New energy vehicle charging interface automatic positioning method and charging gun automatic plug positioning method

Abstract

The application discloses a new energy automobile charging interface automatic positioning method, comprising the following steps: S1: when detecting that an automobile is parked in a charging position, an image of a front windshield containing a vehicle frame number external rectangle area of the automobile is collected by using a camera, and a vehicle frame number external rectangle is labeled in the image to obtain a vehicle frame number image; S2: pixel coordinate information of a vehicle frame number mark is obtained according to the vehicle frame number image, and world coordinate information of the vehicle frame number mark is obtained from the pixel coordinate information of the vehicle frame number mark; text information of the vehicle frame number is obtained through vehicle frame number image recognition, the automobile model is determined according to the recognized vehicle frame number text information, a coordinate transformation matrix between the vehicle frame number mark and the charging interface is obtained according to the automobile model, and then the world coordinate information of the charging interface is obtained. The application further discloses a new energy automobile charging gun automatic plug-in positioning method, which can accurately position the charging interface positions of automobiles of different models to realize automatic plug-in of the charging gun.

Description

TECHNICAL FIELD

[0001] The application belongs to the technical field of new energy vehicle charging, and particularly relates to a new energy vehicle charging interface automatic positioning method and a charging gun automatic plug-in positioning method. BACKGROUND

[0002] The vehicle identification code (frame number, VIN code) can be said to be the identity card of the automobile. The frame number of each vehicle is different. The frame number is composed of 17 English and numbers. The production date, manufacturer, vehicle model, vehicle configuration and other related information of the automobile can be found from the frame number. The frame number is unique, and each vehicle has its own frame number. The frame number is generally located at the lower part of the door, the left side of the front windshield and the driver's door column and the like. In many complex scenes, the information of the vehicle can be identified through the frame number.

[0003] In the automobile industry, new energy vehicles include pure electric vehicles and hybrid electric vehicles. Pure electric vehicles do not consume fossil fuels during use and do not emit carbon dioxide. Hybrid electric vehicles can reduce fuel consumption and reduce carbon dioxide emissions. Compared with traditional fuel vehicles, they have the advantages of energy saving and environmental protection. New energy vehicles can store electric energy through charging, and therefore have a charging interface. At present, the charging method of new energy vehicles is generally to charge the new energy vehicles by hand. The degree of automation is low. With the development of intelligent new energy vehicles, intelligent vehicles will realize autonomous and automatic charging, which will become an important function of new energy vehicles to solve the range anxiety and improve the user experience. For example, a new energy vehicle charging interface visual identification positioning method and a charging system are disclosed in Chinese Patent No. CN111179292A. A monocular camera is added to the existing new energy vehicle charging gun to collect images of the new energy vehicle charging interface in real time. The collected real-time images are processed to obtain a binary image of the fixed bolt or marker block of the new energy vehicle charging interface. Then, the obtained binary image is subjected to erosion and inflation processing to remove noise points in the image. The Canny edge detection operator is used to detect the edge pixel points of the fixed bolt or marker block in the image obtained in the previous step, and the edge pixel points are clustered to obtain the point set of each edge pixel point. The center point of each edge pixel point set is calculated to obtain the pixel coordinates of the corresponding center point. Finally, the positions and attitudes of the new energy vehicle charging interface in the camera coordinate system are calculated according to the pixel coordinates of all the center points, and a three-dimensional mechanical arm is used to control the new energy vehicle charging gun to be connected with the new energy vehicle charging interface to realize automatic charging of the new energy vehicle. However, in actual scenarios, the method has the following disadvantages:

[0004] 1) Before the world recognition positioning of the charging interface, the position of the charging interface on the car needs to be found accurately first; however, due to various brands and models of cars, the charging interface positions of different models of cars are different, and in addition, the positions and postures of the cars when parked at the charging position are not unified, therefore, how to accurately find the charging interface of the car of different models and different parking postures becomes an industry difficulty;

[0005] 2) When the fixing bolt of the charging interface of the new energy vehicle is photographed, it is often easily blocked, and the setting of the marker block needs to be manually marked, which is time-consuming and laborious. SUMMARY

[0006] Therefore, the purpose of the present application is to provide a new energy vehicle charging interface automatic positioning method and a charging gun automatic plug-in positioning method, which can accurately position the charging interface position of different models of cars to realize the automatic plug-in of the charging gun.

