A mobile charging robot control method and system based on artificial intelligence

Abstract

The present invention relates to the field of data processing, and the present invention relates to a mobile charging robot control method and system based on artificial intelligence. By acquiring historical images of vehicles to be charged, the probability of different weather conditions is analyzed using an XGBoost model based on the edge sharpness and high-frequency components of the image, and the corresponding image processing algorithm is selected to generate an optimized vehicle image. The candidate frame position and confidence of the license plate are obtained through the target detection algorithm trained by the YOLOV5 deep learning model. The candidate frame confidence under different weather conditions is integrated to accurately locate the license plate position information, and finally the actual position of the charging port is inferred based on the vehicle model and the license plate position, and the charging robot is controlled to align with the charging port. The automatic charging operation is finally completed. The present invention uses weather probability analysis and image optimization, combined with a target detection algorithm, to accurately locate the vehicle charging port and realize automated charging operations.

Description

Technical Field

[0001] The present invention relates to the field of data processing, and in particular to a mobile charging robot control method and system based on artificial intelligence. Background Art

[0002] At present, the proportion of new energy vehicles in my country has reached nearly 30%, but the construction speed of charging facilities has lagged significantly, and there are still many deficiencies at the charging end, and "mileage anxiety" still exists. In this case, mobile charging robots came into being. They integrate battery packs and charging and discharging systems on mobile chassis, and can be flexibly moved to the target vehicle to charge the vehicle through remote control or automatic driving. At present, mobile charging robots need to identify the vehicle to be charged when working, so as to use the robot arm to connect the charging gun to the vehicle charging port. The key link is that the mobile charging robot needs to accurately locate according to the posture of the vehicle to be charged. The existing method is to preset a target position in the parking space where the vehicle to be charged is located, so that the mobile charging robot navigates to the target position, and then performs the next related operations based on the target position. However, the above-mentioned preset target position is based on the premise that the vehicle to be charged is parked at a standard parking angle in the parking space. Due to the different parking habits of different users, if the vehicle to be charged fails to park in the parking space at the standard parking angle, the mobile charging robot may not be able to charge the vehicle to be charged normally.

[0003] At present, the Chinese patent application document with publication number CN115880681A discloses a control method for a mobile charging robot based on the license plate of the vehicle to be charged. The method obtains the position adjustment value and posture adjustment value corresponding to the mobile charging robot through the license plate in the vehicle image, and corrects the posture of the mobile charging robot, thereby solving the problem that the mobile charging robot cannot charge the vehicle to be charged normally when the vehicle to be charged is not parked in the parking space according to the standard parking angle. However, this method does not take into account that the vehicle license plate recognition may be affected by adverse weather conditions such as rain, snow, and fog. These conditions will cause the recognition rate of the license plate recognition system to decrease, resulting in the inability to provide accurate posture correction for the subsequent mobile charging robot. Summary of the invention

[0004] In view of the problem that the system recognition rate decreases due to the above-mentioned weather conditions, in the first aspect, the present invention proposes a mobile charging robot control method based on artificial intelligence, including: obtaining a set number of images of vehicles to be charged and the initial position of the license plate and the initial position of the charging port when the vehicles to be charged are parked in standard parking spaces; obtaining the probability of rainy days, snowy days and foggy days corresponding to the images of the vehicles to be charged based on a classifier; obtaining the accurate position of the candidate frame of the license plate of the vehicles to be charged, specifically:

[0005]

[0006] in Indicates the exact position of the candidate frame of the license plate; , and They respectively represent the normalized values ​​of the probabilities of rainy days, snowy days, and foggy days obtained by the classifier for the image of the vehicle to be charged; , and Respectively represent the positions of the license plate candidate frames in rainy, snowy and foggy scenes; obtain the exact position of the license plate of the vehicle to be charged according to the exact position of the license plate candidate frame; calculate the coordinate difference between the initial position of the license plate and the exact position of the license plate to determine the offset of the vehicle to be charged, and determine the actual position of the vehicle charging port based on the offset and the initial position of the charging port; control the charging robot to move to the vehicle charging port based on the actual position and complete the charging operation.

