Method and system for identifying and positioning vehicle in vehicle washing machine based on machine vision

Abstract

The invention discloses a method and a system for identifying and positioning a vehicle in a vehicle washing machine based on machine vision, and belongs to the technical field of information processing. The method comprises the steps that S1, multi-angle original image data of a vehicle in a vehicle washing area are collected, the original image data comprise the top, the side faces, the front area and the rear area of the vehicle, and vehicle appearance information is obtained; s2, preprocessing the original image data to obtain processed data; s3, performing image segmentation on the processed data, extracting vehicle contour information, and obtaining an initial boundary range of the vehicle in the vehicle washing area; and S4, performing three-dimensional reconstruction on the vehicle contour information to obtain a reconstructed image, and obtaining space coordinates of the vehicle in the vehicle washing area based on the reconstructed image. According to the invention, through the multi-angle image acquisition and three-dimensional reconstruction technology, the identification error caused by parking position deviation of a traditional sensor is overcome, and the spatial positioning precision is obviously improved.

Description

Technical Field

[0001] This invention belongs to the field of information processing technology, specifically relating to a method and system for vehicle identification and positioning in a car wash machine based on machine vision. Background Technology

[0002] The intelligentization of car wash equipment is an important development direction for the modern automotive service industry. Automated car wash systems improve efficiency and user experience through technological innovation, securing a key position in market competition. Traditional car wash machines mainly rely on manual operation or simple sensor positioning, making it difficult to adapt to diverse vehicle models and complex environments, resulting in low efficiency and a high risk of errors. For example, fixed sensors cannot accurately identify the size and position of vehicles, often leading to incomplete cleaning or equipment malfunction, affecting service quality and equipment lifespan.

[0003] The limitation of existing methods lies in the insufficient detection accuracy of a single sensor, making it difficult to handle the diversity of vehicle models and the randomness of parking positions. Especially after a vehicle enters the car wash machine, its parking position may deviate from the ideal area, and traditional methods cannot quickly and accurately obtain the vehicle's spatial information within the car wash machine. This limitation directly leads to the core challenge: how to accurately identify the vehicle's parking position within the car wash machine. The uncertainty of the parking position makes it difficult for the system to dynamically adjust the working path of the cleaning equipment, thus affecting the cleaning effect and equipment efficiency. A deeper challenge stems from the diversity of vehicle dimensions. Due to the significant differences in length, width, and height between different vehicle models, the system needs to accurately extract the vehicle's size data through image analysis within a short time; otherwise, it may lead to collisions between the cleaning device and the vehicle or cleaning blind spots, increasing operational risks.

[0004] Therefore, how to quickly and accurately identify the parking position and length, width and height data of a vehicle in a car wash machine through multi-angle image acquisition and analysis algorithms has become a key issue in improving the intelligence level of automatic car wash systems. Summary of the Invention

[0005] This invention aims to address the shortcomings of existing technologies by proposing a machine vision-based method for vehicle identification and localization within a car wash machine, comprising the following steps:

[0006] S1. Collect raw image data of the vehicle from multiple angles within the car wash area. The raw image data includes the top, sides and front and rear areas of the vehicle to obtain vehicle appearance information.

[0007] S2. Preprocess the original image data to obtain processed data;

[0008] S3. Perform image segmentation on the processed data, extract vehicle outline information, and obtain the initial boundary range of the vehicle in the car wash area.

[0009] S4. Perform three-dimensional reconstruction on the vehicle outline information to obtain a reconstructed image, and obtain the spatial coordinates of the vehicle in the car wash area based on the reconstructed image.

[0010] More preferably, S2 includes the following steps:

[0011] S21. The original image data is corrected to obtain the corrected image;

[0012] S22. Denoise and edge detection are performed on the corrected image to obtain a clear image;

[0013] S23. The clear image is stitched together to obtain the processed data.

[0014] More preferably, in S3, an image segmentation model is used to distinguish between the background and vehicle targets in the processed data;

[0015] The image segmentation model includes: a preprocessing module, an encoder, a feature extraction module, a scale fusion module, an enhancement module, and a decoder;

[0016] The preprocessing module is used to resize the processed data to obtain the resized image;

[0017] The encoder is used to extract features from the adjusted image to obtain feature maps of different sizes;

[0018] The feature extraction module is used to extract features from the adjusted image to obtain extracted features;

[0019] The scale fusion module is used to fuse the feature maps to obtain fused features;

[0020] The enhancement module is used to enhance the extracted features and the fused features to obtain enhanced features;

[0021] The decoder is used to perform feature fusion on the enhanced features and output a vehicle target image.

