Method, system and battery swap platform for detecting vehicle parking state in battery swap platform

Abstract

The present application relates to a method and system for detecting a parking state of a vehicle in a battery swap platform and a battery swap platform comprising such a system. The method comprises the following steps: a) acquiring an image of the battery swap platform on which a vehicle is parked by using a panoramic camera; b) inputting the image into a model, wherein the model is constructed based on a plurality of sets of training data comprising sample images and annotation information, the annotation information depicting whether a vehicle in the sample image reaches a preset position of the battery swap platform; and c) processing the image based on the model to determine whether the vehicle is parked in the preset position of the battery swap platform. According to the method, system and battery swap platform of the present application, the parking state of the vehicle in the battery swap platform can be automatically detected by applying a deep learning-based model, thereby preventing misoperation on the vehicle and thus preventing damage to the battery.

Description

Technical Field

[0001] The present invention relates to a method, a system for detecting the parking status of a vehicle in a battery swapping platform, and a battery swapping platform including such a system. Background Technology

[0002] Currently, there are two main modes of energy replenishment for electric vehicles: vehicle charging and battery swapping (i.e., battery replacement). Vehicle charging can be further divided into AC slow charging and DC fast charging. AC slow charging takes a long time and is limited by parking availability. Furthermore, while DC fast charging offers high power and short charging time, it puts a greater strain on the power grid and reduces battery lifespan. Conversely, battery swapping provides rapid energy replenishment for electric vehicles while minimizing damage to the battery. In addition, battery swapping can achieve peak-shaving and energy storage for the power grid, thereby improving the overall utilization efficiency of power equipment.

[0003] In battery swapping, specialized equipment is typically required to remove the depleted battery from the vehicle chassis and then reinstall a new or fully charged battery. These battery swapping platforms are usually configured to be automatic or semi-automatic, and before the swap can begin, the electric vehicle must be accurately parked in its designated spot on the platform, which is especially important in automated battery swapping scenarios. Automatic detection of the vehicle's parking status within the battery swapping platform not only prevents accidental vehicle operation but also reduces the operating costs of the swapping process. Summary of the Invention

[0004] The purpose of this invention is to prevent accidental operation of the vehicle during battery swapping and thereby reduce battery swapping time.

[0005] Furthermore, the present invention aims to solve or alleviate other technical problems existing in the prior art.

[0006] To address the aforementioned technical problems, one objective of this invention is to provide a method for detecting the parking status of a vehicle in a battery swapping platform.

[0007] According to one aspect of the present invention, a method for detecting the parking status of a vehicle in a battery swapping platform is provided, comprising the following steps:

[0008] a) Using a panoramic camera device to acquire images of the battery swapping platform, on which vehicles are parked;

[0009] b): Input the image into the model, wherein the model is constructed based on multiple sets of training data containing sample images and annotation information, the annotation information depicting whether the vehicle has reached a preset location on the battery swapping platform in the sample images; and

[0010] c): Based on the model, the image is processed to determine whether the vehicle has parked in the preset position of the battery swapping platform.

[0011] According to one aspect of the present invention, a method for detecting vehicle parking status in a battery swapping platform is provided, wherein a calibration image is constructed to detect whether imaging distortion exists in the panoramic camera device, the calibration image being based on a source image of the battery swapping platform when the vehicle is not parked and an outline image of the wheel fixing device of the battery swapping platform.

[0012] According to one aspect of the present invention, a method for detecting vehicle parking status in a battery swapping platform is provided, wherein the outlined image is converted into an RGB image and a grayscale image, and a calibration image is constructed based on the RGB image and the normalized grayscale image of the grayscale image, wherein the value of the normalized grayscale image is 0 or 1.

[0013] According to one aspect of the present invention, a method for detecting vehicle parking status in a battery swapping platform involves performing ROI processing on images of the battery swapping platform acquired by a panoramic camera device during the construction of training data to obtain sample images, wherein the height and / or width of the sample images have a deviation from a preset value that follows a normal distribution.

