interact with the user until the parking state of the specified parking space meets the automatic parking requirement again.

[0012] Further, the recognition module comprises:

[0013] The gray processing unit is configured to perform gray processing on the vehicle surrounding pictures in the database.

[0014] The data expansion unit is configured to perform data expansion on the gray-processed vehicle surrounding pictures to obtain expanded images.

[0015] The training unit is configured to input the expanded images into the convolutional neural network to train the convolutional neural network to obtain a trained convolutional neural network model.

[0016] The recognition unit is configured to input the real-time acquired vehicle surrounding pictures into the trained convolutional neural network model to perform image segmentation, generate a plurality of comparison areas after segmentation, obtain gray values of the comparison areas, compare the gray values of the comparison areas with preset gray values, and perform parking space recognition on the vehicle surrounding pictures according to the comparison results.

[0017] The selection unit is configured to select a target parking space from the parking spaces.

[0018] Further, the gray processing unit has a gray calculation mode when performing gray processing on the vehicle surrounding pictures, as follows.

[0019]

[0020] Wherein, Gray is the gray value, R is the value of R channel, B is the value of B channel, G is the value of G channel, the graying refers to the process of unifying the values of R channel, B channel and G channel to obtain the gray value.

[0021] Further, the recognition unit inputs the real-time acquired vehicle surrounding picture into the trained convolutional neural network model, performs image segmentation, acquires the gray value Fj of each segmented comparison area, j is an integer and j≥1, compares the gray value Fj of each segmented comparison area with the preset gray value F0, and performs parking space recognition according to the comparison result, wherein,

[0022] When Fj<F0, the recognition unit determines that the comparison area is a parking space, and the parking state of the area is parkable;

[0023] When Fj≥F0, the recognition unit determines that the comparison area is a non-parking space, and the parking state of the area is unparkable.

[0024] Further, the selection unit acquires the distance L i between each parking space and the rear of the vehicle through the radar, i≥1 and i is an integer, and the selection unit selects the parking space with the smallest distance between the parking space and the rear of the vehicle as the target parking space.

[0025] Further, the pushing module identifies the obstacles in the driving route of the target parking space through radar scanning, and judges the state of the driving route of the target parking space according to the obstacle identification result, wherein,

[0026] When there are obstacles in the driving route of the target parking space, the pushing module determines that the driving route of the target parking space is in an abnormal state;

[0027] When there are no obstacles in the driving route of the target parking space, the pushing module determines that the driving route of the target parking space is in a normal state.

[0028] Further, when the state of the driving route is in a normal state, the pushing module acquires the width D of the target parking space through the radar, compares the width D of the target parking space with the preset width D0, and judges the parking state of the target parking space according to the comparison result, wherein,

[0029] When D≥D0, the pushing module determines that the parking state of the target parking space is automatic parking;

[0030] When D<D0, the pushing module determines that the parking state of the target parking space is unable to automatically park;

[0031] The preset width D0 is set as D1+r, D1 is a body width of the vehicle, and r is a parking safety width of the vehicle, which refers to a safety distance for passengers to get on and off the vehicle and to maintain non-interference with other vehicles after parking.

[0032] Further, after determining the parking state of the target parking space, the pushing module pushes the parking state of the target parking space to the car central control screen, when the parking state of the target parking space is automatic parking, the pushing module pushes the vehicle surrounding picture corresponding to the target parking space to the car central control screen, when the parking state of the target parking space is not automatic parking, the pushing module obtains other parking spaces and pushes them to the car central control screen, interacts with the user, and obtains the operation instruction of the user on the central control screen to obtain the specified parking space.

[0033] Further, the feedback module identifies obstacles in the driving route of the specified parking space through radar, and judges the state of the driving route of the specified parking space according to the obstacle identification result, wherein,

[0034] When there are obstacles in the driving route of the specified parking space, the feedback module determines that the state of the driving route of the specified parking space is an abnormal state;

[0035] When there are no obstacles in the driving route of the specified parking space, the feedback module determines that the state of the driving route of the specified parking space is a normal state, the feedback module judges the parking state of the specified parking space, and when the parking state is not automatic parking, the feedback module pushes the parking state of the specified parking space to the car central control screen through the pushing module, and continuously interacts with the user until the parking state of the specified parking space meets the automatic parking requirement again.