[0007] Through an image processing method based on weather probability analysis, the present invention uses a classifier to obtain the candidate frame position of the license plate of the vehicle to be charged under complex weather conditions such as rainy days, snowy days, and foggy days, and weightedly fuses the candidate frame positions of the license plate in different weather scenes to determine the exact position of the license plate. Compared with the prior art, the present invention uses probabilistic normalization processing of multiple weather scenes to achieve higher accuracy in license plate recognition under severe weather conditions, thereby accurately calculating the license plate offset and inferring the actual position of the charging port. Ultimately, the charging robot can automatically move to the charging port and complete the charging operation through path planning and posture adjustment, overcoming the limitations of traditional methods with low recognition accuracy and inaccurate docking and charging in complex weather conditions, greatly improving the robustness and intelligence of the system under various weather conditions.

[0008] Furthermore, the method for obtaining the position of the license plate candidate frame in the rainy, snowy and foggy scenes is as follows: in response to the probability of rainy days in the image of the vehicle to be charged, a deraining algorithm based on a convolutional neural network is used for smoothing; in response to the probability of snowy days in the image of the vehicle to be charged, a desnowing algorithm based on sparse coding is used for smoothing; in response to the probability of foggy days in the image of the vehicle to be charged, a defogging algorithm based on a dark channel prior is used for smoothing; the position of the license plate candidate frame in the smoothed image under different weather scenes is marked and input into the YOLOV5 target detection model for training; the position of the license plate candidate frame in the rainy, snowy and foggy scenes is obtained respectively through the trained YOLOV5 target detection model.

[0009] The deraining, desnowing and defogging algorithms based on different weather conditions improve the clarity of images in bad weather conditions and optimize the license plate recognition effect. Compared with the existing single algorithm, the image processing for different weather scenes improves the processing effect and accuracy.

[0010] Furthermore, the weight , and Other ways to obtain it include:

[0011]

[0012] in , and Respectively represent the confidence of the license plate candidate frame in rainy, snowy and foggy scenes; , and They respectively represent the probabilities of rainy days, snowy days and foggy days obtained by the classifier of the image of the vehicle to be charged; the confidence of the license plate candidate frame is obtained through the YOLOV5 target detection model.

[0013] A weighted calculation method for license plate position based on different weather probability weights is used to further improve the accuracy of license plate position by weighted summing of the confidence of license plate candidate frames under different weather conditions. Compared with traditional methods, comprehensive weather model processing has higher accuracy and adaptability.

[0014] Furthermore, the training process of the classifier is as follows: calculating the mean value of the gradient amplitude of all edge pixels in the image of the vehicle to be charged and the proportion of high-frequency components to obtain the characteristic parameters of the historical image; obtaining the weather label of the image of the vehicle to be charged; and inputting the characteristic parameters and weather labels as training data into the XGBoost classifier to obtain a trained classifier.

[0015] Using the XGBoost classifier to classify weather provides efficient and accurate weather probability calculation. Compared with other machine learning models, XGBoost has stronger generalization ability and faster convergence speed, and is suitable for real-time scenarios.

[0016] Furthermore, the method also includes grayscale processing of the image of the vehicle to be charged and noise reduction using a median filter.

[0017] Furthermore, the edge pixels are obtained through Canny edge detection.

[0018] Furthermore, the high-frequency component ratio is obtained by performing Fourier transform on the image of the vehicle to be charged.

[0019] Furthermore, the path planning is obtained through an RRT path planning algorithm.

[0020] The RRT path planning algorithm provides an optimal path planning solution, allowing the robot to safely and quickly approach the vehicle charging port in a complex environment. Compared with other path planning algorithms, the RRT algorithm has better performance in handling dynamic obstacles and irregular paths.

[0021] In a second aspect, the present invention provides an artificial intelligence-based mobile charging robot control system, comprising a processor and a memory, wherein the memory stores computer program instructions, and when the computer program instructions are executed by the processor, the artificial intelligence-based mobile charging robot control method of the present invention is implemented.