[0007] To achieve the above purpose, the present application provides the following technical solutions:

[0008] The present application first proposes a new energy vehicle charging interface automatic positioning method, comprising the following steps:

[0009] S1: When a car is detected to be parked in the charging position, an image of the front windshield of the car containing the outer rectangle of the vehicle identification number is collected by using a camera, and an outer rectangle of the vehicle identification number is labeled in the image to obtain a vehicle identification number image;

[0010] S2: The pixel coordinate information of the vehicle identification number is obtained according to the vehicle identification number image, and the world coordinate information of the vehicle identification number is obtained from the pixel coordinate information of the vehicle identification number; the text information of the vehicle identification number is obtained by recognizing the vehicle identification number image, the model of the car is determined according to the recognized text information of the vehicle identification number, the coordinate transformation matrix between the vehicle identification number and the charging interface is obtained according to the model of the car, and then the world coordinate information of the charging interface is obtained.

[0011] Further, in step one, a vehicle identification number recognition model is used to label the outer rectangle of the vehicle identification number in the image.

[0012] Further, the construction method of the vehicle identification number recognition model comprises the following steps:

[0013] 11) Image collection: images of the front windshield of different models of cars containing the outer rectangle of the vehicle identification number are collected from different angles and different heights by using cameras deployed in the charging station to constitute an original image set;

[0014] 12) Preprocessing: The original image set is preprocessed and augmented to obtain an enhanced image set; the enhanced image set is cropped and classified and labeled to mark the bounding rectangle of the vehicle identification number, and the center point coordinates, height, width and angle information of the target box are obtained. Finally, it is converted into a data format that conforms to the training of the vehicle identification number recognition model to obtain the enhanced dataset.

[0015] 13) Create a vehicle identification number (VIN) recognition model network;

[0016] 14) The vehicle identification number (VIN) recognition model was trained using the augmented dataset.

[0017] Furthermore, the vehicle identification number (VIN) recognition model employs an improved YOLOv5 object detection convolutional neural network model. In step 13), the method for improving the YOLOv5 object detection convolutional neural network model is as follows:

[0018] 131) Generate anchors using the K-means++ clustering algorithm;

[0019] 132) Replace the YOLOv5 backbone network with the more lightweight MobileNetv3 network;

[0020] 133) Replace the Neck network structure with a weighted bidirectional feature pyramid network (BiFPN) structure.

[0021] Furthermore, in step 131), the step of generating anchors using the K-means++ clustering algorithm is as follows:

[0022] 1311) Randomly select a sample x0 from dataset X as the initial cluster center c1; k = 1;

[0023] 1312) Calculate x for each sample m Compared with existing cluster centers c i The shortest distance d(x) between m ,c i ):

[0024] d(x m ,c i ) = 1 - IOU(x m ,c i )

[0025] Where IOU represents the degree of overlap between anchors and the bounding box;

[0026] 1313) Calculate the probability that each sample will be selected as the next cluster center using the following formula: probability p m :

[0027]

[0028] 1314) The next cluster center is selected according to the roulette method; k = k + 1;

[0029] 1315) Determine whether k is equal to K: if yes, K cluster centers have been selected, stop iteration; if no, perform step 1312).

[0030] Further, in the 132), the function of replacing the backbone network of YOLOv5 with MobileNetv3 network is implemented as follows:

[0031] 1321) Use 1*1 convolution to upgrade the input feature to exp size, and pass through the nonlinear activation function NL;

[0032] 1322) The output feature is passed through a 3*3 DW convolution, and the nonlinear activation function NL, at this time the channel number expsize does not change;

[0033] 1323) The output feature is passed into the SE block, first global average pooling is performed on each channel to obtain a 1*1*exp size vector; then the vector is passed into the first layer of fully connected layer, and the activation function is ReLU, to obtain a vector with an output of 1*1*(exp size / 4); then the output is passed into the second layer of fully connected layer, and the activation function is h-sigmoid, to obtain a vector with an output of 1*1*exp size; finally, each element in the vector is multiplied by the corresponding channel feature matrix of the feature map input to the SE layer, to obtain a new feature map with the same size as the feature map input to the SE layer. Figure One

[0034] 1324) Finally, 1*1 convolution dimension reduction processing is performed.

