Strip-shaped object recognition method and device for parking and computer system
By using bar convolution kernels in image information, the problem of low accuracy in recognition of long strip objects in the prior art is solved, and a more accurate length Strip object recognition.
Patent Information
- Application Number
- CN202411545549.9
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-10-31
- Publication Date
- 2025-02-14
AI Technical Summary
The existing intelligent driving system cannot accurately identify long objects in the parking lot, such as curbs and pillars, resulting in easy collisions when parking automatically.
The first and second convolution kernels are used to extract features from the vertical and horizontal directions of the image information respectively, and feature fusion is performed to more accurately identify the strip-like object.
By fusing vertical and horizontal features, the extracted features are more comprehensive, so that the characteristics of long strip objects can be learned more accurately and the accuracy of identification results can be improved.
Smart Images
Figure CN119445535A_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the field of unmanned driving technology, and in particular to a method, device, computer system, computer-readable storage medium and computer program product for recognizing long strip objects for parking. Background Art
[0002] As a hot emerging technology, driverless technology has attracted more and more attention. Among them, the most popular in the past two years is the integrated driving and parking technology. As a part of the integrated driving and parking technology, parking plays a decisive role in the field of intelligent driving.
[0003] When parking, the parking lot environment is relatively complex. In order to park safely, the intelligent driving system needs to segment the objects around the parking lot, and then perform intelligent parking based on the segmentation results. However, there are many long objects in the parking lot, such as curbs and pillars, and the existing intelligent driving system cannot extract the accurate features of these objects, so the recognition effect of these objects is poor, which can easily cause the vehicle to collide during automatic parking. Therefore, an object recognition method with higher recognition accuracy is urgently needed. Summary of the invention
[0004] The present invention aims to solve at least one of the technical problems existing in the prior art. To this end, the present invention provides a method, device, computer system, computer readable storage medium and computer program product for recognizing long strip objects for parking to solve the problem of low accuracy in recognizing long strip objects.
[0005] In a first aspect, the present invention provides a method for identifying a long strip object for parking, the method comprising:
[0006] Acquire image information of the parking area;
[0007] Using a feature extraction layer of an object recognition model, a first strip convolution kernel and a second strip convolution kernel are used to extract features from the vertical direction and the horizontal direction of the image information respectively, so as to obtain a first feature and a second feature of the image information;
[0008] Performing feature fusion on the first feature and the second feature to obtain a fused feature;
[0009] According to the fusion feature, the long strip object in the image information is identified.
[0010] The above-mentioned method for recognizing long strip objects for parking, when extracting features of image information, extracts the first feature and the second feature of the image information from the vertical direction and the horizontal direction respectively, and then fuses the first feature and the second feature to obtain the fused feature. That is to say, the fused feature in the above-mentioned embodiment is more comprehensive than the features directly extracted from the image information in the traditional technology, so that the features of the long strip objects can be learned more accurately, and thus the recognition result of the long strip objects determined according to the fused feature is also more accurate.
[0011] In another embodiment, the feature extraction layer using the object recognition model uses a first strip convolution kernel and a second strip convolution kernel to extract features from the vertical direction and the horizontal direction of the image information respectively to obtain the first feature and the second feature of the image information, including:
[0012] Using a first strip-shaped convolution kernel to extract features from a vertical direction of the image information using a first feature extraction layer of the object recognition model to obtain a first feature of the image information;
[0013] The second feature extraction layer of the object recognition model uses a second strip convolution kernel to extract features from the horizontal direction of the image information to obtain the second feature of the image information.
[0014] In the above embodiment, the first feature extraction layer and the second feature extraction layer of the object recognition model are used to extract the first feature and the second feature in different directions of the image information respectively, thereby ensuring the comprehensiveness of the extracted image information features, thereby being able to more accurately learn the features of the elongated object.
[0015] In another embodiment, the step of fusing the first feature and the second feature to obtain a fused feature includes:
[0016] Performing image interpolation processing on the first feature and the second feature respectively so that the vector dimensions of the first feature and the second feature are consistent;
[0017] A sum operation is performed on the first feature and the second feature having the same vector dimension to obtain the fused feature.
[0018] In the above embodiment, the first feature and the second feature are first subjected to image interpolation processing so that the vector dimensions of the first feature and the second feature are unified, and then the first feature and the second feature with unified vector dimensions are summed to obtain a fused feature. When an object recognition model is subsequently used to perform object recognition based on the fused feature, the recognition accuracy of the model can be improved, while the robustness of the model can be improved and the risk of overfitting can be reduced.