[0022] The technical effects of the present invention are:

[0023] Through a variety of image processing algorithms based on weather probability analysis, combined with a deep learning model, the vehicle license plate position is accurately identified, and the actual position of the charging port is calculated based on the license plate offset, and the charging robot is controlled to automatically complete the charging operation through path planning and posture adjustment. This method improves the recognition accuracy and automated charging efficiency under complex weather conditions, and has stronger adaptability and robustness. BRIEF DESCRIPTION OF THE DRAWINGS

[0024] By reading the following detailed description with reference to the accompanying drawings, the above and other objects, features and advantages of the exemplary embodiments of the present invention will become readily understood. In the accompanying drawings, several embodiments of the present invention are shown in an exemplary and non-limiting manner, and the same or corresponding reference numerals represent the same or corresponding parts, wherein:

[0025] Figure 1 is a flow chart schematically illustrating a method for controlling a mobile charging robot based on artificial intelligence according to an embodiment of the present invention;

[0026] Figure 2 is a grayscale schematic diagram schematically showing the mobile charging robot in the embodiment of the present invention when charging;

[0027] Figure 3 It is a block diagram schematically showing the structure of an artificial intelligence-based mobile charging robot control system in an embodiment of the present invention. DETAILED DESCRIPTION

[0028] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative work are within the scope of protection of the present invention.

[0029] The specific embodiments of the present invention are described in detail below with reference to the accompanying drawings.

[0030] Embodiment of the control method of the mobile charging robot based on artificial intelligence:

[0031] like Figure 1As shown, the artificial intelligence-based mobile charging robot control method of the present invention includes:

[0032] S1. Obtain historical images of vehicles to be charged taken by a mobile charging robot; and obtain image feature parameters.

[0033] When the mobile charger robot is working, it controls the charging gun through the mechanical arm to insert it into the car charging port for charging. Figure 2 However, when it is working outdoors, in order to identify the real-time weather conditions of the mobile charging robot during charging, it is necessary to process and analyze the images it obtains. The details are as follows: First, 500 historical RGB images of the vehicles to be charged taken by the mobile charging robot can be obtained. The pixel size of each image can be set to In this embodiment, the empirical value can be taken ; Then all historical RGB images are grayed out and converted into grayscale images; then the median filter can be used to remove noise in the grayscale image, such as salt and pepper noise, to ensure image quality; further, the Canny edge detection algorithm can be used to obtain all edge pixels in the image and the gradient amplitude corresponding to the edge pixels, and the average gradient amplitude of all edge pixels is counted to obtain the edge sharpness of the image. , specifically:

[0034]

[0035] in Indicates the image The gradient amplitude of edge pixels; Represents the total number of edge pixels in the image. By calculating the edge sharpness of the image, the clarity of the edge is quantified. A high edge sharpness indicates a clear object outline, while a low edge sharpness indicates a blurred object outline. The above median filter and Canny edge detection algorithm are well-known technologies and will not be described here.

[0036] In sunny days and other conditions with sufficient light, the edges of vehicles and license plates are relatively clear and have high edge sharpness, which can accurately distinguish between objects and backgrounds. In foggy days, when fog is dispersed in the air, the sharpness of object edges will be significantly reduced, and the boundary between the edges of vehicles or license plates and the background may become blurred. In rainy and snowy weather, the edge sharpness of local areas of the image will be reduced due to the influence of raindrops or snowflakes. Therefore, the edge sharpness of the image can be used as a characteristic parameter to determine the real-time weather of the charging scene.

[0037] In order to avoid the phenomenon of overfitting and poor robustness of a single feature parameter in the subsequent training model, the frequency distribution in the image can also be analyzed by Fourier transform in this embodiment to distinguish the image features under different weather conditions. On sunny days, the high-frequency components in the image are more detailed; while on rainy, foggy, snowy and other weather conditions, the low-frequency components in the image are more smooth areas. The proportion of high-frequency components in the image is counted and recorded as , specifically:

[0038]

[0039] in represents the value in the frequency domain after Fourier transform, Represents the value of the high-frequency region, Represents the value of all frequency regions.