[0036] On the other hand, the application provides a method for identifying a parking space of a vehicle based on a convolutional neural network, comprising,

[0037] Step S1, collecting a vehicle surrounding picture in real time;

[0038] Step S2, obtaining the vehicle surrounding picture collected in real time and storing it in a database;

[0039] Step S3, generating a convolutional neural network model according to the obtained vehicle surrounding picture, inputting the vehicle surrounding picture collected in real time into the convolutional neural network model, obtaining a parking space, and selecting the closest parking space to the vehicle as a target parking space;

[0040] Step S4, determining the parking state of the target parking space and pushing the parking state of the target parking space to the car central control screen to interact with the user through the car central control screen;

[0041] Step S5, when the parking state of the target parking space is in automatic parking, the vehicle is controlled to automatically park into the target parking space;

[0042] Step S6, the parking state of the specified parking space interacted by the user is judged, and when the parking state of the specified parking space is automatic parking, the specified parking space is fed back, and the vehicle is controlled to automatically park into the specified parking space, when the parking state of the specified parking space is not parking, the specified parking space is fed back to the pushing module, and the parking state of the specified parking space is sent to the car central screen, and the user continues to interact, until the parking state of the specified parking space meets the automatic parking demand again.

[0043] Compared with the prior art, the system has the beneficial effects that the acquisition module acquires the vehicle surrounding picture in real time to identify the parking space according to the actual situation, thereby improving the identification accuracy of the parking space and the safety of automatic parking, after acquiring the vehicle surrounding picture, the acquisition module acquires the vehicle surrounding picture through the intelligent communication controller of the car, and stores the acquired parking lot picture in the database, thereby transmitting the vehicle surrounding situation to the system of the car, improving the identification accuracy of the parking space and the safety of automatic parking, the identification module obtains the trained convolutional neural network model through training the convolutional neural network according to the parking lot picture acquired by the intelligent controller, thereby training through deep learning to realize more accurate image segmentation, improving the identification accuracy of the parking space and the safety of automatic parking, the identification module inputs the real-time acquired vehicle surrounding picture into the trained convolutional neural network model to identify the car parking space, thereby accurately obtaining the target parking space, improving the identification accuracy of the parking space and the safety of automatic parking, the pushing module judges the parking state of the target parking space and pushes the parking state of the target parking space to the car central screen, and interacts with the user through the car central screen, thereby visualizing the parking space, further improving the identification accuracy of the parking space and the safety of automatic parking, the control module controls the vehicle to automatically park into the target parking space when the parking state of the target parking space is automatic parking, to realize automatic parking and improve the identification accuracy of the parking space and the safety of automatic parking, the feedback module judges the parking state of the specified parking space interacted by the user, and when the parking state of the specified parking space is automatic parking, the specified parking space is fed back to the control module, when the parking state of the specified parking space is not parking, the feedback module feeds back the specified parking space to the pushing module, the pushing module sends the parking state of the specified parking space to the car central screen, and continuously interacts with the user, until the parking state of the specified parking space meets the automatic parking demand again, to intelligently acquire the parking space and adjust according to the actual situation, further improving the identification accuracy of the parking space and the safety of automatic parking.

[0044] Especially, the gray unit performs initial gray processing on the vehicle surrounding picture in the database through a gray algorithm, which is more complex than a general gray algorithm and more accurately determines the gray value, thereby improving the recognition accuracy of the parking space and the safety of automatic parking.

[0045] Especially, the selection unit selects the closest parking space as the target parking space to adapt to the driving condition of the vehicle, thereby automatically parking according to the actual situation and further improving the recognition accuracy of the parking space and the safety of automatic parking.

[0046] Especially, the pushing module determines the parking state of the target parking space according to the road condition and the target parking space width, to ensure safety in the actual driving process and avoid danger during automatic parking, thereby further improving the recognition accuracy of the parking space and the safety of automatic parking.