[0019] In another embodiment, the identifying the long strip object in the image information according to the fusion feature includes:
[0020] Using a preset convolution kernel to perform convolution processing on the fusion feature, changing the number of channels of the fusion feature, and obtaining a target fusion feature; the number of channels of the target fusion feature is the same as the number of channels of the image information;
[0021] The long strip object in the image information is identified according to the target fusion feature.
[0022] In the above embodiment, the fusion feature is convolved with a preset convolution kernel to change the number of channels of the fusion feature, thereby obtaining a target fusion feature having the same number of channels as the number of channels of the image information, so as to facilitate subsequent calculations.
[0023] In another embodiment, the identifying the long strip object in the image information according to the target fusion feature includes:
[0024] Calculating the target fusion feature using the activation function in the object recognition model to obtain the weights corresponding to each channel of the fusion feature;
[0025] A product operation is performed on the weight and the image information to obtain a target feature, and a recognition result of the long strip object in the image information is obtained based on the target feature.
[0026] In the above embodiment, the weight corresponding to the target fusion feature is calculated, and then the weight is multiplied by the original input image information to obtain the final target feature, and the target feature is involved in the subsequent recognition of long strip objects. Since the fusion feature integrates the features of the image in the horizontal and vertical directions, it can better learn the features of the long strip objects, making the recognition result more accurate.
[0027] In another embodiment, the weight ranges from 0 to 1.
[0028] In a second aspect, the present invention provides a strip-shaped object recognition device for parking, the device comprising:
[0029] An information acquisition module, used to acquire image information of a parking area;
[0030] A feature extraction module, configured to use a feature extraction layer of an object recognition model to extract features from the vertical direction and the horizontal direction of the image information using a first strip convolution kernel and a second strip convolution kernel, respectively, to obtain a first feature and a second feature of the image information;
[0031] A feature fusion module, used for fusing the first feature and the second feature to obtain a fused feature;
[0032] The object recognition module is used to recognize the long strip object in the image information according to the fusion feature.
[0033] In a third aspect, the present invention provides a computer system comprising a memory and a processor, wherein the memory stores a computer program, and the processor implements the steps of the method described in any one of the first aspects when executing the computer program.
[0034] In a fourth aspect, the present invention provides a computer-readable storage medium having a computer program stored thereon, wherein the computer program, when executed by a processor, implements the steps of the method described in any one of the first aspects.
[0035] In a fifth aspect, the present invention provides a computer program product, comprising a computer program, which, when executed by a processor, implements the steps of the method described in any one of the first aspects.
[0036] One or more of the above embodiments of the present invention have at least one or more of the following beneficial effects:
[0037] When extracting features of image information, a first feature and a second feature of the image information are extracted respectively, the first feature includes features in the vertical direction of the image information, and the second feature includes features in the horizontal direction of the image information. The first feature and the second feature are then fused to obtain a fused feature. That is to say, the fused feature in the above embodiment, compared to the traditional technology of directly extracting a single feature from the image information, the feature in the present application includes features extracted from the vertical and horizontal directions of the image information, that is, the extracted features are more comprehensive, and therefore the fused features obtained based on the fusion of these features are more comprehensive, so that the features of the long object can be learned more accurately, thereby making the recognition result of the long object determined based on the fused features more accurate.
[0038] Additional aspects and advantages of the present invention will be given in part in the following description and in part will be obvious from the following description, or will be learned through practice of the present invention. BRIEF DESCRIPTION OF THE DRAWINGS
[0039] The disclosure of the present invention will become more easily understood with reference to the accompanying drawings. It is easy for those skilled in the art to understand that these drawings are only for illustrative purposes and are not intended to limit the scope of protection of the present invention. In addition, similar numbers in the figures are used to represent similar components, among which:
[0040] Figure 1 is a flow chart of a method for recognizing a long strip object for parking in an embodiment of the present application;
[0041] Figure 2 is a schematic diagram of the structure of an object recognition model in an embodiment of the present application;
[0042] Figure 3a and Figure 3b They are schematic diagrams of the first row convolution kernel and the second row convolution kernel in the embodiment of the present application respectively;
[0043] Figure 4 is a schematic diagram of the structure of a strip-shaped object recognition device for parking in an embodiment of the present application;
[0044] Figure 5 It is a schematic diagram of the structure of the computer system in the embodiment of the present application. DETAILED DESCRIPTION
[0045] Some embodiments of the present invention are described below with reference to the accompanying drawings. It should be understood by those skilled in the art that these embodiments are only used to explain the technical principles of the present invention and are not intended to limit the protection scope of the present invention.