[0040] S2. Determine the current weather probability and select the corresponding image processing algorithm; mark the license plates of the vehicles to be charged in the images processed by the algorithm to complete the construction of the data set of vehicles to be charged.

[0041] In step S1, the edge sharpness and high-frequency component ratio of 500 historical images were obtained, and the weather label of each historical image was further annotated. And the proportion of high-frequency components Construct the feature matrix as follows:

[0042]

[0043] And the weather labels of all images constitute the label set ,have ; Each sample, that is, each historical image corresponds to a weather label, and finally the feature matrix is ​​combined with the label set The input is input into the XGBoost algorithm for training. The input of the algorithm also includes some major hyperparameters. In this embodiment, the empirical values ​​are respectively taken as follows: learning rate ; Maximum depth of the tree ; Subsample proportion ; Feature sampling ratio for each tree ; Minimum splitting loss ; L2 regularization term ; L1 regularization term The implementer of the above experience value can also select other values ​​according to the actual situation, and this algorithm belongs to the known technology, and the rest will not be described in detail.

[0044] After obtaining the trained XGBoost model, the feature parameters of the real-time image of the vehicle to be charged are obtained and input into the model, and the model outputs the probability of various types of weather corresponding to the image of the vehicle to be charged. In one embodiment, an image of the vehicle to be charged taken by a mobile charging robot is input, and the probability of the charging scene weather category corresponding to the image is output, such as 0.7 for sunny days, 0 for rainy days, 0.1 for snowy days, and 0.2 for foggy days. At this time, the probability of sunny days is greater than the probabilities of other weather conditions, and the weather of the image is considered to be sunny. Since the image obtained on a sunny day is clear enough, no further processing is performed on the case where the weather category of the image of the vehicle to be charged is sunny in this embodiment.

[0045] In another embodiment, the XGBoost model outputs the charging scene weather category probabilities for an image as follows: sunny 0, rainy 0.7, snowy 0.1, foggy 0.2. This indicates that when the image of the vehicle to be charged is acquired in real time, the weather conditions are complicated. To facilitate the subsequent labeling of the license plates of the vehicle to be charged, different image processing algorithms are used for different weather probabilities. Since there are three types of weather probabilities in the classification results of the above model, the images are input into three corresponding optimization processing algorithms respectively. In this embodiment, a deraining algorithm based on a convolutional neural network, a defogging algorithm based on a dark channel prior, and a desnowing algorithm based on sparse coding can be used to finally obtain three optimized images. The above-mentioned optimization algorithms belong to the known technology and will not be described in detail here.

[0046] The deraining, desnowing and defogging algorithms based on different weather conditions improve the clarity of images in bad weather conditions and optimize the license plate recognition effect. Compared with the existing single algorithm, the image processing for different weather scenes improves the processing effect and accuracy.

[0047] After obtaining the above three images, the Python third-party tool Iabelimg is used to annotate the license plate of the vehicle to be charged in each image, thereby completing the construction of a dataset of vehicles to be charged. The total dataset then includes all 500 images of vehicles to be charged that have been processed by different optimization algorithms, as well as the license plate position annotated in each image.

[0048] Using the XGBoost classifier to classify weather provides efficient and accurate weather probability calculation. Compared with other machine learning models, XGBoost has stronger generalization ability and faster convergence speed, and is suitable for real-time scenarios.

[0049] S3. Train a deep learning model; use the trained optimal model to perform target detection on the real-time image to obtain the license plate candidate frame of the vehicle to be charged in the real-time image and its corresponding confidence.

[0050] In step S2, the total data set is obtained, and then the images with license plate labels in the total data set can be input into the YOLOV5 model for training. In this embodiment, the Adam optimization algorithm can be used to adjust the model parameters, and the F1 criterion can be used to evaluate the model performance when training the model. The cutoff condition of the model training can be set to: stop training when the number of training times is greater than the set experience value of 50 times. This algorithm belongs to the known technology, and the specific process will not be repeated.

[0051] At this point, the YOLOV5 model with the best prediction effect is obtained. The model is fed with the image of the vehicle to be charged that has been optimized and subsequently taken, and the model outputs the candidate frame position of the license plate in the image and its corresponding confidence.

