Lane line recognition method, device and medium

Abstract

The embodiment of the present application discloses a lane line recognition method, device and medium. The lane line image to be tested is input into a lane edge preprocessing network to obtain lane line features, wherein the lane edge preprocessing network is composed of a corresponding encoder and decoder; a binary lane line edge map is generated according to the lane line features, so as to obtain lane line coordinates through the binary lane line edge map; the binary lane line edge map and the lane line coordinates are input into a preset lane line recognition network model, and the distance between the lane line coordinates and the estimated lane line is determined through the preset lane line recognition network model, so as to recognize the lane line in the lane line image to be tested according to the distance.

Description

Technical Field

[0001] The present application relates to the field of computer vision technology, and in particular to a lane line recognition method, device and medium. Background Art

[0002] With the continuous development of computer technology and computer hardware, the field of machine vision has made rapid progress and has been widely used in academia and industry. In recent years, the progress of multimodal information technology fusion and the development of various segmentation detection networks have also prompted the development of autonomous driving to a new stage.

[0003] Lane lines are lane markings that guide the direction. They are used to indicate that vehicles should drive in the indicated direction when entering an intersection. Such markings are usually drawn at intersections with heavy traffic to clarify the driving direction, so that each vehicle can drive in its own lane and relieve traffic pressure. Lane line detection is to detect the lane lines on the current road to provide accurate lane line coordinates and the shape of each lane line. As an important part of autonomous driving and automatic assisted driving, how to accurately identify one or more lane lines has become a key and difficult point in the current vision field. Summary of the invention

[0004] The embodiments of the present application provide a lane line recognition method, device and medium for solving the following technical problem: how to accurately identify one or more lane lines.

[0005] The present application embodiment adopts the following technical solutions:

[0006] An embodiment of the present application provides a lane line recognition method, including: inputting a lane line image to be measured into a lane edge preprocessing network to obtain lane line features, wherein the lane edge preprocessing network is composed of a corresponding encoder and a decoder; generating a binary lane line edge map according to the lane line features to obtain lane line coordinates through the binary lane line edge map; inputting the binary lane line edge map and the lane line coordinates into a preset lane line recognition network model, determining the distance between the lane line coordinates and the estimated lane line through the preset lane line recognition network model, so as to recognize the lane line in the lane line image to be measured according to the distance.

[0007] The embodiment of the present application extracts lane line features through a lane edge preprocessing network, and generates a binary lane line edge map based on the extracted features. Not only can the image be processed through an encoder-decoder to make it clearer, but the coordinate position of the lane line in the image can also be obtained. Secondly, the embodiment of the present application can obtain the number of lane lines by inputting the reduced-dimensional image and the lane line coordinates into a preset neural network model, thereby realizing the recognition of multiple lane lines.

[0008] In one implementation of the present application, a binary lane line edge map and lane line coordinates are input into a preset lane line recognition network model, specifically including: inputting the binary lane line edge map into a lane line local detection network to obtain a reduced-dimensional binary lane line edge map; obtaining the number of lane line edge points corresponding to the reduced-dimensional binary lane line edge map; reducing the dimension of the lane line coordinate values ​​according to the number of lane line edge points; and inputting the reduced-dimensional binary lane line edge map and the reduced-dimensional lane line coordinate values ​​into a preset lane line recognition network model.

[0009] The embodiment of the present application can reduce the amount of calculation by performing dimensionality reduction processing on the lane line coordinate values, thereby improving the speed of lane recognition.

[0010] In one implementation of the present application, a binary lane line edge map is input into a lane line local detection network to obtain a binary lane line edge map after dimensionality reduction, specifically including: inputting the binary lane line edge map into a fully connected layer in the lane line local detection network to obtain a first feature map; wherein the fully connected layer is composed of a one-dimensional convolution and a maximum pooling layer; inputting the first feature map into an average pooling layer, adding and averaging multiple channels corresponding to the first feature map to obtain a second average feature map; inputting the second average feature map into a low-dimensional representation layer for dimensionality reduction processing to obtain a binary lane line edge map after dimensionality reduction.