[0035] Further, the activation function h-sigmoid is:

[0036]

[0037] Where, ReLU6 represents the activation function.

[0038] Further, in the second step, the coordinate conversion matrix from the pixel coordinate information of the frame number identifier to the world coordinate information of the frame number identifier is:

[0039]

[0040] Where, [u,v] T ​represents the coordinate of the frame number mark in the image pixel coordinate system O-U-V; fx=f / dx, fy=f / dy, dx and dy represent the physical size of each pixel in the u-axis and v-axis directions, f represents the focal length of the camera; Zc represents the coordinate of the frame number mark in the Z-axis of the camera coordinate system; u0 and v0 represent the coordinates of the origin of the image physical coordinate system in the image pixel coordinate system; R represents an orthogonal unit rotation matrix; T represents a three-dimensional translation vector; [Xw, Yw, Zw] represents the coordinate of the frame number mark in the world coordinate system.

[0041] The application further provides an automatic plug-in positioning method for a new energy vehicle charging gun.

[0042] Step one: obtaining the world coordinate information of the charging interface by using the automatic positioning method for the new energy vehicle charging interface; and obtaining the world coordinate information of the charging gun according to the coordinate transformation matrix between the charging gun and the camera.

[0043] Step two: planning a moving path of the charging gun to the charging interface according to the world coordinate information of the charging interface and the world coordinate information of the charging gun.

[0044] Step three: moving the charging gun to the charging interface position according to the planned moving path; and interacting with the vehicle to open the charging port cover.

[0045] Step four: inserting the charging gun into the charging interface.

[0046] Further, in the step four, the position of the charging gun is compensated in real time by using visual identification during the insertion process, so that the charging gun is aligned with the charging interface.

[0047] The application has the following beneficial effects:

[0048] In view of the fact that the sizes and structures of new energy vehicles of different models are inconsistent, and the positions and postures of the vehicles when parking and charging are inconsistent, which leads to the difficulty in positioning the charging interface, the new energy vehicle charging interface automatic positioning method uses the fact that the positions between the rectangular external frame of the vehicle front windshield frame number and the charging interface of the vehicle are relatively fixed, and through the vehicle frame number text information, the model, structure, size and other information of the vehicle can be obtained, so that the coordinate transformation matrix between the frame number identification and the charging interface of the corresponding vehicle model can be obtained; after the vehicle is parked in the charging position, the image of the frame number external rectangular region is obtained by the camera, and the frame number external matrix is labeled in the image, so that the pixel coordinate information of the frame number identification is converted to the world coordinate system, and combined with the coordinate transformation matrix between the frame number identification and the charging interface, the world coordinate information of the charging interface is obtained, and the automatic positioning of the charging interface is realized; that is, the new energy vehicle charging interface automatic positioning method can realize the positioning of the charging interface of the vehicle of any model, any parking position and any parking posture in the charging position, and has the advantages of good reliability and good applicability. BRIEF DESCRIPTION OF DRAWINGS

[0049] In order to make the purpose, technical scheme and beneficial effects of the present application clearer, the present application provides the following drawings for illustration:

[0050] Figure 1 The principle block diagram of the new energy vehicle charging gun automatic plugging positioning method is shown in the figure.

[0051] Figure 2 The structure diagram of the frame number recognition model is shown in the figure. DETAILED DESCRIPTION

[0052] The present application will be further described below in combination with the drawings and specific embodiments, so that those skilled in the art can better understand the present application and implement it, but the embodiments are not limiting to the present application.

[0053] The new energy vehicle charging gun automatic plugging positioning method of the present embodiment includes the following steps:

[0054] Step one: obtaining the world coordinate information of the charging interface by using the new energy vehicle charging interface automatic positioning method; obtaining the world coordinate information of the charging gun according to the coordinate transformation matrix between the charging gun and the camera.

[0055] Specifically, the new energy vehicle charging interface automatic positioning method of the present embodiment includes the following steps:

[0056] S1: when detecting that the vehicle is parked in the charging position, the image of the vehicle front windshield containing the frame number external rectangular region is collected by the camera, and the frame number external rectangle is labeled in the image, to obtain the frame number image.