[0052] S4. Obtain the position information corresponding to the license plate candidate frame of the current vehicle to be charged; and control the mobile charging robot to complete the charging operation based on the position information and vehicle information.

[0053] In one embodiment, the mobile charging robot first takes an initial image of the vehicle to be charged; then inputs it into the XGBoost classifier to obtain the probability of various types of weather when the initial image is taken. When it is judged to be a non-sunny scene, different optimization algorithms are used to process it according to the weather classification; the processed image is further input into the YOLOV5 model to obtain the candidate frame position of the license plate in each image of the vehicle and the corresponding confidence.

[0054] In this embodiment, the rainy day probability of the initial image output by the XGBoost classifier is recorded as , and there is Then, the initial image is processed using a convolutional neural network-based deraining algorithm to obtain a derained image. After inputting the YOLOV5 model, the center coordinates of the candidate frame of the vehicle license plate to be charged in the derained image are obtained. And the width of the license plate candidate frame ,high and confidence ; If in this embodiment, the result output by the XGBoost classifier of the initial image also includes the probability of snow and the probability of fog, which are recorded as and , and there is , then the same as the above processing steps, the center coordinates of the candidate frames of the vehicle license plate to be charged in the snow-removed image and the fog-removed image are obtained, respectively and ; At the same time, the width, height and confidence of the candidate frame of the vehicle license plate to be charged in the snow-removed image are , and , the width, height and confidence of the candidate frame of the vehicle license plate to be charged in the dehazed image are and .

[0055] In summary, the position information parameters of the candidate frame of the current vehicle to be charged under the comprehensive weather model are obtained, and the calculation process of the central horizontal coordinate of the candidate frame is illustrated as follows:

[0056]

[0057] in Indicates the central horizontal coordinate of the final candidate box; Represents the confidence of the candidate frame of the vehicle license plate to be charged in the rain-free image; Indicates the confidence of the candidate frame of the vehicle license plate to be charged in the snow-removed image; Represents the confidence of the candidate frame of the vehicle license plate to be charged in the dehazed image; Indicates the probability of rainy days after the initial image is output by the XGBoost classifier; It represents the snow probability of the initial image after the XGBoost classifier outputs; Indicates the foggy probability of the initial image output by the XGBoost classifier; The central horizontal coordinate of the candidate frame of the vehicle license plate to be charged in the rain-removed image; The central horizontal coordinate of the candidate frame of the license plate of the vehicle to be charged in the snow-removed image; Represents the central horizontal coordinate of the candidate frame of the vehicle license plate to be charged in the dehazed image.

[0058] Similarly, obtain the remaining position parameters under the comprehensive weather model, including the central ordinate of the candidate box And the width of the license plate candidate frame and high At this point, the position information corresponding to the license plate candidate frame of the vehicle to be charged under the current weather conditions is obtained.

[0059] This embodiment provides a weighted calculation method for license plate position based on different weather probability weights, which further improves the accuracy of license plate position by weighted summation of the confidence of license plate candidate frames under different weather conditions. Compared with traditional methods, comprehensive weather model processing has higher accuracy and adaptability.

[0060] After obtaining the real-time image of the vehicle to be charged, in this embodiment, the trained convolutional neural network model can be first used to identify the captured vehicle image to obtain the specific model of the vehicle. The convolutional neural network belongs to a well-known technology. The input of its training process includes all historical vehicle images taken by the mobile robot and the model labels of the vehicles manually marked. The specific training process will not be repeated here; then, according to the vehicle model, the charging port location and related vehicle parameter information of the vehicle can be obtained from a third-party vehicle database, such as the domestic database "Extreme Speed ​​Data"; finally, combined with the vehicle model recognition result, the preset position of the charging port when the vehicle is parked at the charging position according to the parking standard is accurately located, and the parking standard is that the whole vehicle is parked within the calibrated parking line range.