[0011] The embodiment of the present application can realize information exchange of all coordinate points by inputting the binary lane line edge map into the fully connected layer in the lane line local detection network. Secondly, the embodiment of the present application can reduce the amount of data and the complexity of calculation by reducing the dimension of the feature map, thereby improving the speed of calculation.

[0012] In one implementation of the present application, the lane line image to be tested is input into the lane edge preprocessing network to obtain the lane line features, specifically including: extracting the lane line features of the lane image to be tested through the convolution layer of the encoder in the lane edge preprocessing network; wherein the convolution layer includes a convolution kernel of a preset dimension and a dilated convolution kernel with different dilated coefficients.

[0013] In one implementation of the present application, a binary lane line edge map is generated based on lane line features, specifically including: restoring the feature resolution of lane line features through the micro-pixel convolution layer of the decoder in the lane edge preprocessing network to obtain a binary lane line edge map.

[0014] In one implementation of the present application, before inputting the lane line image to be tested into the lane edge preprocessing network and obtaining the lane line features, the method also includes: inputting the lane line image in the first training set into the encoder in the lane edge preprocessing network, and the encoder extracts lane line features from the lane line image in the first training set; the decoder in the lane edge preprocessing network generates a binary lane line edge map corresponding to the lane line image in the first training set according to the lane line features; determining the pixel value distinction between the lane line image in the first training set and the binary lane line edge map corresponding to the lane line image in the first training set through a preset loss function; when the pixel distinction meets the preset value, the trained encoder and decoder meet the requirements.

[0015] In one implementation of the present application, before the binary lane line edge map and the lane line coordinates are input into the preset lane line recognition network model, the method also includes: taking the lane line edge map in the second training set and the lane line coordinates marked in the lane line edge map as input, and taking the number of lane lines corresponding to the lane line edge map in the second training set as output, training the preset neural network to obtain the preset lane line recognition network model.

[0016] In one implementation of the present application, a preset neural network is trained, specifically including: obtaining the distance difference between the predicted lane line key value and the marked lane line key value, and determining the minimum square error corresponding to the distance difference; obtaining the lane line coordinate value, and determining the minimum predicted distance between the lane line coordinate value and the nearest estimated lane line; based on the minimum square error and the minimum predicted distance, obtaining the loss function value of the weakly supervised learning training of the lane line local detection network.

[0017] An embodiment of the present application provides a lane line recognition device, comprising: at least one processor; and a memory connected to the at least one processor in communication; wherein the memory stores instructions that can be executed by the at least one processor, and the instructions are executed by the at least one processor so that the at least one processor can: input the lane line image to be measured into a lane edge preprocessing network to obtain lane line features, wherein the lane edge preprocessing network is composed of corresponding encoders and decoders; generate a binary lane line edge map according to the lane line features to obtain lane line coordinates through the binary lane line edge map; input the binary lane line edge map and the lane line coordinates into a preset lane line recognition network model, and determine the distance between the lane line coordinates and the estimated lane line through the preset lane line recognition network model to recognize the lane line in the lane line image to be measured based on the distance.

[0018] A non-volatile computer storage medium provided in an embodiment of the present application stores computer executable instructions, and the computer executable instructions are configured to: input a lane line image to be measured into a lane edge preprocessing network to obtain lane line features, wherein the lane edge preprocessing network is composed of corresponding encoders and decoders; generate a binary lane line edge map according to the lane line features to obtain lane line coordinates through the binary lane line edge map; input the binary lane line edge map and the lane line coordinates into a preset lane line recognition network model, and determine the distance between the lane line coordinates and the estimated lane line through the preset lane line recognition network model to recognize the lane line in the lane line image to be measured according to the distance.