[0057] In this embodiment, the frame number recognition model is used to label the frame number bounding rectangle in the image. Specifically, the construction method of the frame number recognition model includes the following steps:

[0058] 11) Collect images: use the camera deployed in the charging station to collect images of different models of automobile front windshield containing frame number bounding rectangle region from different angles and different heights to form a set of original images;

[0059] 12) Preprocessing: preprocessing and data augmentation are performed on the original image set to obtain an enhanced image set. Preprocessing can eliminate the influence of noise, lighting, viewing angle, etc. in the image. Data augmentation operation enriches the original image set by flipping, shifting, mirror symmetry, cropping, brightness and contrast enhancement, etc. The enhanced image set is cropped and classified and labeled. The frame number image is labeled using a labeling software to label the bounding rectangle of the frame number, and the center point coordinates, height, width and angle information of the target box are obtained. Finally, it is converted into a data format that meets the training requirements of the frame number recognition model, and an enhanced data set is obtained;

[0060] 13) Create a frame number recognition model network. Specifically, the frame number recognition model of the present embodiment uses an improved YOLOv5 target detection convolutional neural network model to build a frame number recognition model that can quickly and accurately recognize frame numbers. Specifically, the improvement method of the YOLOv5 target detection convolutional neural network model is as follows:

[0061] 131) Use K-means++ clustering algorithm to generate anchors. The traditional YOLOv5 uses K-means clustering algorithm to generate anchors, but K-means algorithm has certain limitations, and the clustering result is easily affected by the selection of initial value. In order to reduce the influence of the selection of initial value, when selecting the cluster center, the way of selecting one by one can be adopted, and K is not selected at one time, that is, the core idea of K-means++ algorithm is obtained.

[0062] K-means++ algorithm belongs to the derivative of K-means algorithm, and its main solution is the first step of K-means algorithm, that is, the problem of randomly selecting the center point. Specifically, the steps of K-means++ clustering algorithm to generate anchors are as follows:

[0063] 1311) Randomly select a sample x0 from the data set X as the initial cluster center c1; k = 1;

[0064] 1312) Calculate the shortest distance d(x m , c i ) between each sample x m and the current cluster center c i):

[0065] d(x m ,c i )=1-IOU(x m ,c i )

[0066] wherein IOU represents the degree of overlap of the anchors and the bounding box;

[0067] 1313) calculating the probability of each sample being selected as the next cluster center with the following formula: probability p m :

[0068]

[0069] 1314) selecting the next cluster center according to the roulette method; k = k + 1;

[0070] 1315) judging whether k is equal to K: if yes, the K cluster centers have been selected, and the iteration is stopped; if no, step 1312) is executed.

[0071] 132) replacing the backbone network of YOLOv5 with a more lightweight MobileNetv3 network. The YOLOv5 backbone feature extraction network adopts a C3 structure, which brings a large parameter quantity and a slow detection speed, and the application is limited. Replacing the backbone feature extraction network with a more lightweight MobileNetv3 network can realize the lightweight of the network model, balance the speed and accuracy, and facilitate the deployment of the model.

[0072] MobileNetv3 compared with the previous two generations of MobileNet network mainly in the update Block, re-design of time-consuming layer structure in two parts of the improvement. In the update Block, first of all, the attention mechanism (SE module) is added, the attention mechanism here is very simple, that is, the pooling processing is carried out for each channel, and the channel number elements are obtained, and the output vector is obtained through two fully connected layers. Among them, the node number of the first fully connected layer is equal to 1 / 4 of the channel number, and the node number of the second fully connected layer is consistent with the channel number, and the output vector obtained is equivalent to analyzing the importance of each channel of the original feature matrix. The more important the weight is given, the less important the smaller weight is given. Secondly, MobileNetv3 redesigns the activation function. Previously in MobileNetv2, ReLU6 activation function was used. Now the commonly used is swish activation function, that is, x*sigmoid activation function. Although using swish activation function can improve the accuracy of the network, the calculation and derivation time complexity is not friendly to the quantization process. Therefore, the embodiment proposes h-swish activation function, which can improve the calculation speed of the network and is friendly to the quantization process by replacing the swish activation function with h-swish and replacing the sigmoid activation function with h-sigmoid activation function. In terms of redesigning the time-consuming layer structure, first of all, the number of first layer convolution kernels is reduced from 32 to 16. Although this does not change the accuracy, the parameter quantity is reduced and the calculation speed is increased. Secondly, the Last Stage is simplified, and the layer structure is reduced. Similarly, the calculation speed can be improved without losing accuracy.