[0014] According to one aspect of the present invention, a method for detecting the parking status of a vehicle in a battery swapping platform is provided, wherein it is determined whether the center region of the wheel in the sample image has reached a preset position of the wheel fixing device of the battery swapping platform, so as to obtain annotation information belonging to the sample image.

[0015] According to one aspect of the present invention, a method for detecting the parking status of a vehicle in a battery swapping platform is provided, wherein the wheel fixing device is constructed as a V-shaped roller groove, and it is determined whether the projection of the wheel center region is located on the central axis of the V-shaped roller groove, or whether the wheel center region is located between two vertical contact normals, wherein the two vertical contact normals pass through the two end tangent points of the wheel and the V-shaped roller groove in the sample image, respectively, to obtain annotation information belonging to the sample image.

[0016] According to one aspect of the present invention, a method for detecting vehicle parking status in a battery swapping platform is provided, wherein the model is constructed based on a convolutional neural network model, the model having an input layer, a hidden layer including a feature filtering module and a Canoe module, and an output layer, wherein the model is based on the Fish activation function:

[0017]

[0018] According to one aspect of the present invention, a method for detecting vehicle parking status in a battery swapping platform is provided, wherein the Canoe module includes a Reshape layer, a convolutional layer, a Sigmoid layer, a flattening layer, and a fully connected layer, wherein the format of an image from the input layer is converted from NHWC to N1W(HC) based on the Reshape layer.

[0019] Furthermore, according to another aspect of the present invention, a system for detecting the parking status of a vehicle is provided, which can be used in a battery swapping platform, the system comprising:

[0020] A panoramic camera device can be installed at a battery swapping platform, the panoramic camera device being configured to acquire images of the battery swapping platform, wherein the battery swapping platform is used for parking vehicles;

[0021] A judgment device, connected to the panoramic camera device, is configured to process images from the panoramic camera device based on a model and determine whether a vehicle has parked at a preset position on the battery swapping platform. The model is constructed based on multiple sets of training data including sample images and annotation information, where the annotation information indicates whether the vehicle has reached the preset position on the battery swapping platform in the sample images.

[0022] A triggering device is arranged in the battery swapping platform and connected to the judgment device, the triggering device being configured to trigger or stop the battery swapping process of the vehicle based on the judgment result of the judgment device.

[0023] According to another aspect of the invention, the system has a calibration device configured to detect imaging distortion of the panoramic camera based on a calibration image.

[0024] According to another aspect of the invention, the system includes a calibration image generator configured to generate a calibration image based on a source image of the battery swapping platform when the vehicle is not parked and an outline image of the wheel fixing device of the battery swapping platform. According to another aspect of the invention, the system has a preprocessing unit connected to a panoramic camera device, configured to perform ROI processing on the image of the battery swapping platform acquired by the panoramic camera device and output a sample image, wherein the height and / or width of the sample image have a deviation from a preset value that follows a normal distribution.

[0025] According to another aspect of the present invention, the system has a labeling information generator connected to the preprocessing device, which is configured to determine whether the center region of the wheel in the sample image reaches a preset position of the wheel fixing device of the battery swapping platform and generate labeling information belonging to the sample image based on the determination result.

[0026] According to another aspect of the invention, the wheel fixing device is constructed as a V-shaped roller groove.

[0027] According to another aspect of the present invention, the panoramic camera device is constructed as a fisheye camera.

[0028] According to another aspect of the invention, the fisheye camera is arranged such that when the vehicle is parked in a predetermined position on the battery swapping platform, the fisheye camera is aligned with the central region of the wheel.

[0029] Finally, according to another aspect of the invention, a battery swapping platform is proposed, the battery swapping platform comprising such a system.

[0030] The advantages of the method, system, and battery swapping platform for detecting vehicle parking status in a battery swapping platform according to this disclosure include: the vehicle parking status in the battery swapping platform can be automatically detected by applying a deep learning-based model, thereby preventing misoperation of the vehicle and thus preventing damage to the battery. Attached Figure Description

[0031] The invention will now be described in more detail with reference to the accompanying drawings, in which:

[0032] Figure 1 A flowchart illustrating a method for detecting vehicle parking status in a battery swapping platform according to the present invention is shown.