[0046] In existing operations, square convolution kernels (such as 1*1, 3*3 or 5*5) are used for convolution when extracting features, which is more suitable for the recognition of square objects. However, for long, narrow objects, when the feature map obtained by extracting features using the square kernel is used for object recognition, some irrelevant features will inevitably be attributed to the narrow object, resulting in inaccurate recognition results. Based on this, the present application proposes a new method for recognizing long, narrow objects for parking. When extracting features of image information, the first and second features are extracted from the vertical and horizontal directions of the image information using the first strip convolution kernel and the second strip convolution kernel, respectively, and then the first and second features are fused to obtain a fused feature. That is to say, the fused feature obtained in the present application, compared with the traditional technology of extracting a single feature from the image information using the square kernel, the feature in the present application includes the vertical and horizontal features in the image information, that is, the extracted features are more comprehensive, so that the features of the long, narrow objects can be learned more accurately, and the recognition results of the long, narrow objects determined according to the fused features are also more accurate.
[0047] The present invention will be described in detail below through specific embodiments.
[0048] In some embodiments, Figure 1 A method for object recognition is provided, comprising the following steps:
[0049] S100: Acquire image information of a parking area.
[0050] There may be some objects to be identified in the image information. When the method is applied to the intelligent parking system, the objects to be identified in the image information may be long strips of objects in the parking lot, such as curbs (or roadside kerbs), pillars, and other objects.
[0051] In some specific embodiments, the parking area may be photographed by a camera and the obtained initial image may be used as the image information of the parking area. Alternatively, some preprocessing operations may be performed on the initial image and the image after the preprocessing operations may be used as the image information of the parking area.
[0052] It should be noted that the preprocessing operation is not specifically limited in the embodiments of the present application. Without violating the inventive concept of the present application, the preprocessing may include one or a combination of operations such as digitization, geometric transformation, normalization, smoothing, restoration and enhancement. Among them, preprocessing the initial image can eliminate irrelevant information in the image, restore useful real information, enhance the detectability of relevant information and simplify the data to the maximum extent, thereby improving the reliability of subsequent feature extraction.
[0053] S200, utilizing a feature extraction layer of an object recognition model to employ a first strip convolution kernel and a second strip convolution kernel to extract features from the vertical direction and the horizontal direction of the image information respectively, to obtain a first feature and a second feature of the image information.
[0054] As mentioned above, since the existing feature extraction method cannot extract accurate features of long objects such as curbs and pillars, the recognition effect of these objects is poor, which easily causes the vehicle to collide during automatic parking. The convolution kernel used in the convolution layer in other schemes is a square kernel (such as 3*3), which is more suitable for the recognition of square objects. For such long and narrow objects, when the feature map obtained by feature extraction using the square kernel is used for object recognition, some irrelevant features will inevitably be attributed to the narrow object, resulting in inaccurate recognition results. In view of the above problems, in the embodiment of the present application, the feature extraction layer used for feature extraction in the object recognition model is improved, and the square kernel is not used for feature extraction, but the first strip kernel and the second strip kernel are used to extract features from the vertical direction and the horizontal direction of the image information respectively, to obtain the first feature and the second feature, so that the extracted features are more comprehensive, so that the features of the long object can be learned more accurately, and then the recognition result of the long object determined by the fusion feature is also more accurate.
[0055] In some embodiments, the above first feature is a feature extracted from the vertical direction of the image information, and the second feature is a feature extracted from the horizontal direction of the image information. This is only an exemplary and non-restrictive description. Without violating the inventive concept of the present application, the first feature and the second feature can also be features extracted from other directions of the image information, which will not be described one by one here.
[0056] The first feature and the second feature include but are not limited to one or more of the following features extracted from the image information:
[0057] 1. Color feature is a global feature that describes the surface properties of the scene corresponding to the image or image region. Common color feature extraction and matching methods include color histogram, color set, color aggregation vector and color correlation graph. In some specific embodiments, the first feature includes color features extracted from the image information along the vertical direction, and / or the second feature includes color features extracted from the image information along the horizontal direction.