[0073] The expressions of various activation functions are as follows:

[0074] The ReLU6 activation function is:

[0075] ReLU6(x)=min(max(x,0),6)

[0076] The h-sigmoid activation function is:

[0077]

[0078] The h-swish activation function is

[0079]

[0080] The activation function h-swish is applied in the backbone network MobileNetv3, specifically in the first convolutional layer and several subsequent bneck layers. The MobileNetv3 network is divided into three parts: the first part extracts feature layers through a convolutional layer, the second part is a reverse residual structure block, and the third part outputs categories. The steps for replacing the backbone network of YOLOv5 with the MobileNetv3 network are as follows:

[0081] 1321) The input features are upgraded to exp size using a 1*1 convolution, and then passed through a nonlinear activation function NL. In this embodiment, the nonlinear activation function NL includes ReLU and h-swish functions. Here, the nonlinear activation function NL refers to the h-swish function in the first convolutional layer of the MobileNetv3 backbone network.

[0082] 1322) The output features are passed through a 3*3 DW convolution and then through a nonlinear activation function NL. At this time, the channel number exp size does not change.

[0083] 1323) The output features are passed into the SE block (4th, 5th, and 6th bneck). First, each channel is globally averaged to obtain a 1*1*exp size vector. Then, the vector is passed into the first fully connected layer with ReLU as the activation function, resulting in a 1*1*(exp size / 4) vector. Next, the output is passed into the second fully connected layer with h-sigmoid as the activation function, resulting in a 1*1*exp size vector (it can be seen that the channel of the feature matrix does not change after SE). Finally, each element in the vector is multiplied by the corresponding channel feature matrix of the input feature map to the SE layer (11th, 12th, 13th, 14th, and 15th bneck), resulting in a new feature map with the same size as the input feature map. Figure One

[0084] 1324) Finally, a 1*1 convolution is performed for dimension reduction.

[0085] 133) Replace the neck network structure with the weighted bidirectional feature pyramid network BiFPN structure. The YOLOv5 neck network design follows the FPN+PAN structure. In order to improve the detection speed, the neck network structure is replaced with the weighted bidirectional feature pyramid network (BiFPN) structure. Through bidirectional cross-scale connection and weighted feature fusion, more features can be fused without increasing too much cost.

[0086] ​14) Train the VIN recognition model using the augmented dataset. Divide the augmented dataset into a training set and a test set in a 9:1 ratio. Input the sample images from the training set into the VIN recognition model network. Through iterative training, obtain the optimal weight file for VIN target detection and load this weight file into the VIN recognition model network to finally obtain the trained VIN recognition model.

[0087] S2: Obtain the pixel coordinates of the VIN (Vehicle Identification Number) from the VIN image, and then derive the world coordinates of the VIN from these pixel coordinates. Recognize the text information of the VIN from the VIN image, determine the vehicle model based on the recognized text information, and obtain the coordinate transformation matrix between the VIN and the charging interface based on the vehicle model. This matrix, in turn, yields the world coordinates of the charging interface. The coordinate transformation matrix between the VIN and the charging interface is determined based on the vehicle's structure and dimensions. Since the coordinate transformation matrix between the VIN and the charging interface is identical for each vehicle model, it can be pre-stored within the charging control system for different vehicle models.

[0088] Specifically, the method for obtaining the world coordinate information of the vehicle identification number (VIN) from the pixel coordinate information of the VIN is as follows:

[0089] 21) Establish an image pixel coordinate system OUV, where the origin o of the image pixel coordinate system is located at the upper left corner of the frame number bounding rectangle, the u-axis is the width direction of the frame number bounding rectangle, and the v-axis is the height direction of the frame number bounding rectangle.

[0090] 22) Obtain the image pixel coordinates [u,v] of the vehicle identification number (VIN) from the VIN image. T Coordinate transformation can convert the image to its physical coordinate system [x, y]. T Image pixel coordinates describe the position of a point in pixels; the unit is merely a counting unit. Image physical coordinates, on the other hand, describe the position of a point in physical units, representing a point with actual physical meaning. The conversion relationship between the two is as follows:

[0091]

[0092] Where dx and dy represent the physical dimensions of each pixel in the u-axis and v-axis directions, and u0 and v0 represent the coordinates of the origin of the image physical coordinate system in the image pixel coordinate system.