[0047] Especially, when the parking state of the target parking space is unable to automatically park, the pushing module acquires a parking space other than the target parking space and feeds back to the central control screen, acquires the operation of the user on the central control screen, and obtains a specified parking space, so that the user can independently select a parking space, realize the interaction process with the user, thereby further improving the recognition accuracy of the parking space and the safety of automatic parking.

[0048] Especially, the feedback module determines the state of the driving route of the specified parking space, and when the parking state of the specified parking space is unable to automatically park, feeds back the specified parking space to the pushing module, and the pushing module sends the parking state of the specified parking space to the central control screen of the car, and continuously interacts with the user until the parking state of the specified parking space again meets the automatic parking requirement, to automatically park according to the actual situation, thereby improving the recognition accuracy of the parking space and the safety of automatic parking. BRIEF DESCRIPTION OF DRAWINGS

[0049] Figure 1 FIG. 1 is a structural schematic diagram of a convolutional neural network-based automobile parking space recognition system according to the present embodiment;

[0050] Figure 2 FIG. 2 is a structural schematic diagram of a recognition module in the convolutional neural network-based automobile parking space recognition system according to the present embodiment;

[0051] Figure 3 FIG. 3 is a flow schematic diagram of a convolutional neural network-based automobile parking space recognition method according to the present embodiment. DETAILED DESCRIPTION

[0052] In order to make the objects, technical schemes and advantages of the present application clearer, the following further describes the present application with reference to the embodiments. It should be understood that the specific embodiments described herein are only used to explain the present application and should not be used to limit the present application.

[0053] The preferred embodiments of the present application are described below with reference to the drawings. Those skilled in the art should understand that these embodiments are only used to explain the technical principles of the present application and are not used to limit the protection scope of the present application.

[0054] In addition, it should be further explained that, in the description of the present application, unless otherwise explicitly specified and limited, the terms "mounting", "connecting", "connecting" should be understood in a broad sense, for example, it can be fixedly connected, or it can be detachably connected, or integrally connected; it can be mechanically connected, or it can be electrically connected; it can be directly connected, or it can be indirectly connected through an intermediate medium, or it can be the communication inside two elements. Those skilled in the art can understand the specific meaning of the above terms in the present application according to the specific circumstances.

[0055] Please refer to Figure 1 As shown in FIG. 1, which is a structural schematic diagram of a parking space recognition system for a vehicle based on a convolutional neural network according to the present embodiment, the system comprises,

[0056] The acquisition module is used to acquire the surrounding picture of the vehicle in real time through the camera;

[0057] The acquisition module is used to acquire the surrounding picture of the vehicle in real time through the camera;

[0058] The recognition module is used to train the convolutional neural network according to the surrounding picture of the vehicle in the database, generate a convolutional neural network model, and input the real-time acquired surrounding picture of the vehicle into the convolutional neural network model to identify the parking space state of the surrounding of the vehicle, obtain a parking space, and select the closest parking space to the vehicle as the target parking space. The recognition module is connected with the acquisition module;

[0059] The push module is used to determine the parking state of the target parking space and push the parking state of the target parking space to the car central screen to interact with the user through the car central screen. The push module is connected with the recognition module.

[0060] The control module is used to control the vehicle to automatically park into the target parking space when the parking state of the target parking space is in automatic parking. The control module is connected with the push module.

[0061] The feedback module is used for judging the parking state of the specified parking space of the user interaction, and is also used for feeding back the specified parking space to the control module when the parking state of the specified parking space is automatic parking, the control module controls the vehicle to automatically park into the specified parking space, when the parking state of the specified parking space is not parking, the feedback module feeds back the specified parking space to the push module, the push module sends the parking state of the specified parking space to the car central screen, and continuously interacts with the user until the parking state of the specified parking space meets the automatic parking demand again, which is connected with the control module;

[0062] Please refer to Figure 2 The recognition module includes a grayscale unit, a data expansion unit, a training unit and a recognition unit.