[0033] Figure 2 The image shown is a calibration image when the panoramic camera is configured as a fisheye camera;

[0034] Figure 3 Showing according to Figure 2 The outline image of the wheel fixing device of the battery swapping platform in the calibration image;

[0035] Figure 4 The relative positions of the V-shaped roller groove and the wheel according to the present invention are illustrated in a simplified diagram;

[0036] Figures 5a to 5c The output image is shown when the annotation information is constructed;

[0037] Figure 6 A schematic diagram of the structure of the convolutional neural network model according to the present invention is shown;

[0038] Figure 7 The line graph of the Fish activation function of the model is shown;

[0039] Figure 8 A schematic diagram of the system according to the present invention is shown;

[0040] Figure 9A schematic diagram showing the battery swapping platform in the correct parking position of the vehicle. Detailed Implementation

[0041] It is readily understood that, based on the technical solution of this invention, those skilled in the art can propose various interchangeable structural methods and implementations without altering the essential spirit of the invention. Therefore, the following detailed embodiments and accompanying drawings are merely illustrative examples of the technical solution of this invention and should not be considered as the entirety of the invention or as limitations or restrictions on the technical solution of this invention.

[0042] The directional terms such as up, down, left, right, front, back, front, back, top, and bottom mentioned or possibly used in this specification are defined relative to the structures shown in the accompanying drawings. They are relative concepts and may therefore vary depending on their location and usage. Therefore, these or other directional terms should not be interpreted as restrictive.

[0043] The battery swapping platform mentioned here should be understood in two ways. First, it refers to a battery swapping platform or device used for replacing vehicle power batteries, such as a fixed, movable, or foldable platform or device. Second, it can also be understood as a building-type battery swapping station isolated from the outside world. Furthermore, the term "battery swapping platform" can also refer to a charging / swapping platform or charging / swapping station.

[0044] Figure 1 A block diagram illustrates a method for detecting vehicle parking status in a battery swapping platform according to the present invention, the method comprising the following steps:

[0045] a) Using a panoramic camera device to acquire images of the battery swapping platform, on which vehicles are parked;

[0046] b): Input the image into the model, wherein the model is constructed based on multiple sets of training data containing sample images and annotation information, the annotation information depicting whether the vehicle has reached a preset location on the battery swapping platform in the sample images; and

[0047] c): Based on the model, the image is processed to determine whether the vehicle has parked in the preset position of the battery swapping platform.

[0048] It should be noted that the step names mentioned above (and below) are only used to distinguish between steps and facilitate their reference, and do not represent the order of the steps. The flowcharts in the accompanying diagrams are merely examples of how to execute this method. Unless there is a clear conflict, the steps can be executed in various orders or simultaneously.

[0049] Overall, based on the trained model, it is possible to directly visually determine the parking status of vehicles on the battery swapping platform, thereby avoiding misoperations in the automatic or semi-automatic battery swapping process, such as starting battery swapping before the vehicle is parked or before the vehicle is stopped, and thus reducing the time required for the battery swapping process.

[0050] In step a), compared to partial images of the battery swapping platform and vehicles acquired in a common manner, acquiring panoramic images of the battery swapping platform and vehicles parked on it using a panoramic camera not only provides an overall understanding of the vehicle's status (e.g., confirmation of the vehicle's front and rear positions) but also provides a direct view of the vehicle's relationship to its surrounding environment. This offers the possibility of coordination with the main control system of the battery swapping platform, especially the battery swapping station. Furthermore, in step a), only a single panoramic camera can be installed, which essentially eliminates the need to alter the existing hardware of the battery swapping platform. Here, the panoramic camera can be, but is not limited to, a fisheye camera; it also includes, for example, a camera device with a wide-angle lens or other camera devices capable of acquiring panoramic images.

[0051] Next, we will elaborate on the model used in step b). Here, the trained model can be understood as any deep learning or machine learning-based model, such as a neural network model, especially a convolutional neural network model (CNN model) or a recurrent neural network model (RCNN model). Here, the convolutional neural network model can be implemented, for example, in the form of VGGNet, GoogleNet, ResNet, DenseNet, LeNet, etc.