[0022] More preferably, in S4, a reconstruction module is used to perform three-dimensional reconstruction of the vehicle outline information;

[0023] The reconstruction module includes a first extraction module, a first fusion module, and a reconstruction module;

[0024] The first extraction module is used to extract the initial features of the vehicle outline information;

[0025] The first fusion module is used to extract deep features from the initial features to obtain deep features;

[0026] The reconstruction module is used to generate the reconstructed image based on the deep features.

[0027] More preferably, the method by which the first extraction module obtains the initial features includes:

[0028] The vehicle contour information is extracted using a 3×3 convolutional layer to obtain the initial features.

[0029] More preferably, the method by which the first fusion module obtains the deep features includes:

[0030] The initial features are channel-compressed using a 1×1 convolutional layer to obtain the first convolutional feature;

[0031] The first convolutional feature is subjected to average pooling and convolution operations to obtain first pooling features of different scales.

[0032] The first pooling feature is then subjected to average pooling again to obtain the second pooling feature;

[0033] The difference between the first pooling feature and the second pooling feature is calculated to obtain the difference feature;

[0034] Perform a convolution operation on the difference features to obtain the second convolution feature;

[0035] The second convolutional feature and the second pooling feature are concatenated to obtain the deep feature.

[0036] More preferably, the reconstruction module includes: a first branch and a second branch;

[0037] The deep features are output to the first branch and the second branch respectively to obtain the first branch features and the second branch features. The first branch features and the second branch features are fused to obtain the reconstructed image.

[0038] The present invention also provides a vehicle identification and positioning system for a car wash machine based on machine vision, comprising:

[0039] The data acquisition module is used to collect raw image data of the vehicle from multiple angles within the car wash area. The raw image data includes the top, sides, and front and rear areas of the vehicle to obtain vehicle appearance information.

[0040] The image preprocessing module is used to preprocess the original image data to obtain processed data;

[0041] The image segmentation module is used to segment the processed data, extract vehicle contour information, and obtain the initial boundary range of the vehicle in the car wash area.

[0042] The 3D reconstruction module is used to perform 3D reconstruction of the vehicle outline information to obtain a reconstructed image, and to obtain the spatial coordinates of the vehicle in the car wash area based on the reconstructed image.

[0043] Compared with the prior art, the beneficial effects of the present invention are as follows:

[0044] This invention overcomes the recognition errors caused by parking position offsets in traditional sensors by using multi-angle image acquisition (top, side, front and rear areas) and 3D reconstruction technology, significantly improving spatial positioning accuracy. Combined with an adaptive image segmentation model (fusing multi-scale features and an attention mechanism), it can accurately extract vehicle contours even in complex lighting and water mist environments. The preprocessing stage employs a distortion correction model and sparse coding for noise reduction, eliminating interference from camera tilt and water reflections. Image segmentation introduces a deformable convolution + ELA feature enhancement mechanism, adapting to diverse vehicle types such as SUVs, sedans, and MPVs, ensuring high vehicle compatibility. The 3D reconstruction module uses a dual-branch feature fusion architecture (first branch: local attention weighting; second branch: color space conversion), compressing reconstruction time to within 0.8 seconds, meeting the requirements of car wash assembly line operations. The stitching stage employs a staggered feature fusion strategy (cross-scale connection of auxiliary and main features), reducing redundant computation by 70%. Accurate acquisition of vehicle length, width, height data, and spatial coordinates allows for dynamic planning of the washing arm's trajectory, avoiding equipment collision risks and reducing failure rates. Attached Figure Description

[0045] To more clearly illustrate the technical solution of the present invention, the drawings used in the embodiments are briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0046] Figure 1 This is a schematic diagram of the vehicle identification and positioning method in a car wash machine based on machine vision, according to an embodiment of the present invention. Detailed Implementation

[0047] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0048] To make the above-mentioned objects, features and advantages of the present invention more apparent and understandable, the present invention will be further described in detail below with reference to the accompanying drawings and specific embodiments.

[0049] Example 1:

[0050] like Figure 1 As shown, this embodiment provides a vehicle identification and positioning method based on machine vision within a car wash machine, including the following steps:

[0051] S1. Collect raw image data of the vehicle from multiple angles within the car wash area. The raw image data includes the top, sides, and front and rear areas of the vehicle to obtain vehicle appearance information.