[0019] At least one of the above technical solutions adopted in the embodiments of the present application can achieve the following beneficial effects: The embodiments of the present application extract lane line features through a lane edge preprocessing network, and generate a binary lane line edge map based on the extracted features. Not only can the image be processed through an encoder-decoder to make it clearer, but the coordinate position of the lane line in the image can also be obtained. Secondly, the embodiments of the present application can obtain the number of lane lines by inputting the reduced-dimensional image and the lane line coordinates into a preset neural network model, thereby realizing the recognition of multiple lane lines. BRIEF DESCRIPTION OF THE DRAWINGS

[0020] In order to more clearly illustrate the technical solutions in the embodiments of the present application or the prior art, the drawings required for use in the embodiments or the prior art descriptions are briefly introduced below. Obviously, the drawings described below are only some embodiments recorded in the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without creative labor. In the drawings:

[0021] Figure 1 A flow chart of a lane line recognition method provided in an embodiment of the present application;

[0022] Figure 2 A lane edge preprocessing network structure diagram provided in an embodiment of the present application;

[0023] Figure 3 A lane line local detection network structure diagram provided in an embodiment of the present application;

[0024] Figure 4 A schematic diagram of the structure of a lane recognition device provided in an embodiment of the present application. DETAILED DESCRIPTION

[0025] Embodiments of the present application provide a lane line recognition method, device, and medium.

[0026] In order to enable those skilled in the art to better understand the technical solutions in this application, the technical solutions in the embodiments of this application will be clearly and completely described below in conjunction with the drawings in the embodiments of this application. Obviously, the described embodiments are only part of the embodiments of this application, not all of them. Based on the embodiments of this specification, all other embodiments obtained by ordinary technicians in this field without creative work should fall within the scope of protection of this application.

[0027] With the continuous development of computer technology and computer hardware, the field of machine vision has made rapid progress and has been widely used in academia and industry. In recent years, the progress of multimodal information technology fusion and the development of various segmentation detection networks have also prompted the development of autonomous driving to a new stage.

[0028] Lane detection is to detect the lane lines on the current road to provide accurate lane line coordinates and the shape of each lane line. Lane detection is an important part of autonomous driving and automatic assisted driving. How to quickly and accurately identify one or more lane lines has become a key and difficult point in the current vision field.

[0029] In addition, the current lane line detection process is easily interfered by other similar objects, forming noise for model recognition.

[0030] In order to solve the above problems, the embodiments of the present application provide a lane line recognition method, device and medium. The lane line features are extracted through the lane edge preprocessing network, and a binary lane line edge map is generated based on the extracted features. Not only can the image be processed by the encoder-decoder to make it clearer, but the coordinate position of the lane line in the image can also be obtained. Secondly, the embodiment of the present application can obtain the number of lane lines by inputting the reduced-dimensional image and the lane line coordinates into a preset neural network model, thereby realizing the recognition of multiple lane lines.

[0031] The technical solution proposed in the embodiments of the present application is described in detail below with reference to the accompanying drawings.

[0032] Figure 1 A flow chart of a lane line recognition method provided in an embodiment of the present application. Figure 1 As shown, the lane line recognition method includes the following steps:

[0033] S101. The lane line recognition device inputs the lane line image to be tested into the lane edge preprocessing network to obtain the lane line features, wherein the lane edge preprocessing network is composed of a corresponding encoder and decoder.

[0034] In one embodiment of the present application, the lane line recognition device inputs the lane line image in the first training set into the encoder in the lane edge preprocessing network, and the encoder extracts lane line features from the lane line image in the first training set. The decoder in the lane edge preprocessing network generates a binary lane line edge map corresponding to the lane line image in the first training set based on the lane line features. The first loss function is used to determine the pixel value distinction between the lane line image in the first training set and the binary lane line edge map corresponding to the lane line image in the first training set. When the pixel distinction meets the preset value, the trained encoder and decoder meet the requirements.

[0035] Specifically, the lane edge preprocessing network in the embodiment of the present application is composed of a corresponding encoder and decoder. Before processing the lane line image to be tested, it is necessary to obtain the corresponding encoder-decoder structure. Then the lane line image to be tested is input into the encoder and decoder in sequence.