[0052] It is known that before starting the actual detection process, the model needs to be pre-trained using multiple sets of training data, i.e., training is performed using a combination of multiple sample images and their associated annotation information. Optionally, according to this method, before acquiring training image data, the panoramic camera device used is first calibrated to determine whether it has imaging distortion. This can improve the efficiency of training and learning, and thus improve the efficiency of the subsequent detection process. The imaging distortion, for example, is caused by a change in the camera's mounting position. This calibration process can also occur after the model is built or before detection begins.

[0053] According to this method, by comparing the actual image of the battery swapping platform when the vehicle is not parked with the established calibration image, it is possible to infer whether there is imaging distortion in the panoramic camera device. The calibration image I... j Source image I of the battery swapping platform at the following time points sThe outline image I of the wheel fixing device of the battery swapping platform is constructed at the time point before the vehicle is parked on the platform and the panoramic camera is in the correct installation position. The outline image I of the wheel fixing device is saved as a layer, specifically as a PNG image. For example, the outline image I can be divided into a three-channel RGB image I. c A grayscale image with one channel (i.e., the alpha channel) I m Correspondingly, the source image I s The number of channels is, for example, 4 for a PNG image, where only its RGB channels are used when constructing the calibration image. Of course, the number of channels can also be set to 3, for example, for images in JPG or BMP format. Here, the source image I... s RGB image of outlined image I c and grayscale image I m They have the same width and height. Therefore, the construction of the calibration image can include the following steps:

[0054] -Transfer grayscale image I m The image is converted to a grayscale image with an alpha channel value of 0 or 1 through normalization (i.e., normalization, or simply the Norm function). First, the grayscale image I m The value of the Alpha channel is converted from the range of [0, 255] to the range of [0, 1]. Then, the parameter δ is set to finally convert the value of the Alpha channel to 0 or 1.

[0055] -Based on source image I s Normalized grayscale image of the outlined image and RGB image I c To construct calibration image I j .

[0056] Based on this, the construction of the calibration image can also be represented by the following formula:

[0057]

[0058] It should be noted that the construction of this calibration image is not limited to the method mentioned above; it can also be achieved through other means, such as using a pre-defined calibration image. By constructing a calibration image, the effects of imaging distortion caused by internal or external factors during future use of the panoramic camera are prevented, thereby preventing judgment errors and improving its maintainability.

[0059] For example, when the panoramic camera device is constructed as a fisheye camera, the calibration image constructed according to the above method can be as... Figure 2As shown, the outlined image of the wheel fixing device of the battery swapping platform, which will be explained in more detail below, is shown in... Figure 3 This will be explained again in the text.

[0060] The construction of training data includes the creation of sample images and the generation of annotation information. Image acquisition for constructing sample images includes: image acquisition before model building (also known as the initial image acquisition stage) and image acquisition after model building (also known as the iterative image acquisition stage). In the initial image acquisition stage, image data within the acquisition range is recorded, and the recorded video files are saved frame by frame. However, in the iterative image acquisition stage, the built model filters and performs inference operations on the recorded images, outputting images within a preset range. Here, in the iterative data acquisition stage, the model's efficiency in filtering image data increases over time and as the model improves, resulting in increasingly targeted images, thereby gradually improving the model's performance and prediction accuracy.

[0061] Considering the size of the panoramic image and the requirements of the model's input parameters, the acquired image needs to be processed for ROI (Region of Interest) before being input into the model. For example, an ROI of a specified format can be extracted from the image selected during training. Here, the ROI is located at the center of the image and its format is represented as HWC, for example, an image with dimensions of 256×512×3, where H is the image height (in pixels), W is the image width (in pixels), and C is the number of channels. According to this method, to increase the robustness of the training data and simulate the tolerances generated during actual batch installation, the acquired ROI can have a certain deviation from the preset values ​​of the specified format (e.g., the width and height mentioned above), and the randomness of this deviation follows a normal distribution. Specifically, the acquired ROI has random deviations from the original size in the width direction (also referred to as the x-axis of the image coordinate system) and the height direction (also referred to as the y-axis of the image coordinate system). For example, it can have a random deviation of 2%, where the median μ = 0, and the variance in the x-axis direction is σ. x =2.25, variance σ in the y-axis direction y =3.37. The ROI-processed images are used as sample images for subsequent annotation and model training. Additionally, the ROI processing mentioned here can also be performed at the start of the actual detection process, i.e., before step b), on the images acquired in step a).