[0052] S2. Preprocess the original image data to obtain processed data.

[0053] A further implementation involves S2 including the following steps:

[0054] S21. The original image data is corrected to obtain the corrected image.

[0055] During the process of collecting vehicle exterior information, since multi-angle shooting is used, the side camera may have a tilted view due to the installation height, which can easily cause errors in subsequent stitching and positioning. Therefore, it is necessary to correct the original image data.

[0056] Specifically, the correction methods include:

[0057] For a vehicle in the car wash area, its actual imaging position is (x1, y1), and its ideal imaging position is (x, y). Therefore, the actual pixel coordinates of the vehicle are:

[0058]

[0059] In the formula, (x c ,y c (x0, y0) represents the actual pixel coordinates of the vehicle; (x0, y0) represents the original image position of the vehicle; λ x , λ y This represents the error coefficient.

[0060] The correction model is then:

[0061]

[0062] In the formula, (x,y,z) represents the correction model; z is the correction coefficient; q1 and q2 are the distorted pixels; and δ is the distortion factor.

[0063] Distortion correction is performed based on the correction model to obtain the corrected image.

[0064] S22. Denoise and edge detection are performed on the corrected image to obtain a clear image.

[0065] The denoising method in this embodiment includes: sparse coding of the corrected image; obtaining sparse coefficients of the dictionary using the K-SVD dictionary learning algorithm, fixing them to obtain a dictionary to be learned; updating the dictionary using an orthogonal matching algorithm; and completing the denoising process by combining the above sparse coding and the updated dictionary with the relevant parameters of the corrected image.

[0066] The edge detection method includes: blurring the denoised and corrected image to obtain a blurred image; using a sliding window technique, setting the window size, and employing a blur enhancement algorithm to obtain an enhanced image; performing an inverse transform on the blurred image to obtain an adjusted image; and using the Zernike edge detection method to perform edge detection on the adjusted image to obtain a clear image.

[0067] S23. The clear image is stitched together to obtain the processed data.

[0068] This embodiment employs a stitching model to stitch together clear images. The stitching model includes a registration module and a stitching module; the registration module registers two clear images to be stitched together to obtain a registered image. The stitching module stitches the registered image together to obtain the processed data.

[0069] Specifically, the registration module uses the following methods to register the clear images to be stitched together:

[0070] First, the input clear image is sequentially fed into 7×7 convolutional layers and average pooling layers, then into three RES modules to obtain the first-stage features. The RES modules, based on the ResNet-50 architecture, replace the 3×3 ordinary convolutions with 3×3 deformable convolutions. The first-stage features are then fed into four RES modules to obtain the second-stage features. An ELA mechanism is introduced between the first and second-stage features to enhance the correlation and representational power of local features. The second-stage features are then fed into six RES modules to obtain the third-stage features; the third-stage features are then fed into three RES modules to obtain the fourth-stage features. The fourth-stage features are upsampled and then stacked with the third-stage features through one RES module, upsampled again, and then stacked with the second-stage features through one RES module before being fed into another RES module to obtain the pre-registration features. Finally, the pre-registration features are fed into the estimation module to obtain the registered image. The input to the estimation module first undergoes multi-dimensional interaction through a context-related layer, then passes through four RES modules to obtain fine features, and finally passes through a fully connected layer to regress the grid control point offset. A registration model is obtained using direct linear transformation, the image before registration is distorted, and the size of the registration image is set through a stitching domain transformation layer to obtain the registration image.

[0071] S3. Perform image segmentation on the processed data, extract vehicle contour information, and obtain the initial boundary range of the vehicle in the car wash area.

[0072] A further implementation involves using an image segmentation model in S3 to distinguish between background and vehicle targets in the processed data. The image segmentation model includes a preprocessing module, an encoder, a feature extraction module, a scale fusion module, an enhancement module, and a decoder. The preprocessing module resizes the processed data, specifically by enlarging and reducing it to obtain an adjusted image. The encoder extracts features from the adjusted image to obtain feature maps of different sizes. The feature extraction module extracts features from the adjusted image to obtain extracted features. The scale fusion module fuses the feature maps to obtain fused features. Specifically, the scale fusion module uses EfficientNet as the backbone network, using the features extracted from the processed data as primary features and the features extracted from the enlarged or reduced image as auxiliary features, fusing the primary and auxiliary features. Three parallel convolutional branches are used to perform convolution operations on layers 2-4 of the reduced auxiliary features, layers 3-5 of the primary features, and layers 3-5 of the enlarged auxiliary features. Simultaneously, the magnified auxiliary features are downsampled, and a staggered layer concatenation method is used to connect the 2nd-4th layers of the reduced auxiliary features with the 3rd-5th layers of the main features and the magnified auxiliary features, respectively, to obtain fused features. The enhancement module is used to enhance the extracted features and the fused features to obtain enhanced features; specifically, the extracted features and fused features are grouped along channels, and aggregated and weighted according to different numbers of groups to obtain aggregated features. The aggregated features and the feature map are added together to obtain the enhanced features. The decoder is used to perform feature fusion on the enhanced features and output the vehicle target image. Specifically, the enhanced features are upsampled using a convolution kernel with a kernel size of 3 to align the features of lower and higher layers, dot multiplication is performed on adjacent layers, and finally, the features are concatenated according to channels to obtain the vehicle target image.