[0036] In one embodiment of the present application, lane line features are extracted from the lane image to be tested through the convolution layer of the encoder in the lane edge preprocessing network, wherein the convolution layer includes a convolution kernel of a preset dimension and a dilated convolution kernel with different dilated coefficients.

[0037] Specifically, the encoder-decoder is trained using the lane line images in the first training set. The images in the training set are input into the encoder, and the encoder can extract features related to the lane lines, and input the extracted features into the decoder. At this time, the decoder gradually restores the feature resolution based on the received features and generates a binary lane line edge map. The pixel values ​​of the binary lane line edge map and the pixel values ​​of the image input to the encoder are calculated using the loss function to obtain the loss function value, and when the loss function value is less than the preset value, it indicates that the trained encoder and decoder meet the requirements.

[0038] Furthermore, in order to speed up the image processing speed, the encoder in the embodiment of the present application uses a small convolution kernel, and inputs the lane line image in the first training set into the first convolution layer of the encoder, which is composed of a 1×1 convolution kernel and a dilated convolution kernel with different dilated coefficients. Among them, the 1×1 convolution kernel realizes multi-channel level information fusion, and the use of dilated convolution reduces the number of parameters involved in the calculation and simplifies the calculation. The dilated convolution kernel size in the embodiment of the present application is 3×3 dimensions, and the scaling factors are 4, 2, and 1 respectively. The receptive field of the network model will be gradually expanded according to the preset scaling factor.

[0039] Furthermore, the first convolution layer corresponding to the encoder is connected to the average pooling layer, and the average pooling layer forms a first module. The encoder may include four modules and a separate fifth convolution layer, and each module is the same as the first module. In addition, the decoder in the embodiment of the present application uses a micro-pixel convolution layer to gradually restore the feature resolution.

[0040] The first loss function is used to determine the pixel difference between the lane line image in the first training set and the binary lane line edge map corresponding to the lane line image in the first training set. When the pixel difference meets the preset value, the trained encoder and decoder meet the requirements.

[0041] Specifically, the encoder-decoder training loss function is

[0042]

[0043] Among them, B is the batch size, P i is the predicted pixel value of the i-th sample in the current batch, is the nth pixel value in the image, k is the total number of pixels in the image, is the true pixel value of the i-th sample in the current batch. And k = w*h. β is the ratio of positive points to negative points. β can be calculated by the formula:

[0044]

[0045] Among them, w is the width of the image and h is the height of the image.

[0046] Figure 2 A lane edge preprocessing network structure diagram provided in an embodiment of the present application. Figure 2 As shown in the figure, the lane edge preprocessing network consists of a 1×1 convolution kernel and a dilated convolution kernel with different dilated coefficients. The 1×1 convolution kernel realizes multi-channel information fusion, and the use of dilated convolution reduces the overhead of additional parameters and simplifies the calculation. Among them, the dilated convolution kernel size is the same as 3*3 dimensions, and the scaling coefficients are 4, 2, and 1 respectively. The receptive field of the network model will be gradually expanded through different scaling coefficients.

[0047] S102: Generate a binary lane line edge map according to the lane line features, so as to obtain lane line coordinates through the binary lane line edge map.

[0048] In one embodiment of the present application, lane line features are extracted from the lane image to be tested through the convolution layer of the encoder in the lane edge preprocessing network. And the feature resolution of the lane line features is restored through the micro-pixel convolution layer of the decoder in the lane edge preprocessing network to obtain a binary lane line edge map corresponding to the lane line image to be tested.

[0049] Specifically, a lightweight encoder-decoder structure is used to preliminarily encode the lane line image to be tested. The encoder first performs a depth separation convolution layer, in which the depth separation convolution layer uses a 3×3 convolution kernel to perform layer-by-layer feature extraction. Secondly, the decoder structure performs feature resolution recovery and generates a pixel-level binary lane line edge map. The binary lane line edge map generated by the edge pre-detection network will enter the second stage of the lane line local detection network.