[0063] The grayscale unit is used for grayscale processing the vehicle surrounding picture in the database.

[0064] The data expansion unit is used for data expansion of the vehicle surrounding picture after grayscale processing, to obtain an image after data expansion, which is connected with the grayscale unit.

[0065] The training unit is used for inputting the image after data expansion into the convolutional neural network to train the convolutional neural network, to obtain a trained convolutional neural network model.

[0066] The recognition unit is used for inputting the real-time acquired vehicle surrounding picture into the trained convolutional neural network model for image segmentation, to generate a plurality of comparison areas after segmentation, and to acquire the grayscale value of each comparison area, compare the grayscale value of each comparison area with a preset grayscale value, and perform parking space recognition on the vehicle surrounding picture according to the comparison result, the comparison area refers to the image of each area after image segmentation, and the recognition unit is connected with the training unit.

[0067] The selection unit is used for selecting a target parking space from the parking spaces, which is connected with the recognition unit.

[0068] Specifically, the system collects the vehicle surrounding picture in real time through the acquisition module to identify the parking space according to the actual situation, thereby improving the identification accuracy of the parking space and the safety of automatic parking. After the vehicle surrounding picture is collected, the acquisition module acquires the vehicle surrounding picture through the intelligent communication controller of the automobile and stores the acquired parking lot picture in the database, thereby transmitting the vehicle surrounding situation to the system of the automobile, improving the identification accuracy of the parking space and the safety of automatic parking. The identification module acquires the parking lot picture through the intelligent controller, obtains the trained convolutional neural network model through the training of the convolutional neural network, thereby realizing more accurate image segmentation through deep learning training, improving the identification accuracy of the parking space and the safety of automatic parking. The identification module inputs the real-time acquired vehicle surrounding picture into the trained convolutional neural network model to identify the parking space of the automobile, thereby accurately obtaining the target parking space, improving the identification accuracy of the parking space and the safety of automatic parking. The pushing module determines the parking state of the target parking space and pushes the parking state of the target parking space to the automobile center screen, interacts with the user through the automobile center screen, thereby visualizing the parking space, further improving the identification accuracy of the parking space and the safety of automatic parking. When the parking state of the target parking space is for automatic parking, the control module controls the vehicle to automatically park in the target parking space to realize automatic parking and improve the identification accuracy of the parking space and the safety of automatic parking. The feedback module determines the parking state of the specified parking space interacted by the user and feeds back the specified parking space to the control module when the parking state of the specified parking space is that the vehicle can automatically park. When the parking state of the specified parking space is that the vehicle cannot park, the feedback module feeds back the specified parking space to the pushing module. The pushing module sends the parking state of the specified parking space to the automobile center screen and continuously interacts with the user until the parking state of the specified parking space meets the demand of automatic parking, thereby intelligently acquiring the parking space and adjusting according to the actual situation, further improving the identification accuracy of the parking space and the safety of automatic parking.

[0069] Specifically, the setting of the automobile camera is not specifically limited in the embodiment, and can be freely set by those skilled in the art, as long as the vehicle surrounding situation is surveyed, such as setting the automobile camera at the four corners of the automobile or using the automobile 360-degree view to collect images.

[0070] Specifically, the gray processing unit has a gray calculation mode when processing the vehicle surrounding picture, such as the following,

[0071]

[0072] Wherein, Gray is the gray value, R is the value of R channel, B is the value of B channel, G is the value of G channel, the graying refers to the process of unifying the values of R channel, B channel and G channel to obtain the gray value.

[0073] Specifically, the graying unit performs initial graying on the vehicle surrounding picture in the database through a gray algorithm to preliminarily process the image, the gray algorithm is more complex than a general gray algorithm, and the gray value is more accurately determined, thereby improving the recognition accuracy of the parking space and the safety of automatic parking.

[0074] Specifically, the data expansion unit of the embodiment is not specifically limited in the data expansion manner, and a person skilled in the art can freely set it, as long as the requirement of obtaining more learning input of the convolutional neural network is met, such as setting the picture flipping and rotation as the data expansion manner.