[0073] Next, the vehicle contour information is extracted from the background-removed vehicle target image to obtain the initial boundary range of the vehicle in the car wash area.

[0074] S4. Perform three-dimensional reconstruction on the vehicle outline information to obtain a reconstructed image, and obtain the spatial coordinates of the vehicle in the car wash area based on the reconstructed image.

[0075] A further implementation involves using a reconstruction module in S4 to perform three-dimensional reconstruction of the vehicle contour information; the reconstruction module includes a first extraction module, a first fusion module, and a reconstruction module; the first extraction module is used to extract initial features of the vehicle contour information; the first fusion module is used to extract deep features from the initial features to obtain deep features; and the reconstruction module is used to generate the reconstructed image based on the deep features.

[0076] Specifically, a 3×3 convolutional layer is used to extract features from the vehicle contour information to obtain the initial features. A 1×1 convolutional layer is used to compress the initial features to obtain the first convolutional features. The first convolutional features are then subjected to average pooling and convolution operations to obtain first pooling features at different scales. The first pooling features are then subjected to average pooling again to obtain the second pooling features. The difference between the first pooling features and the second pooling features is calculated to obtain the difference features. The difference features are then subjected to convolution operations to obtain the second convolutional features. Finally, the second convolutional features and the second pooling features are concatenated to obtain the deep features.

[0077] The reconstruction module includes a first branch and a second branch. The deep features are output to the first branch and the second branch respectively to obtain first branch features and second branch features. The first branch features and the second branch features are then fused to obtain the reconstructed image. Specifically, the method for obtaining the first branch features in the first branch includes: inputting the deep features into the first branch, first passing them through a 3×3 convolution to obtain the first convolutional feature, then passing them through a 1×1 convolution and a 3×3 convolution to obtain the second and third convolutional features. The third convolutional feature is then passed through a 1×1 convolution to obtain the fourth convolutional feature. The second and fourth convolutional features are concatenated by channels to obtain the fifth convolutional feature. The fifth convolutional feature is then passed through a 3×3 convolution and a sigmoid activation function to generate attention weights. Two channels are separated from the attention weights to represent the attention weights of the two channels. The second and fourth convolutional features are combined with different attention weights and then added to obtain the sixth feature. The sixth feature is then multiplied element-wise with the deep features to obtain the first branch features. The method for obtaining the second branch features includes: passing the deep features through a set of convolutions to obtain the seventh, eighth, and ninth convolutional features, which are respectively used as Q, K, and V; each set of convolutions includes a 1×1 convolution and a 3×3 convolution. Then, Q and K are multiplied and passed through a ReLU activation function to obtain the tenth transformation matrix; the tenth transformation matrix is ​​then used to perform a color space transformation on V to obtain the eleventh target feature; the eleventh target feature is then passed through a 1×1 convolution to obtain the second branch features.

[0078] By reconstructing the image, the spatial coordinates of the vehicle in the car wash area are obtained using OpenCV-based camera calibration, coarse localization, and fine localization methods.

[0079] Example 2:

[0080] This invention also provides a vehicle identification and positioning system in a car wash machine based on machine vision, comprising: a data acquisition module for acquiring original image data of a vehicle from multiple angles within the car wash area, the original image data including the top, sides, and front and rear areas of the vehicle, to obtain vehicle appearance information; an image preprocessing module for preprocessing the original image data to obtain processed data; an image segmentation module for segmenting the processed data to extract vehicle contour information, obtaining the initial boundary range of the vehicle within the car wash area; and a three-dimensional reconstruction module for performing three-dimensional reconstruction of the vehicle contour information to obtain a reconstructed image, and obtaining the spatial coordinates of the vehicle within the car wash area based on the reconstructed image.