[0058] 2. Texture feature is a global feature that describes the surface properties of the scene corresponding to the image or image area. Commonly used texture feature extraction methods can be roughly divided into five categories: statistical method, geometric method, model method, signal processing method and structural method. In some specific embodiments, the first feature includes texture features extracted from the image information along the vertical direction, and / or the second feature includes texture features extracted from the image information along the horizontal direction.
[0059] 3. Shape features are features that describe the shape of an object in an image. Common shape feature extraction methods include edge detection, Hough transform, etc. For example, the Canny operator is a commonly used edge detection operator that can detect edge points in an image and connect them into contours. In some specific embodiments, the first feature includes a shape feature extracted from the image information in a vertical direction, and / or the second feature includes a shape feature extracted from the image information in a horizontal direction.
[0060] 4. Spatial relationship features are features that describe the spatial relationship between objects in an image. For example, HoG (Histogram of Oriented Gradients) is a commonly used spatial relationship feature extraction method, which constructs features by statistically analyzing the gradient direction histogram of a local area of an image. In some specific embodiments, the first feature includes a spatial relationship feature extracted from the image information along the vertical direction, and / or the second feature includes a spatial relationship feature extracted from the image information along the horizontal direction.
[0061] 5. Local features are features extracted from local areas in an image. For example, SIFT (Scale-Invariant Feature Transform) feature is a commonly used local feature extraction method, which can extract features by detecting key points in an image and describing the local appearance around these key points. In some specific embodiments, the first feature includes local features extracted from the image information along the vertical direction, and / or the second feature includes local features extracted from the image information along the horizontal direction.
[0062] It should be noted that in the embodiments of the present application, the specific composition of the first feature and the second feature is not limited, and can be set according to actual needs without violating the inventive concept of the present application.
[0063] It should be noted that in the embodiments of the present application, the specific implementation of the object recognition model is not limited. Without violating the inventive concept of the present application, the object recognition model in the embodiments of the present application can be any one of a segmentation model or a recognition model based on a neural network. For example, the object recognition model can be a convolutional neural network (CNN) model, etc. Among them, a convolutional neural network (CNN) is a type of feedforward neural network (Feedforward Neural Networks) that includes convolution calculations and has a deep structure.
[0064] S300: Fusing the first feature and the second feature to obtain a fused feature.
[0065] Among them, in order to improve the recognition accuracy of the model and improve the robustness of the model, reduce the risk of overfitting, etc., in the embodiment of the present application, it is also necessary to fuse the first feature and the second feature extracted by the aforementioned steps, that is, to combine different features together.
[0066] Reference Figure 2 In some specific embodiments, a feature fusion layer may be added to the object recognition model, and the first feature and the second feature may be fused using the fusion function Concat in the feature fusion layer. It should be noted that in the embodiments of the present application, the specific implementation of the feature fusion layer and the fusion function used are not limited, and may be set according to actual needs without violating the inventive concept of the present application.
[0067] S400: Identify a long strip object in the image information according to the fusion feature.
[0068] Specifically, the long strip objects in the image information are identified based on the fusion features obtained in the previous steps. Since the fusion features include the vertical features and the horizontal features in the image information, the extracted features are more comprehensive, so that the features of the long strip objects can be learned more accurately, and the recognition results of the long strip objects determined by the fusion features are also more accurate.
[0069] In one embodiment, the object recognition model may be an object recognition model or an object segmentation model.
[0070] As a preferred implementation, in the embodiment of the present application, the feature extraction layer using the object recognition model uses a first strip convolution kernel and a second strip convolution kernel to extract features from the vertical direction and the horizontal direction of the image information respectively to obtain the first feature and the second feature of the image information, including:
[0071] Using a first strip-shaped convolution kernel to extract features from a vertical direction of the image information using a first feature extraction layer of the object recognition model to obtain a first feature of the image information;
[0072] The second feature extraction layer of the object recognition model uses a second strip convolution kernel to extract features from the horizontal direction of the image information to obtain the second feature of the image information.
[0073] In one embodiment, the height of the first strip convolution kernel is the same as the height of the image information; the width of the second strip convolution kernel is the same as the width of the image information. Thus, the first strip convolution kernel with the same height as the height of the image information can be used to extract the features of the image information in the vertical direction, and the second strip convolution kernel with the same width as the width of the image information can be used to extract the features of the image information in the horizontal direction.