[0062] Then, the collected sample images can be labeled with the help of a binary classifier. If the wheel has reached the preset position of the battery swapping platform, the label information "1" is given; otherwise, the label information "0" is given. This labeling process can be completed with the help of a commonly used binary classifier.

[0063] Optionally, according to this method, if the center region of the wheel in the acquired sample image has reached the preset position of the wheel fixing device of the battery swapping platform, then a label "1" is given; otherwise, a label "0" is given. Optionally, the wheel fixing device can be optionally constructed as a V-shaped roller groove, such as... Figure 4 As shown, a cylindrical object can slide to the bottom of the V-shaped roller groove due to the influence of gravity. Similarly, when the wheel reaches the preset position precisely, the central area of ​​the wheel, especially the center of the hub, can be projected onto the central axis of the V-shaped roller groove. If this is the case, a label "1" is given; otherwise, a label "0" is given.

[0064] Furthermore, according to this method, an output image for the acquired sample images is obtained by generating an image and a mask tensor through a drawing function, similar to how calibration images are constructed, to simplify the image annotation process. The output image includes the contact tangent between the wheel and the V-shaped roller groove, and a contact normal perpendicular to it, particularly vertical in the image. The contact tangent and contact normal are determined by the two end tangent points of the wheel and the V-shaped roller groove. More precisely, these end tangent points are the intersections of the boundary lines of the wheel and the V-shaped roller groove, and can be imaginary or actual end tangent points. (As in...) Figure 5a As shown, when the wheel is not in contact with the V-shaped roller groove, the contact tangent and the vertical contact normal pass through two imaginary end points of tangency. In this case, the contact tangent is perpendicular to the central axis of the V-shaped roller groove. The annotation "0" is given in this case, and this image can be set up not for training the model, which will be discussed later.

[0065] In contrast, as in Figure 5b and 5c As shown, when the wheel reaches contact with the V-shaped roller groove, that is, when there are two actual end tangent points, the contact tangent line passes through these two end tangent points, and the contact normal extends vertically upward from each end tangent point. Here, if the center area of ​​the wheel is located between these two contact normals, for example, at the center position between these two contact normals, then the label information "1" is given; otherwise, the label information "0" is given.

[0066] Additionally, when the vehicle is not fully in contact with the V-shaped roller groove, meaning there is only one actual end tangent point, the contact tangent line passes through this end tangent point and is perpendicular to the central axis of the V-shaped roller groove. Furthermore, two contact normal lines pass through the actual end tangent point and an imaginary end tangent point determined by this contact tangent line, respectively. In this case, if the wheel center area is located between these two contact normal lines, a label "1" is given; otherwise, a label "0" is given.

[0067] The annotation information obtained by the above method can be saved as an annotation file in LBL format, and together with the corresponding sample images mentioned above, form multiple sets of training data, which can also be called training datasets. It should be noted that the pre-annotation of sample images is not limited to the methods mentioned above.

[0068] As mentioned above, the model used in this method can be constructed in the form of a convolutional neural network model, which has an input layer, a hidden layer, and an output layer. The hidden layer includes a feature filtering module and a Canoe module, such as... Figure 6 As shown. The advantages of using the Canoe module are: it implements an attention mechanism while minimizing computational load, improving the model's compatibility with complex scenes; it strengthens the guidance on the positional relationship between the wheel center and wheel fixing devices such as grooves, enhancing the model's targetedness and accuracy. The feature filtering module includes convolutional layers and max-pooling layers, such as 3×3 convolutional layers and 2×2 max-pooling layers, and the Fish activation function is applied in the feature filtering module:

[0069]

[0070] Where x is the feature matrix of the image.