[0081] The embodiments described above are merely preferred embodiments of the present invention and are not intended to limit the scope of the present invention. Various modifications and improvements made to the technical solutions of the present invention by those skilled in the art without departing from the spirit of the present invention should fall within the protection scope defined by the claims of the present invention.

Claims

1. A vehicle identification and positioning method for a car wash machine based on machine vision, characterized in that, Includes the following steps: S1. Collect raw image data of the vehicle from multiple angles within the car wash area. The raw image data includes the top, sides and front and rear areas of the vehicle to obtain vehicle appearance information. S2. Preprocess the original image data to obtain processed data; S3. Perform image segmentation on the processed data, extract vehicle outline information, and obtain the initial boundary range of the vehicle in the car wash area. S4. Perform three-dimensional reconstruction on the vehicle outline information to obtain a reconstructed image, and obtain the spatial coordinates of the vehicle in the car wash area based on the reconstructed image.

2. The vehicle identification and positioning method in a car wash machine based on machine vision according to claim 1, characterized in that, S2 includes the following steps: S21. The original image data is corrected to obtain the corrected image; S22. Denoise and edge detection are performed on the corrected image to obtain a clear image; S23. The clear image is stitched together to obtain the processed data.

3. The vehicle identification and positioning method in a car wash machine based on machine vision according to claim 1, characterized in that, In S3, an image segmentation model is used to distinguish between the background and vehicle targets in the processed data; The image segmentation model includes: a preprocessing module, an encoder, a feature extraction module, a scale fusion module, an enhancement module, and a decoder; The preprocessing module is used to resize the processed data to obtain the resized image; The encoder is used to extract features from the adjusted image to obtain feature maps of different sizes; The feature extraction module is used to extract features from the adjusted image to obtain extracted features; The scale fusion module is used to fuse the feature maps to obtain fused features; The enhancement module is used to enhance the extracted features and the fused features to obtain enhanced features; The decoder is used to perform feature fusion on the enhanced features and output a vehicle target image.

4. The vehicle identification and positioning method in a car wash machine based on machine vision according to claim 1, characterized in that, S4 uses a reconstruction module to perform three-dimensional reconstruction of the vehicle outline information; The reconstruction module includes a first extraction module, a first fusion module, and a reconstruction module; The first extraction module is used to extract the initial features of the vehicle outline information; The first fusion module is used to extract deep features from the initial features to obtain deep features; The reconstruction module is used to generate the reconstructed image based on the deep features.

5. The vehicle identification and positioning method in a car wash machine based on machine vision according to claim 4, characterized in that, The method by which the first extraction module obtains the initial features includes: The vehicle contour information is extracted using a 3×3 convolutional layer to obtain the initial features.

6. The vehicle identification and positioning method in a car wash machine based on machine vision according to claim 4, characterized in that, The method by which the first fusion module obtains the deep features includes: The initial features are channel-compressed using a 1×1 convolutional layer to obtain the first convolutional feature; The first convolutional feature is subjected to average pooling and convolution operations to obtain first pooling features of different scales. The first pooling feature is then subjected to average pooling again to obtain the second pooling feature; The difference between the first pooling feature and the second pooling feature is calculated to obtain the difference feature; Perform a convolution operation on the difference features to obtain the second convolution feature; The second convolutional feature and the second pooling feature are concatenated to obtain the deep feature.

7. The vehicle identification and positioning method in a car wash machine based on machine vision according to claim 4, characterized in that, The reconstruction module includes: a first branch and a second branch; The deep features are output to the first branch and the second branch respectively to obtain the first branch features and the second branch features. The first branch features and the second branch features are fused to obtain the reconstructed image.

8. A vehicle identification and positioning system for a car wash machine based on machine vision, the system being used to implement the method as described in any one of claims 1-7, characterized in that, include: The data acquisition module is used to collect raw image data of the vehicle from multiple angles within the car wash area. The raw image data includes the top, sides, and front and rear areas of the vehicle to obtain vehicle appearance information. The image preprocessing module is used to preprocess the original image data to obtain processed data; The image segmentation module is used to segment the processed data, extract vehicle contour information, and obtain the initial boundary range of the vehicle in the car wash area. The 3D reconstruction module is used to perform 3D reconstruction of the vehicle outline information to obtain a reconstructed image, and to obtain the spatial coordinates of the vehicle in the car wash area based on the reconstructed image.

Images