[0061] After obtaining the preset position of the charging port of the vehicle to be charged, the position information corresponding to the license plate candidate frame obtained based on the comprehensive weather model, that is, the position center coordinates of the license plate candidate frame And the width and height of the license plate , determine the actual location of the charging port when the vehicle to be charged is in the state of autonomous parking. The specific process is as follows:

[0062] First, the standard position of the license plate when the vehicle to be charged is parked according to the standard is determined based on the identified vehicle model and vehicle parameters; then the position information corresponding to the license plate candidate frame obtained by the comprehensive weather model is obtained, and its offset relative to the standard position is calculated; finally, according to the license plate offset, the preset position of the charging port is adjusted to determine the actual position of the charging port.

[0063] By obtaining the actual position of the charging port, in this embodiment, the RRT path planning algorithm can be used to plan the shortest path in the environment of the charging station; then the PID control algorithm is used to adjust the position and angle of the robot, and the posture of the robot is adjusted in real time, so that the robot approaches the charging port from the current starting position according to the set shortest path. When the robot approaches the charging port, the robot arm is aligned based on the actual position of the charging port; finally, after confirming that the charging head is completely aligned with the charging port, the robot arm performs the docking action, inserts the charging head into the charging port, and completes the charging connection.

[0064] The RRT path planning algorithm provides an optimal path planning solution, allowing the robot to safely and quickly approach the vehicle charging port in a complex environment. Compared with other path planning algorithms, the RRT algorithm has better performance in handling dynamic obstacles and irregular paths.

[0065] Through an image processing method based on weather probability analysis, the present invention uses a classifier to obtain the candidate frame position of the license plate of the vehicle to be charged under complex weather conditions such as rainy days, snowy days, and foggy days, and weightedly fuses the candidate frame positions of the license plate in different weather scenes to determine the exact position of the license plate. Compared with the prior art, the present invention uses probabilistic normalization processing of multiple weather scenes to achieve higher accuracy in license plate recognition under severe weather conditions, thereby accurately calculating the license plate offset and inferring the actual position of the charging port. Ultimately, the charging robot can automatically move to the charging port and complete the charging operation through path planning and posture adjustment, overcoming the limitations of traditional methods with low recognition accuracy and inaccurate docking and charging in complex weather conditions, greatly improving the robustness and intelligence of the system under various weather conditions.

[0066] Embodiment of the mobile charging robot control system based on artificial intelligence:

[0067] On the other hand, the present invention also provides a mobile charging robot control system based on artificial intelligence. Figure 3 As shown, the artificial intelligence-based mobile charging robot control system includes a processor and a memory, and the memory stores computer program instructions. When the computer program instructions are executed by the processor, an artificial intelligence-based mobile charging robot control method according to the first aspect of the present invention is implemented.

[0068] The artificial intelligence-based mobile charging robot control system also includes other components familiar to those skilled in the art, such as communication interfaces, and their settings and functions are known in the art, so they will not be described in detail here.

[0069] In the present invention, the aforementioned memory may be any tangible medium containing or storing a program that can be used by or in combination with an instruction execution system, apparatus or device. For example, a computer-readable storage medium may be any appropriate magnetic storage medium or magneto-optical storage medium, such as a resistive random access memory RRAM (Resistive Random Access Memory), a dynamic random access memory DRAM (Dynamic Random Access Memory), a static random access memory SRAM (Static Random-Access Memory), an enhanced dynamic random access memory EDRAM (Enhanced Dynamic Random Access Memory), a high-bandwidth memory HBM (High-Bandwidth Memory), a hybrid memory cube HMC (Hybrid Memory Cube), etc., or any other medium that can be used to store the required information and can be accessed by an application, a module, or both. Any such computer storage medium may be part of a device or accessible or connectable to a device. Any application or module described in the present invention may be implemented using computer-readable / executable instructions that may be stored or otherwise maintained by such a computer-readable medium.

[0070] In the description of this specification, "plurality" or "several" means at least two, such as two, three or more, etc., unless otherwise clearly and specifically defined.