[0071] like Figure 7 As shown, the Fish activation function used here has the following benefits:

[0072] - It has similar function properties to the commonly used Mish activation function and has the ability to provide gradients near the zero point;

[0073] - Compared to the Mish activation function, it is simpler to calculate, reducing four exponential operations;

[0074] - In the positive data segment, a gradient of approximately 0.75 is provided, which is much gentler than the gradient of the Fish activation function used here, which provides a gradient of 1.0 compared to the Mish activation function.

[0075] In addition, other types of activation functions can certainly be used while ensuring the required computational speed and accuracy.

[0076] Furthermore, the feature filtering module can be implemented using a graphics processing unit (GPU) or a general-purpose processor (GPP). For example, in Figure 6 In one embodiment shown, an input image with dimensions of 256×512×3 can be converted into an image of 16×32×128 through four feature filtering steps and then input into the subsequent Canoe module.

[0077] According to this method, the Canoe module includes the following:

[0078] - The Reshape layer transforms the input image from the NHWC sorting format to the N1W(HC) format. This method can enhance the guidance on the positional relationship between the wheel center region and the wheel fixing device (e.g., the positional relationship between the wheel center region and the central axis of the V-shaped roller groove) and improve the model's judgment targeting and accuracy.

[0079] - Convolutional Layer: Uses 1×1 convolutional kernels for model computation to reduce the computational space of the model and to perform combined distillation of the original features. Its output layer number is... - The Sigmoid layer transforms image features into intensity signals ranging from 0 to 1, and then further distills the features using element-wise multiplication. Its output format is... The transformation function for Sigmoid is:

[0080]

[0081] - Flat layer, The format is eventually converted to Format;

[0082] - Fully Connected Layer: Performs operations on the fully connected layer and outputs a feature map of size N (256) to the output layer.

[0083] In the output layer, the Softmax function is used to output the judgment result as a probability value, and Softmax cross-entropy loss is employed. Alternatively, the output layer can directly output "1" or "0" to indicate whether the vehicle has correctly parked on the battery swapping platform, or other methods are also feasible.

[0084] The invention also includes a system 100 for detecting vehicle parking status in a battery swapping platform, which inFigure 8 The system is illustrated in a block diagram, which includes:

[0085] A panoramic camera device 110 is installed at the battery swapping platform and configured to acquire images of the battery swapping platform, which is used for parking vehicles.

[0086] The judgment device 120 is connected to the panoramic camera device 110 in a signal transmission manner and is configured to determine whether the vehicle is parked in a preset position on the battery swapping platform in the image from the panoramic camera device 110, and output the judgment result in the form of a signal; and

[0087] Triggering device 130, which is arranged in or at the battery swapping platform and connected to judgment device 120, can directly or indirectly trigger or stop the battery swapping process of the vehicle based on a signal issued by judgment device 120.

[0088] Based on the output signal of the judgment device 120, the triggering device 130 can directly or indirectly control the battery swapping process. For example, in a fully automated battery swapping platform, the triggering device can automatically activate or deactivate the connected device for implementing the battery swapping process, thus avoiding accidental operation of the vehicle when it is not accurately parked, which could damage the battery. The triggering device 130 can be configured as a switch to turn the device for implementing the battery swapping operation on or off, or as an alarm or similar device to provide visual or audible warnings to workers in a semi-automated battery swapping platform.

[0089] Additionally, the panoramic camera device 110 can optionally be configured as a fisheye camera, as described in the method according to the invention. This fisheye camera can be positioned next to the battery swapping platform and can acquire images of the platform, on which vehicles awaiting battery swapping operations are parked. The following section discusses... Figure 9 To elaborate further, the battery swapping platform includes a wheel fixing device constructed as a V-shaped roller chute. The V-shaped roller chute is as follows: Figure 4 As shown, it has two support plates at an angle, particularly an obtuse angle, to each other, comprising rows of rollers. When the vehicle is accurately parked, the central area of ​​the wheel, particularly the hub center, projects onto the central axis of the V-shaped roller groove. Figure 4 The center is perpendicular to the plane of the attached drawing.