[0074] Further references Figure 2 As shown, in some specific embodiments, the feature extraction layer in the object recognition model includes a first feature extraction layer and a second feature extraction layer, wherein the first feature extraction layer is used to extract the features in the vertical direction of the image information (i.e., the first feature), and the second feature extraction layer is used to extract the features in the horizontal direction of the image information (i.e., the second feature). In specific implementation, it is assumed that the input x (i.e., image information) size is C*H*W, where C represents the number of channels, H represents height, and W represents width. The first feature extraction layer is represented as conv(H, 1), and the second feature extraction layer is represented as conv(1, W). In some specific embodiments, the first strip convolution kernel can be H*1 (e.g., Figure 3a As shown), the second strip convolution kernel can be 1*W (as shown Figure 3b As shown in the figure, the first feature extraction layer uses a convolution kernel with a size of H*1 to extract features from the vertical direction of the image information (i.e., the first feature), and the second feature extraction layer uses a convolution kernel with a size of 1*W to extract features from the horizontal direction of the image information (i.e., the second feature). The first feature and the second feature are represented by R C×W and R C×H .
[0075] As a preferred implementation manner, in the embodiment of the present application, the first feature and the second feature are subjected to feature fusion to obtain a fused feature, including:
[0076] Performing image interpolation processing on the first feature and the second feature respectively so that the vector dimensions of the first feature and the second feature are consistent;
[0077] A sum operation is performed on the first feature and the second feature with the same vector dimension to obtain the fused feature.
[0078] In some specific embodiments, before the first feature and the second feature are fused, the vector dimensions of the first feature and the second feature may be unified first, so as to facilitate the subsequent fusion process. Specifically, the first feature and the second feature may be subjected to image interpolation processing to unify the vector dimensions of the first feature and the second feature. The first feature and the second feature after the vector dimensions are unified are respectively represented as r C×W and r C×H . Among them, the present application scheme does not limit the specific algorithm used for image interpolation processing. Under the premise of not violating the concept of the present application, it can be selected according to actual needs. For example, image interpolation processing can adopt nearest neighbor interpolation (Nearest-neighbor), bilinear interpolation (Bilinear), bicubic interpolation (bicubic), lanczos interpolation, edge-directed interpolation (Edge-directed interpolation), example-based interpolation, deep learning and other algorithms, which will not be repeated here.
[0079] After unifying the vector dimensions of the first feature and the second feature, the first feature and the second feature with the same vector dimensions are added to obtain the fusion feature, that is, y = r C×W +r C×H , then the size of y is r C×H×W .
[0080] As a preferred implementation, in the embodiment of the present application, the identifying the long strip object in the image information according to the fusion feature includes:
[0081] Using a preset convolution kernel to perform convolution processing on the fusion feature, changing the number of channels of the fusion feature, and obtaining a target fusion feature; the number of channels of the target fusion feature is the same as the number of channels of the image information;
[0082] The long strip object in the image information is identified according to the target fusion feature.
[0083] As mentioned above, since the image information in the embodiment of the present application can be the initial image of the parking area taken by the camera, or it can be the image obtained after some preprocessing operations are performed on the initial image, since the initial image is usually three-channel, and the image obtained after the preprocessing operation is usually multi-channel, and after the image information undergoes a series of operations such as feature extraction and feature fusion, the number of channels of the fused feature obtained may be greater than the number of channels of the image information, or it may be less than the number of channels of the image information, that is, the number of channels of the fused feature may be different from the number of channels of the image information, which will be detrimental to the subsequent product calculation with the image information. Therefore, further reference is made to Figure 2 As shown, in the present embodiment, after the first feature and the second feature are fused to obtain the fused feature, the fused feature can also be convolved using a preset convolution kernel to change the number of channels of the fused feature, thereby obtaining a target fused feature having the same number of channels as the number of channels of the image information, so as to facilitate subsequent calculations.
[0084] It should be noted that the implementation of this application does not specifically limit the preset convolution kernel, and it can be selected according to actual needs without violating the concept of this application.
[0085] In one embodiment, in a convolutional neural network, the number of channels generally includes the number of input channels and the number of output channels. The number of input channels for the initially input picture sample (i.e., image information) depends on the type of picture. For example, if it is a color picture, i.e., RGB type, the number of input channels is 3, and if it is gray, the number of input channels is 1. After the convolution is completed, the number of output channels depends on the number of convolution kernels, i.e., the number of output channels is related to the convolution kernel. For two or more layers of convolution, the number of input channels of each layer is the number of output channels of the previous layer.