[0071] Although this specification has shown and described a number of embodiments of the present invention, it will be apparent to those skilled in the art that such embodiments are provided by way of example only. Those skilled in the art will conceive of many modifications, changes and alternatives without departing from the ideas and spirit of the present invention. It should be understood that in the practice of the present invention, various alternatives to the embodiments of the present invention described herein may be employed.

Claims

1. A mobile charging robot control method based on artificial intelligence, characterized in that: The method comprises: Obtain a set number of images of vehicles to be charged, as well as the initial positions of the license plates and charging ports of the vehicles to be charged when they are parked in standard parking spaces; Based on the classifier, the probability of rainy days, snowy days and foggy days corresponding to the image of the vehicle to be charged is obtained; Get the exact position of the candidate frame of the vehicle license plate to be charged, specifically: ； in Indicates the exact position of the candidate frame of the license plate; , and They respectively represent the normalized values ​​of the probabilities of rainy days, snowy days, and foggy days obtained by the classifier for the image of the vehicle to be charged; , and Respectively represent the positions of the license plate candidate frames in rainy, snowy and foggy scenes; obtain the exact position of the license plate of the vehicle to be charged according to the exact position of the license plate candidate frame; The coordinate difference between the initial position of the license plate and the accurate position of the license plate is calculated to determine the offset of the vehicle to be charged, and the actual position of the vehicle charging port is determined based on the offset and the initial position of the charging port; based on the actual position, the charging robot is controlled to move to the vehicle charging port and complete the charging operation.

2. The method for controlling a mobile charging robot based on artificial intelligence according to claim 1, characterized in that: The method for obtaining the position of the license plate candidate frame in the rainy, snowy and foggy scenes is as follows: In response to the probability of rainy days in the image of the vehicle to be charged, a deraining algorithm based on a convolutional neural network is used for smoothing; in response to the probability of snowy days in the image of the vehicle to be charged, a desnowing algorithm based on sparse coding is used for smoothing; in response to the probability of foggy days in the image of the vehicle to be charged, a defogging algorithm with a dark channel prior is used for smoothing; the positions of the license plate candidate frames in the smoothed images under different weather scenes are marked and input into the YOLOV5 target detection model for training; the positions of the license plate candidate frames in rainy, snowy and foggy scenes are obtained respectively through the trained YOLOV5 target detection model.

3. The method for controlling a mobile charging robot based on artificial intelligence according to claim 2, characterized in that: Weight , and Other ways to obtain it include: ； in , and Respectively represent the confidence of the license plate candidate frame in rainy, snowy and foggy scenes; , and They respectively represent the probabilities of rainy days, snowy days and foggy days obtained by the classifier of the image of the vehicle to be charged; the confidence of the license plate candidate frame is obtained through the YOLOV5 target detection model.

4. The method for controlling a mobile charging robot based on artificial intelligence according to claim 1, characterized in that: The training process of the classifier is: The characteristic parameters of the historical image are obtained by calculating the mean value of the gradient amplitude of all edge pixels in the image of the vehicle to be charged and the proportion of high-frequency components; the weather label of the image of the vehicle to be charged is obtained; and the characteristic parameters and the weather label are input into the XGBoost classifier as training data to obtain a trained classifier.

5. The method for controlling a mobile charging robot based on artificial intelligence according to claim 4 is characterized in that: The method also includes grayscale processing of the image of the vehicle to be charged and noise reduction using a median filter.

6. The method for controlling a mobile charging robot based on artificial intelligence according to claim 4 is characterized in that: The edge pixels are obtained through Canny edge detection.

7. The method for controlling a mobile charging robot based on artificial intelligence according to claim 4, characterized in that: The high-frequency component ratio is obtained by performing Fourier transform on the image of the vehicle to be charged.

8. The method for controlling a mobile charging robot based on artificial intelligence according to claim 1, characterized in that: Controlling the movement of the charging robot also includes obtaining the optimal path through the RRT path planning algorithm.

9. A mobile charging robot control system based on artificial intelligence, characterized in that: It includes a processor and a memory, wherein the memory stores computer program instructions, and when the computer program instructions are executed by the processor, the artificial intelligence-based mobile charging robot control method described in any one of claims 1 to 8 is implemented.

Images