[0075] Specifically, in the training of the convolutional neural network model in this embodiment, the convolutional neural network is provided with a convolution module, a pooling layer and a residual module, the convolution module is set to use a 3*3 convolution kernel, the step length of the convolution kernel is 1, the padding strategy is vaild, the activation function used is a relu activation function, in the vaild strategy, the size of the picture is reduced by 2 each time the convolution is performed, the padding refers to a processing process of operating from a new boundary, including same and vaild, the same is padding, the vaild is not padding, and the output is directly calculated, the relu activation function is a commonly used activation function in artificial neural networks, usually refers to a nonlinear function represented by a ramp function and its variants, the pooling layer is set to use a 2*2 maximum pooling, the padding strategy is vaild, the size of the picture is changed to 1 / 2 of the original size through the pooling operation of the pooling layer, the residual module is set to consist of two residual blocks, in the residual module, the input parameter x is added to the output result to avoid gradient disappearance, the output result refers to the result obtained after the residual block convolution network convolution in the residual block, the cross-entropy loss function is set as the loss function, and the loss value is calculated through the cross-entropy function, the loss value L is compared with the preset loss value L0, according to the comparison result, it is determined whether the convolutional neural network calculation process needs to be compensated, wherein when L>L0, the compensation unit determines that the convolutional neural network calculation process needs to be compensated, when L≤L0, the compensation unit determines that the convolutional neural network calculation process does not need to be compensated, when compensation is needed, the residual parameter x is modified to compensate the convolutional neural network calculation process, the modification mode of the residual parameter x is linear increase, and when the loss value in the convolutional neural network calculation is less than or equal to the preset loss value, the trained convolutional neural network model is obtained. It can be understood that those skilled in the art can also set other training methods, and this embodiment is not limited in detail, as long as the convolutional neural network model meets the calculation requirements of the vehicle surrounding picture.

[0076] Specifically, the recognition unit inputs the real-time acquired vehicle surrounding picture into the trained convolutional neural network model, performs image segmentation, acquires the gray value Fj of each comparison area after segmentation, j is an integer and j≥1, compares the gray value Fj of each comparison area after segmentation with the preset gray value F0, and performs parking space recognition according to the comparison result, wherein,

[0077] When Fj<F0, the recognition unit determines that the comparison area is a parking space, and the parking state of the area is parkable;

[0078] When Fj≥F0, the recognition unit determines that the comparison area is a non-parking space, and the parking state of the area is unparkable.

[0079] Specifically, the selecting unit acquires the distance L i between each parking space and the tail of the vehicle through radar, i≥1 and i is an integer, and selects the parking space with the smallest distance between the parking space and the tail of the vehicle as the target parking space.

[0080] Specifically, the selecting unit selects the closest parking space as the target parking space to adapt to the driving condition of the vehicle, thereby automatically parking according to the actual situation and further improving the recognition accuracy of the parking space and the safety of automatic parking.

[0081] Specifically, the pushing module identifies the obstacles in the driving route of the target parking space through radar scanning, and judges the state of the driving route of the target parking space according to the identification result of the obstacles, wherein,

[0082] When the driving route of the target parking space has obstacles, the pushing module determines that the driving route of the target parking space is in an abnormal state.

[0083] When the driving route of the target parking space has no obstacles, the pushing module determines that the driving route of the target parking space is in a normal state.

[0084] Specifically, in the embodiment, when the target parking space driving route is set, the line from the midpoint of the tail of the car to the midpoint of the tail of the target parking space is set as the hypotenuse, a right triangle is drawn according to the hypotenuse, and the route formed by the two right angles of the right triangle is set as the target parking space driving route. It can be understood that those skilled in the art can also set other ways to obtain the target parking space driving route, and the present embodiment is not limited in detail, as long as the planning requirements of the target parking space driving route are met.

[0085] Specifically, the present embodiment does not specifically limit the identification method of the obstacles, and those skilled in the art can freely set it, as long as the identification requirements of the obstacles are met, such as setting an echo positioning sensor to identify the obstacles.