[0086] As a preferred implementation, in the embodiment of the present application, the step of identifying the long strip object in the image information according to the target fusion feature includes:
[0087] Calculating the target fusion feature using the activation function in the object recognition model to obtain the weights corresponding to each channel of the fusion feature;
[0088] A product operation is performed on the weight and the image information to obtain a target feature, and a recognition result of the long strip object in the image information is obtained based on the target.
[0089] Further references Figure 2As shown, in the specific implementation, the activation function (sigmoid) of the object recognition model can be used to calculate the target fusion feature, obtain the weights corresponding to each channel of the fusion feature, and finally perform channel-by-channel multiplication operation on the weights corresponding to each channel of the fusion feature and each channel of the image information, so as to obtain a feature map with channel attention for the target feature, and finally obtain the recognition result of the long strip object in the image information based on the target feature.
[0090] In one embodiment, after obtaining the target features, the recognition result of the long strip object in the image information can be obtained based on the object recognition model or the object segmentation model. For example, the target features can be input into the middle layer of the backbone network of the object recognition model or the object segmentation model for further calculation to obtain the object recognition result.
[0091] In one embodiment, the object recognition model or object segmentation model may be a deeplab object recognition model or a unet object recognition model.
[0092] Among them, the activation function (sigmoid) is an S-shaped function commonly seen in biology, also known as an S-shaped growth curve. In information science, due to its monotonic and inverse monotonic properties, the Sigmoid function is often used as an activation function of a neural network to map variables between 0 and 1.
[0093] In the embodiment of the present application, the recognition result of the long strip object includes but is not limited to whether the object to be recognized in the image is a curb, a pillar, etc., which will not be elaborated here.
[0094] As a preferred implementation, in the embodiment of the present application, the value range of the weight is 0 to 1.
[0095] When the activation function (sigmoid) of the sigmoid layer in the object recognition model is used to calculate the target fusion feature, the weight value corresponding to the fusion feature can range from 0 to 1.
[0096] It should be understood that, although the steps in the flowcharts involved in the above embodiments are displayed in sequence according to the indication of the arrows, these steps are not necessarily executed in sequence according to the order indicated by the arrows. Unless there is a clear explanation in this article, the execution of these steps is not strictly limited in order, and these steps can be executed in other orders. Moreover, at least a part of the steps in the flowcharts involved in the above embodiments may include multiple steps or multiple stages, and these steps or stages are not necessarily executed at the same time, but can be executed at different times, and the execution order of these steps or stages is not necessarily carried out in sequence, but can be executed in turn or alternately with other steps or at least a part of the steps or stages in other steps.
[0097] Based on the same inventive concept, the embodiment of the present application also provides a long strip object recognition device for parking for implementing the long strip object recognition method for parking involved above. The implementation scheme for solving the problem provided by the device is similar to the implementation scheme recorded in the above method, so the specific limitations in one or more object recognition device embodiments provided below can refer to the limitations of the long strip object recognition method for parking above, and will not be repeated here.
[0098] In some specific embodiments, Figure 4 As shown, a strip-shaped object recognition device for parking is provided, comprising:
[0099] An information acquisition module 10 is used to acquire image information of a parking area;
[0100] A feature extraction module 20, configured to use a feature extraction layer of an object recognition model to extract features from the vertical direction and the horizontal direction of the image information using a first strip convolution kernel and a second strip convolution kernel, respectively, to obtain a first feature and a second feature of the image information;
[0101] A feature fusion module 30, configured to fuse the first feature and the second feature to obtain a fused feature;
[0102] The object recognition module 40 is used to recognize the long strip object in the image information according to the fusion feature.
[0103] In some specific embodiments, the feature extraction module 20 is specifically used to:
[0104] Using a first strip-shaped convolution kernel to extract features from a vertical direction of the image information using a first feature extraction layer of the object recognition model to obtain a first feature of the image information;
[0105] The second feature extraction layer of the object recognition model uses a second strip convolution kernel to extract features from the horizontal direction of the image information to obtain the second feature of the image information.
[0106] In some specific embodiments, the feature fusion module 30 is specifically used to:
[0107] Performing image interpolation processing on the first feature and the second feature respectively so that the vector dimensions of the first feature and the second feature are consistent;
[0108] A sum operation is performed on the first feature and the second feature with the same vector dimension to obtain the fused feature.