[0090] According to the invention, the fisheye camera 110 is arranged such that it can be aligned with the central area of ​​the wheel and accurately and substantially without distortion display the wheel and / or the wheel fixing device of the battery swapping platform in the central area of ​​the image it captures. Furthermore, the system can also include two or more panoramic cameras arranged on both sides of the battery swapping platform, for example, one fisheye camera per wheel, to improve the accuracy of the system. Figure 9 In the embodiment shown, the system 100 includes two fisheye cameras respectively arranged next to the wheel fixing device, which are schematically shown by boxes for clarity.

[0091] Optionally, the system 100 further includes a calibration device 140 configured to detect imaging distortion of the panoramic camera device 110 based on a calibration image, the imaging distortion being caused by a change in the position of the panoramic camera device. Optionally, the calibration device 140 further includes a calibration image generator 141 configured to generate or store a calibration image thereon, with reference to the above explanation of the method according to the invention.

[0092] Optionally, the system 100 may also include a preprocessing unit 150 connected to the panoramic camera device 110, configured to extract an image of a predetermined size from the image from the panoramic camera device. For example, when constructing a sample image, the acquired image can be processed for a region of interest (ROI) and the sample image can be output. Additionally, during detection, the acquired image can be cropped to a predetermined size and output to the judgment device 120. Refer to the above explanation regarding the method according to the invention.

[0093] Optionally, the system 100 may also include a labeling information generator 160 connected to the preprocessing device 150, which is configured to determine whether the center area of ​​the wheel in the input sample image reaches the preset position of the wheel fixing device of the battery swapping platform and generate corresponding labeling information based on the determination result, and input the labeling information into the determination device 120.

[0094] Optionally, the annotation information generator 160 is configured to determine whether the projection of the wheel's central region lies on the central axis of the V-shaped roller groove, or to determine whether the wheel's central region lies between two vertical contact normals, which in the sample image pass through the two end points of tangency between the wheel and the V-shaped roller groove, respectively. Based on this determination result, annotation information corresponding to the sample image is generated. Please refer to the above explanation regarding the method according to the present invention.

[0095] Optionally, the determining device 120 includes a processor 121 on which a model is stored, the model being constructed based on a convolutional neural network model having an input layer, a hidden layer including a feature filtering module and a Canoe module, and an output layer, as explained above regarding the method according to the invention.

[0096] Finally, the present invention also relates to a battery swapping platform including such a system as to be connected to or integrated into the main control system of the battery swapping platform.

[0097] Overall, the present invention has the following advantages compared to the prior art:

[0098] 1. Inexpensive cameras can be used as the data acquisition source for panoramic camera devices;

[0099] 2. Installation is simple; once installed, simply deploy the algorithm model to gain algorithmic capabilities.

[0100] 3. Deep convolutional neural networks can be used to build models and achieve computer vision judgment. There is no need to locate and observe elements. The data can be directly input into the overall judgment algorithm logic to obtain the precise location information of the vehicle.

[0101] 4. There is no strong dependence on the hardware facilities involved. As long as the algorithm development and data collection standards described later are met, the training, development and implementation of the model can be realized.

[0102] 5. The model can be upgraded based on actual operating conditions to achieve better quality;

[0103] 6. Flexible installation: In real-world scenarios, only one location is needed to complete the judgment. For scenarios requiring higher accuracy, multiple sensor groups can be installed simultaneously with algorithms to obtain more reliable judgment outputs through a voting process.

[0104] 7. High accuracy and reliability, compatible with multiple data formats, including but not limited to (color images, infrared images, scanned images, point cloud images, etc.);

[0105] 8. It requires minimal changes to the hardware structure, is highly operable, and is suitable for large-scale batch applications.

[0106] It should be understood that all the above embodiments are exemplary and not restrictive, and any modifications or variations made by those skilled in the art to the specific embodiments described above under the concept of the present invention should be within the legal protection scope of the present invention.