[0050] S103, inputting the binary lane line edge map and the lane line coordinates into a preset lane line recognition network model, determining the distance between the lane line coordinates and the estimated lane line through the preset lane line recognition network model, and identifying the lane line in the lane line image to be tested according to the distance.

[0051] In one embodiment of the present application, the lane line edge map in the second training set and the lane line coordinates marked in the lane line edge map are used as inputs, and the number of lane lines corresponding to the lane line edge map in the second training set is used as output to train the preset neural network to obtain a preset lane line recognition network model.

[0052] Specifically, the embodiment of the present application first trains the neural network model through a preset second training set to obtain a preset lane line recognition network model, wherein the second training set includes a plurality of lane line coordinates and a lane line edge map with the number of lane lines marked.

[0053] In one embodiment of the present application, a binary lane line edge map is input into a lane line local detection network to obtain a binary lane line edge map after dimensionality reduction. The number of lane line edge points corresponding to the binary lane line edge map after dimensionality reduction is obtained. The lane line coordinate values ​​are reduced in dimension according to the number of lane line edge points. The binary lane line edge map after dimensionality reduction and the lane line coordinate values ​​after dimensionality reduction are input into a preset lane line recognition network model.

[0054] Specifically, in the first stage of the lane line local detection network, the acquired lane line coordinates are used as input values, and the input value dimension is reduced from w×h to n×2, where n is the number of lane line edge points in the binary lane line edge map.

[0055] In one embodiment of the present application, the binary lane line edge map is input into the fully connected layer, the average pooling layer, and the low-dimensional representation layer in the lane line local detection network to obtain a binary lane line edge map after dimensionality reduction.

[0056] Specifically, the binary lane edge map is input into the fully connected layer in the lane local detection network to obtain the first feature map. The fully connected layer consists of a one-dimensional convolution and a maximum pooling layer. The first feature map is input into the average pooling layer, and the multi-channels corresponding to the first feature map are added and averaged to obtain the second average feature map. The second average feature map is input into the low-dimensional representation layer for dimensionality reduction processing to obtain the binary lane edge map after dimensionality reduction.

[0057] Furthermore, the lane line local detection network adopts a multi-layer shared fully connected layer, which is composed of a one-dimensional convolution and a maximum pooling layer. The size of the one-dimensional convolution kernel is the same as the feature dimension of each lane line coordinate point. The multi-layer shared fully connected layer realizes the variability and diversity of the input order. After passing through the above fully connected layer, the lane line local detection network enters the nonlinear eigenvalue aggregation stage to realize the information exchange of all coordinate points. Secondly, the first stage result will enter the average pooling layer. Finally, the first stage of the lane line local detection network will generate a low-dimensional representation layer to obtain a binary lane line edge map after dimensionality reduction. To enter the second stage deep neural network.

[0058] Figure 3 A lane line local detection network structure diagram provided in an embodiment of the present application. Figure 3 As shown in the figure, the lane line local detection network uses a multi-layer shared fully connected layer, which consists of a one-dimensional convolution and a maximum pooling layer. The size of the one-dimensional convolution kernel is the same as the feature dimension of each point. The output of the pooling layer is input into the neural network. After training, when the loss function value is lower than the preset threshold, a deep neural network model is generated.

[0059] In one embodiment of the present application, the second stage deep neural network will solve the problem of the uncertainty of the number of lane lines. The lane lines in the image are fitted as a second-order function, and the second stage deep neural network is continuously iterated to generate a confidence score for the lane line prediction.

[0060] Specifically, through the function

[0061]

[0062] Determine the function coefficients for each lane line. [K1, K2, K3] are the key values ​​of the lane line, [P2, P1, P0] are the second-order coefficients, including linear and constants, and h is the pixel value corresponding to the wide edge of the image.

[0063] Furthermore, the lane line local detection network can be trained by supervised learning or weakly supervised learning. The supervised learning training loss function can achieve function and model convergence by reducing the difference in the L2 norm distance between the average value of the predicted key value and the actual lane line label key value.