[0093] 23) Transformation between the image physical coordinate system and the camera coordinate system. Let the vehicle identification number (VIN) be located at coordinates [Xc, Yc, Zc] in the camera coordinate system. The transformation relationship between the two is as follows:

[0094]

[0095] where f represents the focal length of the camera. Therefore, the conversion from the image pixel coordinate system to the camera coordinate system is:

[0096]

[0097] where the parameters fx, fy, u0 and v0 are called the intrinsic parameters of the camera, and the matrix K is called the intrinsic matrix of the camera. The intrinsic parameters of the camera are obtained by Zhang Zhengyou's calibration algorithm, and then the matrix K is obtained.

[0098] 24) Conversion between the camera coordinate system and the world coordinate system. This conversion is based on the rigid body transformation of the object, which only changes the spatial position (translation) and orientation (rotation) of the object, but does not change its shape. Let the coordinates of the frame number identification in the world coordinate system be [Xw, Yw, Zw], and the conversion relationship is as follows:

[0099]

[0100] where R represents an orthogonal unit rotation matrix, and T represents a three-dimensional translation vector. The translation vector is the translation amount along the axis direction. For the rotation matrix, assuming that the coordinates of point P before rotation are [X, Y, Z], and O-X-Y rotates around the Z axis by an angle of θ, let the coordinates of point P after rotation be P' = [X', Y', Z'], then the conversion relationship between the two is:

[0101]

[0102] Similarly, the transformation matrices r1, r2 for rotation around the X axis and the Y axis can be obtained:

[0103]

[0104] The rotation matrix R is the product of the three degrees of freedom rotation transformation matrices:

[0105] R = r1 x r2 x r3

[0106] Finally, the coordinate conversion matrix from the pixel coordinate information of the frame number identification to the world coordinate information of the frame number identification is obtained as:

[0107]

[0108] That is:

[0109]

[0110] where the matrix [R T] is called the extrinsic matrix of the camera, and the Perspective-n-Points (PnP) method can be used to solve the extrinsic matrix of the camera.

[0111] The coordinate transformation matrix between the charging gun and the camera is determined by the structure of the charging device and the relative position between the charging gun and the camera, and will not be repeated. Through the coordinate transformation matrix between the charging gun and the camera, the world coordinate information of the charging gun can be obtained.

[0112] Step two: According to the world coordinate information of the charging interface and the world coordinate information of the charging gun, the moving path of the charging gun to the charging interface is planned. When planning the moving path, the structure of the charging position, the size information of the car, the parking position and attitude, etc. should be combined to avoid collision between the charging pile and the structure in the charging position and the car.

[0113] Step three: Move the charging gun to the charging interface position according to the planned moving path; at the same time, open the charging port cover through Bluetooth interaction between the charging gun and the car.

[0114] Step four: Insert the charging gun into the charging interface, and compensate the position of the charging gun in real time by using visual identification during the insertion process, so as to align the charging interface.

[0115] The above-mentioned embodiments are only preferred embodiments for fully illustrating the present application, and the protection scope of the present application is not limited thereto. The equivalent substitutions or transformations made by the skilled in the art on the basis of the present application are within the protection scope of the present application. The protection scope of the present application is subject to the claims.

Claims

1. A new energy vehicle charging interface automatic positioning method, characterized in that: Comprising the following steps: S1: After detecting the parking of the car in the charging position, an image of the front windshield containing the outer rectangle of the vehicle identification number is collected by the camera, and the outer rectangle of the vehicle identification number is labeled in the image to obtain the vehicle identification number image; S2: Obtain the pixel coordinate information of the vehicle identification number from the vehicle identification number image, and obtain the world coordinate information of the vehicle identification number from the pixel coordinate information of the vehicle identification number; obtain the text information of the vehicle identification number through vehicle identification number image recognition, determine the vehicle model according to the recognized vehicle identification number text information, obtain the coordinate transformation matrix between the vehicle identification number and the charging interface, and then obtain the world coordinate information of the charging interface; In the step S1, a vehicle identification number recognition model is used to label the outer rectangle of the vehicle identification number in the image; The construction method of the vehicle identification number recognition model comprises the following steps: 11) Image collection: Collect images of the front windshield containing the outer rectangle of the vehicle identification number of different models from different angles and heights by using the camera deployed in the charging station to form an original image set; 12) Preprocessing: Preprocess and data augment the original image set to obtain an enhanced image set; Crop and classify the enhanced image set, label the outer rectangle of the vehicle identification number, obtain the center point coordinates, height, width and angle information of the target frame, and finally convert them into a data format suitable for vehicle identification number recognition model training to obtain an enhanced data set; 13) Create a vehicle identification number recognition model network; 14) Train the vehicle identification number recognition model using the enhanced data set.