Claims

1. A method for detecting vehicle parking status in a battery swapping platform, characterized in that, Includes the following steps: a): Using a panoramic camera device to acquire images of a battery swapping platform, wherein a vehicle is parked on the battery swapping platform and the battery swapping platform is used to replace the vehicle's power battery; b): Inputting the image into a model, wherein the model is constructed based on multiple sets of training data containing sample images and annotation information, the annotation information depicting whether the vehicle has reached a preset location on the battery swapping platform in the sample images; and c): Based on the model, the image is processed to determine whether the vehicle has parked in the preset position of the battery swapping platform. Specifically, it is determined whether the projection of the wheel center region in the sample image is located on the central axis of the wheel fixing device constructed as a V-shaped roller groove, or whether the wheel center region is located between two vertical contact normals. The two vertical contact normals pass through the two end tangent points of the wheel and the V-shaped roller groove in the sample image, respectively, to obtain the annotation information belonging to the sample image.

2. The method according to claim 1, characterized in that, A calibration image is constructed to detect whether the panoramic camera device has imaging distortion. The calibration image is based on the source image of the battery swapping platform when the vehicle is not parked and the outline image of the wheel fixing device of the battery swapping platform.

3. The method according to claim 2, characterized in that, The outlined image is converted into an RGB image and a grayscale image. The calibration image is constructed based on the normalized grayscale image of the RGB image and the grayscale image, wherein the value of the normalized grayscale image is 0 or 1.

4. The method according to claim 1, characterized in that, When constructing training data, the images of the battery swapping platform acquired by the panoramic camera device are processed by ROI to obtain sample images, wherein the height and / or width of the sample images have a deviation from a preset value that follows a normal distribution.

5. The method according to claim 1, characterized in that, The model is built based on a convolutional neural network model, which has an input layer, hidden layers including a feature filtering module and a Canoe module, and an output layer. The model is based on the Fish activation function.

6. The method according to claim 5, characterized in that, The Canoe module includes a Reshape layer, a convolutional layer, a Sigmoid layer, a flattening layer, and a fully connected layer. The Reshape layer converts the format of the image from the input layer from NHWC to N1W(HC).

7. A system for detecting the parking status of a vehicle, the system being usable in a battery swapping platform, characterized in that, The system includes: A panoramic camera device can be installed at a battery swapping platform. The panoramic camera device is configured to acquire images of the battery swapping platform, which is used to park vehicles and perform battery swapping operations. A judgment device is connected to the panoramic camera device. The judgment device is configured to process the image from the panoramic camera device based on a model and determine whether the vehicle has parked in a preset position on the battery swapping platform. The model is constructed based on multiple sets of training data containing sample images and annotation information. The annotation information is depicted in the sample images to indicate whether the vehicle has arrived at the preset position on the battery swapping platform. A triggering device, disposed within the battery swapping platform and connected to the judgment device, is configured to trigger or stop the battery swapping process for the vehicle based on a judgment result from the judgment device; and The annotation information generator is configured to determine whether the projection of the wheel center region in the sample image is located on the central axis of the wheel fixing device constructed as a V-shaped roller groove, or to determine whether the wheel center region is located between two vertical contact normals, which pass through the two end tangent points of the wheel and the V-shaped roller groove in the sample image, and to generate annotation information belonging to the sample image based on the determination result.

8. The system according to claim 7, characterized in that, The system has a calibration device configured to detect imaging distortion of the panoramic camera device based on a calibration image.

9. The system according to claim 8, characterized in that, The calibration device includes a calibration image generator configured to generate a calibration image based on a source image of the battery swapping platform when the vehicle is not parked and an outline image of the wheel fixing device of the battery swapping platform.

10. The system according to claim 7, characterized in that, The system has a preprocessing unit connected to the panoramic camera device, which is configured to perform ROI processing on the images of the battery swapping platform acquired by the panoramic camera device and output sample images, wherein the height and / or width of the sample images have a deviation from a preset value that follows a normal distribution.

11. The system according to claim 7, wherein the panoramic camera device is configured as a fisheye camera.

12. The system according to claim 11, characterized in that, The fisheye camera is arranged such that when the vehicle is parked in a preset position on the battery swapping platform, the fisheye camera is aligned with the center area of ​​the wheel.

13. A battery swapping platform, characterized in that, The battery swapping platform includes the system according to any one of claims 7 to 12.

Images