[0064] In one embodiment of the present application, the distance difference between the predicted lane line key value and the marked lane line key value is obtained, and the minimum square error corresponding to the distance difference is determined. The lane line coordinate value is obtained, and the minimum predicted distance between the lane line coordinate value and the nearest estimated lane line is determined. Based on the minimum square error and the minimum predicted distance, the loss function value of the weakly supervised learning training of the lane line local detection network is obtained.

[0065] Specifically, the lane line local detection network can be trained using weakly supervised learning, which greatly reduces the data labeling work. Weakly supervised training only requires the number of lane lines as labeling information and uses the min-distance loss function. The min-distance loss function is obtained by calculating the distance between each lane line coordinate input point and the nearest estimated lane line.

[0066] Furthermore, the output of the min-distance loss function is similar to the output of the supervised learning training loss function.

[0067]

[0068] Get the output value of the min-distance loss function. Among them, is the second-order coefficient, is the key value of the lane line. h is the distance between each lane line coordinate input point and the nearest estimated lane line.

[0069] Specifically, the overall loss function of the lane line local detection network weakly supervised learning training will combine the L2 norm loss function and the min-distance loss function:

[0070] L=L2+α*L min-distance

[0071] Among them, L is the overall loss function value, L2 is the L2 norm loss function, α is the constant coefficient, L min-distance is the output value of the min-distance loss function.

[0072] The output value of the L2 loss function is obtained by obtaining the distance difference between the predicted lane key value and the marked lane key value, and determining the minimum square error corresponding to the distance difference. The output value of the overall loss function can be obtained by adding the output value of the L2 loss function and the output value of the min-distance loss function.

[0073] In one embodiment of the present application, the lane line local detection network will stop after the loss function value generated by the deep neural network is lower than a certain threshold. At this point, the deep neural network training is completed. The binary lane line edge map corresponding to the lane line image to be tested and the generated lane line coordinates are input into the preset lane line recognition network model to obtain the number of lanes corresponding to the lane line image to be tested.

[0074] The embodiment of the present application uses an encoder-decoder in the lane edge pre-detection network to achieve the mapping from the vision group image stitching database (IPM) image to the binary lane edge projection map. Combining convolution with deep neural network, a weakly supervised learning training process that reduces manual labeling is implemented to complete the detection of multiple lane lines.

[0075] Figure 4 A schematic diagram of the structure of a lane line recognition device provided in an embodiment of the present application. Figure 4 As shown, the lane line recognition device includes:

[0076] at least one processor; and,

[0077] a memory communicatively connected to the at least one processor; wherein,

[0078] The memory stores instructions executable by the at least one processor, the instructions being executed by the at least one processor to enable the at least one processor to:

[0079] Inputting the lane line image to be measured into a lane edge preprocessing network to obtain lane line features, wherein the lane edge preprocessing network is composed of a corresponding encoder and a decoder;

[0080] Generate a binary lane line edge map according to the lane line features, so as to obtain lane line coordinates through the binary lane line edge map;

[0081] The binary lane line edge map and the lane line coordinates are input into a preset lane line recognition network model. The preset lane line recognition network model is used to determine the distance between the lane line coordinates and the estimated lane line, so as to identify the lane line in the lane line image to be tested according to the distance.

[0082] The embodiment of the present application also includes a non-volatile computer storage medium storing computer executable instructions, wherein the computer executable instructions are configured as follows:

[0083] Inputting the lane line image to be measured into a lane edge preprocessing network to obtain lane line features, wherein the lane edge preprocessing network is composed of a corresponding encoder and a decoder;

[0084] Generate a binary lane line edge map according to the lane line features, so as to obtain lane line coordinates through the binary lane line edge map;

[0085] The binary lane line edge map and the lane line coordinates are input into a preset lane line recognition network model. The preset lane line recognition network model is used to determine the distance between the lane line coordinates and the estimated lane line, so as to identify the lane line in the lane line image to be tested according to the distance.