[0086] Specifically, when the state of the driving route is normal, the pushing module acquires the width D of the target parking space through radar, compares the width D of the target parking space with a preset width D0, and judges the parking state of the target parking space according to the comparison result, wherein,

[0087] When D≥D0, the pushing module determines that the parking state of the target parking space is that it can be automatically parked.

[0088] When D<D0, the pushing module determines that the parking state of the target parking space is that it cannot be automatically parked.

[0089] The preset width D0 is set as D1+r, D1 is a body width of the vehicle, and r is a parking safety width of the vehicle, which refers to a safety distance for passengers to get on and off the vehicle and to maintain non-interference with other vehicles after parking.

[0090] Specifically, the pushing module determines the parking state of the target parking space according to the road condition and the target parking space width, so as to ensure safety in actual driving process, avoid danger in automatic parking, and further improve the recognition accuracy of the parking space and the safety of automatic parking.

[0091] Specifically, after determining the parking state of the target parking space, the pushing module pushes the parking state of the target parking space to the car central control screen. When the parking state of the target parking space is that automatic parking is possible, the pushing module pushes the vehicle surrounding picture corresponding to the target parking space to the car central control screen. When the parking state of the target parking space is that automatic parking is impossible, the pushing module acquires other parking spaces and pushes them to the car central control screen, interacts with the user, acquires the operation instruction of the user on the central control screen during the interaction, and obtains the specified parking space.

[0092] Specifically, when the parking state of the target parking space is that automatic parking is impossible, the pushing module acquires the parking spaces other than the target parking space and feeds back to the central control screen, acquires the operation of the user on the central control screen, and obtains the specified parking space, so that the user can autonomously select the parking space, realize the interaction process with the user, and further improve the recognition accuracy of the parking space and the safety of automatic parking.

[0093] Specifically, the interaction refers to pushing the parking spaces other than the target parking space to the car central control screen and acquiring the parking space selected by the user from the parking spaces other than the target parking space. The embodiment does not specifically limit the content and mode of the interaction, and a person skilled in the art can freely set them, as long as the user's requirement for specifying the parking space is met. For example, the content of the interaction is to select all the parking spaces around the vehicle, and a selection window of whether to perform automatic parking is popped up. The mode of the interaction is to push the parking space selection and the vehicle surrounding picture to the central control screen, acquire the operation area of the user on the central control screen, and obtain the specified parking space according to the information corresponding to the operation area.

[0094] Specifically, the feedback module identifies the obstacles in the driving route of the specified parking space through radar and judges the state of the driving route of the specified parking space according to the obstacle identification result, wherein,

[0095] When there are obstacles in the driving route of the specified parking space, the feedback module determines that the state of the driving route of the specified parking space is an abnormal state.

[0096] When there is no obstacle in the designated parking space driving route, the feedback module determines that the state of the designated parking space driving route is a normal state, the feedback module judges the parking state of the designated parking space, and when the parking state is unable to automatically park, the feedback module pushes the parking state of the designated parking space to the car central screen through the pushing module, and continuously interacts with the user until the parking state of the designated parking space again meets the automatic parking requirement.

[0097] Specifically, the feedback module determines the state of the designated parking space driving route, and when the parking state of the designated parking space is unable to automatically park, the feedback module feeds back the designated parking space to the pushing module, and the pushing module sends the parking state of the designated parking space to the car central screen, and continuously interacts with the user until the parking state of the designated parking space again meets the automatic parking requirement, so as to automatically park according to the actual situation, thereby improving the recognition accuracy of the parking space and the safety of automatic parking.

[0098] Specifically, in the embodiment, when the designated parking space driving route is set, the line connecting the midpoint of the tail of the car to the midpoint of the tail of the designated parking space is set as the hypotenuse, and a right triangle is drawn according to the hypotenuse, and the route formed by the two right angles of the right triangle is set as the designated parking space driving route. It can be understood that those skilled in the art can also set other ways to obtain the designated parking space driving route, and the embodiment is not limited specifically, as long as the planning requirement of the designated parking space driving route is met.