[0109] In some specific embodiments, the object recognition module 40 is further used to:
[0110] Using a preset convolution kernel to perform convolution processing on the fusion feature, changing the number of channels of the fusion feature, and obtaining a target fusion feature; the number of channels of the target fusion feature is the same as the number of channels of the image information;
[0111] The long strip object in the image information is identified according to the target fusion feature.
[0112] In some specific embodiments, the object recognition module 40 is specifically used to:
[0113] Calculating the target fusion feature using the activation function in the object recognition model to obtain the weights corresponding to each channel of the fusion feature;
[0114] A product operation is performed on the weight and the image information to obtain a target feature, and a recognition result of the long strip object in the image information is obtained based on the target feature.
[0115] In some specific embodiments, the weight ranges from 0 to 1.
[0116] Each module in the above-mentioned strip-shaped object recognition device for parking can be implemented in whole or in part by software, hardware, or a combination thereof. Each of the above-mentioned modules can be embedded in or independent of a processor in a computer system in the form of hardware, or can be stored in a memory in a computer system in the form of software, so that the processor can call and execute operations corresponding to each of the above modules.
[0117] In some specific embodiments, a computer system is provided, including a memory and a processor, wherein a computer program is stored in the memory, and when the processor executes the computer program, the method for recognizing a long strip object for parking as described in any one of the above items is implemented.
[0118] in, Figure 5 The computer system 1500 is shown as an example, and may include a processor 1510, a video display adapter 1511, a disk drive 1512, an input / output interface 1513, a network interface 1514, and a memory 1520. The processor 1510, the video display adapter 1511, the disk drive 1512, the input / output interface 1513, the network interface 1514, and the memory 1520 may be communicatively connected via a communication bus 1530.
[0119] Among them, the processor 1510 can be implemented by a general-purpose CPU (Central Processing Unit), a microprocessor, an application-specific integrated circuit (Application Specific Integrated Circuit, ASIC), or one or more integrated circuits, etc., to execute relevant programs to implement the technical solution provided by the present invention.
[0120] The memory 1520 can be implemented in the form of ROM (Read Only Memory), RAM (Random Access Memory), static storage device, dynamic storage device, etc. The memory 1520 can store an operating system 1521 for controlling the operation of the electronic device and a basic input and output system (BIOS) for controlling the low-level operation of the electronic device. In addition, a web browser 1523, a data storage management system 1524, and a device identification information processing system 1525, etc. can also be stored. The above-mentioned device identification information processing system 1525 can be an application program that specifically implements the operations of the aforementioned steps in the embodiment of the present invention. In short, when the technical solution provided by the present invention is implemented by software or firmware, the relevant program code is stored in the memory 1520 and is called and executed by the processor 1510.
[0121] The input / output interface 1513 is used to connect the input / output module to realize information input and output. The input / output module can be configured in the device as a component (not shown in the figure), or it can be externally connected to the device to provide corresponding functions. The input device may include a keyboard, a mouse, a touch screen, a microphone, various sensors, etc., and the output device may include a display, a speaker, a vibrator, an indicator light, etc.
[0122] The network interface 1514 is used to connect to a communication module (not shown) to realize communication interaction between the device and other devices. The communication module can realize communication through a wired mode (such as USB, network cable, etc.) or a wireless mode (such as mobile network, WIFI, Bluetooth, etc.).
[0123] The bus includes a path that transmits information between the various components of the device (eg, processor 1510, video display adapter 1511, disk drive 1512, input / output interface 1513, network interface 1514, and memory 1520).
[0124] In addition, the electronic device can also obtain information on specific collection conditions from the virtual resource object collection condition information database for use in condition judgment, etc.
[0125] It should be noted that although the above device only shows the processor 1510, the video display adapter 1511, the disk drive 1512, the input / output interface 1513, the network interface 1514, the memory 1520, the bus, etc., in the specific implementation process, the device may also include other components necessary for normal operation. In addition, it can be understood by those skilled in the art that the above device may also only include the components necessary for implementing the solution of the present invention, and does not necessarily include all the components shown in the figure.
[0126] In some specific embodiments, a computer-readable storage medium is provided, on which a computer program is stored. When the computer program is executed by a processor, the method for recognizing a long strip object for parking as described in any one of the above items is implemented.