2. The method of claim 1, wherein: The vehicle identification number recognition model uses an improved YOLOv5 target detection convolutional neural network model, and the improvement method of the YOLOv5 target detection convolutional neural network model in step 13) is: 131) Use the K-means++ clustering algorithm to generate anchors; 132) Replace the backbone network of YOLOv5 with a more lightweight MobileNetv3 network; 133) Replace the neck network structure with a weighted bidirectional feature pyramid network BiFPN structure.

3. The method of claim 2, wherein: In step 131), the K-means++ clustering algorithm generates anchors in the following steps: 1311) randomly select a sample from the dataset ;​​​ 1312) Calculate the shortest distance between each sample and the current cluster centers :​ ) wherein, represents the degree of overlap of the anchors with the Bounding box; 1313) The probability that each sample is selected as the next cluster center is calculated with the following formula: Probability = exp(-distance) / Σ exp(-distance) for all samples in the cluster : 1314) The next cluster center is chosen according to the roulette method; ; 1315) Judgment Is it equal to If so, then it has been selected. If no cluster centers are found, stop iterating; otherwise, proceed to step 1312.

4. The method of claim 2, wherein: In step 132), the function implementation steps for replacing the backbone network of YOLOv5 with a MobileNetv3 network are: 1321) Use 1*1 convolution to increase the dimension of the input feature to exp size, and pass it through a nonlinear activation function NL; 1322) Pass the output feature through a 3*3 DW convolution, and pass it through a nonlinear activation function NL, at which time the channel number exp size does not change; 1323) The output feature is input into the SE block, first, global average pooling is performed on each channel to obtain a 1*1*exp size vector; then the vector is input into the first fully connected layer, and the activation function is ReLU, to obtain a 1*1*(exp size / 4) vector; then the output is input into the second fully connected layer, and the activation function is , to obtain a 1*1*exp size vector; finally, each element in the vector is multiplied by the corresponding channel feature matrix of the feature map input into the SE layer to obtain a new feature map with the same size as the feature map; 1324) Finally, perform 1*1 convolution dimension reduction processing.

5. The method of claim 4, wherein: The activation function is is: Wherein, ReLU6 represents the activation function.

6. The method of claim 1, wherein: In step S2, the coordinate conversion matrix from the pixel coordinate information of the vehicle identification number to the world coordinate information of the vehicle identification number is: wherein [u, v] T denotes the coordinate of the frame number identification in the image pixel coordinate system O-U-V; , , and denote the physical size of each pixel in the u-axis and v-axis directions, and f denotes the focal length of the camera; denotes the Z-axis coordinate of the frame number identification in the camera coordinate system; and denote the coordinates of the image physical coordinate system origin in the image pixel coordinate system; denotes the orthogonal unit rotation matrix; denotes the three-dimensional translation vector; and [Xw, Yw, Zw] denotes the coordinate of the frame number identification in the world coordinate system.

7. A new energy vehicle charging gun automatic plugging and positioning method, characterized in that: Comprising the following steps: Step one: Obtain the world coordinate information of the charging interface by using the new energy vehicle charging interface automatic positioning method according to any one of claims 1-6; obtain the world coordinate information of the charging gun according to the coordinate transformation matrix between the charging gun and the camera; Step two: according to the world coordinate information of the charging interface and the world coordinate information of the charging gun, the moving path of the charging gun to the charging interface is planned; Step three: the charging gun is moved to the charging interface position according to the planned moving path; meanwhile, the charging port cover is opened by interacting with the automobile; Step four: the charging gun is inserted into the charging interface.

8. The automatic plugging and positioning method of the new energy vehicle charging gun according to claim 7, characterized in that: In step four, during the insertion of the charging gun, the position of the charging gun is compensated in real time by using visual recognition, so that it is aligned with the charging interface.

Images