[0099] Please refer to Figure 3 The flowchart of the automobile parking space recognition method based on the convolutional neural network of the embodiment, comprising,

[0100] Step S1, real-time collection of vehicle surrounding pictures is performed;

[0101] Step S2, the real-time collected vehicle surrounding pictures are obtained and stored in a database;

[0102] Step S3, a convolutional neural network model is generated according to the obtained vehicle surrounding pictures, and the real-time obtained vehicle surrounding pictures are input into the convolutional neural network model to obtain a parking space, and the closest parking space to the vehicle is selected as a target parking space;

[0103] Step S4, the parking state of the target parking space is determined, and the parking state of the target parking space is pushed to the car central screen, and the user is interacted through the car central screen;

[0104] Step S5, when the parking state of the target parking space is in automatic parking, the vehicle is controlled to automatically park in the target parking space;

[0105] Step S6, judging the parking state of the specified parking space according to the user interaction, and when the parking state of the specified parking space is automatic parking, feeding back the specified parking space and controlling the vehicle to automatically park in the specified parking space, when the parking state of the specified parking space is not automatic parking, feeding back the specified parking space to the pushing module and sending the parking state of the specified parking space to the car central screen to continuously interact with the user until the parking state of the specified parking space meets the automatic parking demand again.

[0106] So far, the technical solutions of the present application have been described in combination with the preferred embodiments shown in the drawings, but those skilled in the art can easily understand that the protection scope of the present application is obviously not limited to these specific embodiments. Those skilled in the art can make equivalent changes or replacements to the related technical features without departing from the principles of the present application, and the technical solutions after these changes or replacements will all fall within the protection scope of the present application.