[0127] In some specific embodiments, a computer program product is provided, including a computer program, which, when executed by a processor, implements the method for recognizing a long strip object for parking as described in any one of the above items.
[0128] In the description of this specification, the description with reference to the terms "one embodiment", "some embodiments", "examples", "specific examples", or "some examples" means that the specific features, structures, materials or characteristics described in conjunction with the embodiment or example are included in at least one embodiment or example of the present invention. In this specification, the schematic representation of the above terms does not necessarily refer to the same embodiment or example. Moreover, the specific features, structures, materials or characteristics described may be combined in any one or more embodiments or examples in a suitable manner.
[0129] In addition, the terms "first" and "second" are used for descriptive purposes only and should not be understood as indicating or implying relative importance or implicitly indicating the number of the indicated technical features. Therefore, the features defined as "first" and "second" may explicitly or implicitly include at least one of the features. In the description of the present invention, the meaning of "plurality" is at least two, such as two, three, etc., unless otherwise clearly and specifically defined.
[0130] Although the embodiments of the present invention have been shown and described above, it is to be understood that the above embodiments are exemplary and are not to be construed as limitations of the present invention. A person skilled in the art may change, modify, replace and vary the above embodiments within the scope of the present invention.
Claims
1. A method for identifying long strip objects for parking, characterized in that: The method comprises: Acquire image information of the parking area; Using a feature extraction layer of an object recognition model, a first strip convolution kernel and a second strip convolution kernel are used to extract features from the vertical direction and the horizontal direction of the image information respectively, so as to obtain a first feature and a second feature of the image information; Performing feature fusion on the first feature and the second feature to obtain a fused feature; According to the fusion feature, the long strip object in the image information is identified.
2. The method for recognizing a long strip object for parking according to claim 1, characterized in that: The feature extraction layer using the object recognition model uses a first strip convolution kernel and a second strip convolution kernel to extract features from the vertical direction and the horizontal direction of the image information respectively to obtain a first feature and a second feature of the image information, including: Using a first strip-shaped convolution kernel to extract features from a vertical direction of the image information using a first feature extraction layer of the object recognition model to obtain a first feature of the image information; The second feature extraction layer of the object recognition model uses a second strip convolution kernel to extract features from the horizontal direction of the image information to obtain the second feature of the image information.
3. The method for recognizing a long strip object for parking according to claim 1, characterized in that: The step of fusing the first feature and the second feature to obtain a fused feature includes: Performing image interpolation processing on the first feature and the second feature respectively so that the vector dimensions of the first feature and the second feature are consistent; A sum operation is performed on the first feature and the second feature with the same vector dimension to obtain the fused feature.
4. The method for recognizing a long strip object for parking according to claim 1, characterized in that: The step of identifying the long strip object in the image information according to the fusion feature includes: Using a preset convolution kernel to perform convolution processing on the fusion feature, changing the number of channels of the fusion feature, and obtaining a target fusion feature; the number of channels of the target fusion feature is the same as the number of channels of the image information; The long strip object in the image information is identified according to the target fusion feature.
5. The method for recognizing a long strip object for parking according to claim 4, characterized in that: The identifying the long strip object in the image information according to the target fusion feature includes: Calculating the target fusion feature using the activation function in the object recognition model to obtain the weights corresponding to each channel of the fusion feature; A product operation is performed on the weight and the image information to obtain a target feature, and a recognition result of the long strip object in the image information is obtained based on the target feature.
6. The method for recognizing a long strip object for parking according to claim 5, characterized in that: The value range of the weight is 0 to 1.
7. A strip-shaped object recognition device for parking, characterized in that: The device comprises: An information acquisition module, used to acquire image information of a parking area; A feature extraction module, configured to use a feature extraction layer of an object recognition model to extract features from a vertical direction and a horizontal direction of the image information using a first strip convolution kernel and a second strip convolution kernel, respectively, to obtain a first feature and a second feature of the image information; A feature fusion module, used for fusing the first feature and the second feature to obtain a fused feature; The object recognition module is used to recognize the long strip object in the image information according to the fusion feature.
8. A computer system comprising a memory and a processor, wherein the memory stores a computer program, wherein: When the processor executes the computer program, the steps of the method according to any one of claims 1 to 6 are implemented.
9. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the method according to any one of claims 1 to 6 are implemented.
10. A computer program product, comprising a computer program, characterized in that When the computer program is executed by a processor, the steps of the method according to any one of claims 1 to 6 are implemented.