Claims

1. A convolutional neural network-based automobile parking space recognition system, characterized by, The system comprises, The acquisition module is used for collecting the vehicle surrounding picture in real time through the camera; The acquisition module is used for acquiring the vehicle surrounding picture collected in real time and storing it in the database; The recognition module is used for training the convolutional neural network according to the vehicle surrounding picture in the database, generating a convolutional neural network model, and inputting the vehicle surrounding picture acquired in real time into the convolutional neural network model to identify the parking space state of the vehicle surrounding, obtain a parking space, and select the closest parking space to the vehicle as the target parking space; The push module is used for judging the parking state of the target parking space and pushing the parking state of the target parking space to the car central screen to interact with the user through the car central screen; The control module is used for controlling the vehicle to automatically park into the target parking space when the parking state of the target parking space is automatic parking; The feedback module is used for judging the parking state of the specified parking space interacted by the user, and is also used for feeding back the specified parking space to the control module when the parking state of the specified parking space is automatic parking, and the control module controls the vehicle to automatically park into the specified parking space, and feeding back the specified parking space to the push module when the parking state of the specified parking space is not automatic parking, and the push module pushes the parking state of the specified parking space to the car central screen to continuously interact with the user until the parking state of the specified parking space meets the automatic parking demand again; The recognition module comprises: The gray processing unit is used for performing gray processing on the vehicle surrounding picture in the database; The data expansion unit is used for performing data expansion on the vehicle surrounding picture after the gray processing to obtain an image after the data expansion; The training unit is used for inputting the image after the data expansion into the convolutional neural network to train the convolutional neural network to obtain a trained convolutional neural network model; The recognition unit is used for inputting the vehicle surrounding picture acquired in real time into the trained convolutional neural network model to perform image segmentation, generating a plurality of comparison areas after the segmentation, obtaining the gray value of each comparison area, comparing the gray value of each comparison area with a preset gray value, and identifying the parking space of the vehicle surrounding according to the comparison result; The selection unit is used for selecting the target parking space from the parking space; The gray processing unit has a gray calculation mode when performing the gray processing on the vehicle surrounding picture, as follows, wherein Gray is the gray value, R is the value of the R channel, B is the value of the B channel, and G is the value of the G channel; The recognition unit inputs the vehicle surrounding picture acquired in real time into the trained convolutional neural network model to perform image segmentation, obtains the gray value Fj of each comparison area after the segmentation, compares the gray value Fj of each comparison area after the segmentation with the preset gray value F0, and identifies the parking space according to the comparison result, wherein when Fj < F0, the recognition unit determines that the comparison area is the parking space, the parking state of the area is automatic parking, and the parking space is set as the automatic parking space; When Fj≥F0, the identification unit determines that the comparison area is a non-parking space, and the parking state of the area is unparkable. The selection unit obtains the distance Li between each parking space and the rear of the vehicle by radar, i≥1 and i is an integer, and selects the parking space with the smallest distance between the parking space and the rear of the vehicle as the target parking space. 2.The CNN-based vehicle parking space recognition system according to claim 1, wherein, The pushing module identifies the obstacles in the driving route of the target parking space by radar scanning, and judges the state of the driving route of the target parking space according to the obstacle identification result, wherein, When there are obstacles in the driving route of the target parking space, the pushing module determines that the driving route of the target parking space is in an abnormal state; When there are no obstacles in the driving route of the target parking space, the pushing module determines that the driving route of the target parking space is in a normal state. 3.The CNN-based vehicle parking space recognition system according to claim 2, wherein, When the state of the driving route is in a normal state, the pushing module obtains the width D of the target parking space by radar, compares the width D of the target parking space with the preset width D0, and judges the parking state of the target parking space according to the comparison result, wherein, When D≥D0, the pushing module determines that the parking state of the target parking space is automatic parking; When D<D0, the pushing module determines that the parking state of the target parking space is not automatic parking. 4.The CNN-based vehicle parking space recognition system according to claim 3, wherein, After determining the parking state of the target parking space, the pushing module pushes the parking state of the target parking space to the car central control screen. When the parking state of the target parking space is automatic parking, the pushing module pushes the vehicle surrounding picture corresponding to the target parking space to the car central control screen. When the parking state of the target parking space is not automatic parking, the pushing module obtains other parkable parking spaces and pushes them to the car central control screen. The pushing module interacts with the user and obtains the operation instruction of the user on the central control screen to obtain the specified parking space. 5.The CNN-based vehicle parking space recognition system according to claim 4, wherein, The feedback module identifies the obstacles in the driving route of the specified parking space by radar, and judges the state of the driving route of the specified parking space according to the obstacle identification result, wherein, When there are obstacles in the driving route of the specified parking space, the feedback module determines that the state of the driving route of the specified parking space is abnormal; When there are no obstacles in the driving route of the specified parking space, the feedback module determines that the state of the driving route of the specified parking space is normal, judges the parking state of the specified parking space, and when the parking state is not automatic parking, pushes the parking state of the specified parking space to the car central control screen through the pushing module, and continuously interacts with the user until the parking state of the specified parking space meets the automatic parking requirement.

6. A method applied to the convolutional neural network-based vehicle parking space recognition system according to any one of claims 1-5, characterized in that, The steps include, Step S1, real-time collection of vehicle surrounding pictures; Step S2, obtaining the real-time collected vehicle surrounding pictures and storing them in the database; Step S3, generating a convolutional neural network model according to the obtained vehicle surrounding pictures, inputting the real-time obtained vehicle surrounding pictures into the convolutional neural network model, obtaining parkable parking spaces, and selecting the closest parkable parking space to the vehicle as the target parking space; Step S4, determining the parking state of the target parking space and pushing the parking state of the target parking space to the car central control screen to interact with the user through the car central control screen; Step S5, when the parking state of the target parking space is for automatic parking, controlling the vehicle to automatically park into the target parking space; Step S6, judging the parking state of the specified parking space interacted by the user, and when the parking state of the specified parking space is for automatic parking, feeding back the specified parking space and controlling the vehicle to automatically park into the specified parking space, when the parking state of the specified parking space is not for parking, feeding back the specified parking space to the pushing module and sending the parking state of the specified parking space to the car central control screen to continuously interact with the user until the parking state of the specified parking space meets the automatic parking demand again.

Images