[0086] Each embodiment in this application is described in a progressive manner, and the same or similar parts between the embodiments can be referred to each other, and each embodiment focuses on the differences from other embodiments. In particular, for the device, equipment, and non-volatile computer storage medium embodiments, since they are basically similar to the method embodiments, the description is relatively simple, and the relevant parts can be referred to the partial description of the method embodiment.

[0087] The above describes specific embodiments of the present application. Other embodiments are within the scope of the appended claims. In some cases, the actions or steps recorded in the claims can be performed in an order different from that in the embodiments and still achieve the desired results. In addition, the processes depicted in the accompanying drawings do not necessarily require the specific order or continuous order shown to achieve the desired results. In some embodiments, multitasking and parallel processing are also possible or may be advantageous.

[0088] The above is only an embodiment of the present application and is not intended to limit the present application. For those skilled in the art, the embodiments of the present application may have various changes and variations. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the embodiments of the present application should be included in the scope of the claims of the present application.

Claims

1. A lane line recognition method, characterized in that: The method comprises: Inputting the lane line image to be measured into a lane edge preprocessing network to obtain lane line features, wherein the lane edge preprocessing network is composed of a corresponding encoder and a decoder; Generate a binary lane line edge map according to the lane line features, so as to obtain lane line coordinates through the binary lane line edge map; Inputting the binary lane line edge map and the lane line coordinates into a preset lane line recognition network model, determining the distance between the lane line coordinates and the estimated lane line through the preset lane line recognition network model, so as to recognize the lane line in the lane line image to be tested according to the distance; The step of generating a binary lane line edge map according to the lane line features specifically includes: Performing feature resolution restoration on the lane line features through the micro-pixel convolution layer of the decoder in the lane edge preprocessing network to obtain a binary lane line edge map; The step of inputting the binary lane line edge map and the lane line coordinates into a preset lane line recognition network model specifically includes: Inputting the binary lane line edge map into a lane line local detection network to obtain a binary lane line edge map after dimensionality reduction; Obtaining the number of lane line edge points corresponding to the binary lane line edge graph after dimensionality reduction; Performing dimension reduction on the lane line coordinate value according to the number of lane line edge points; The reduced-dimensional binary lane line edge map and the reduced-dimensional lane line coordinate values ​​are input into the preset lane line recognition network model.

2. A lane line recognition method according to claim 1, characterized in that: The step of inputting the binary lane line edge map into a lane line local detection network to obtain a binary lane line edge map after dimensionality reduction specifically includes: Inputting the binary lane edge map into a fully connected layer in a lane local detection network to obtain a first feature map; wherein the fully connected layer is composed of a one-dimensional convolution and a maximum pooling layer; Input the first feature map into the average pooling layer, add and average multiple channels corresponding to the first feature map, and obtain a second average feature map; The second average feature map is input into the low-dimensional representation layer for dimensionality reduction processing to obtain a binary lane line edge map after dimensionality reduction.

3. A lane line recognition method according to claim 1, characterized in that: The lane line image to be measured is input into the lane edge preprocessing network to obtain the lane line features, specifically including: The lane line features are extracted from the lane image to be tested through the convolution layer of the encoder in the lane edge preprocessing network; wherein the convolution layer includes a convolution kernel of a preset dimension and a dilated convolution kernel with different dilated coefficients.

4. A lane line recognition method according to claim 1, characterized in that: Before inputting the lane line image to be measured into the lane edge preprocessing network to obtain the lane line features, the method further includes: Inputting the lane line image in the first training set into an encoder in the lane edge preprocessing network, wherein the encoder extracts lane line features from the lane line image in the first training set; The decoder in the lane edge preprocessing network generates a binary lane line edge map corresponding to the lane line image in the first training set according to the lane line features; Determining, by using a preset loss function, the degree of difference in pixel values ​​between the lane line image in the first training set and the binary lane line edge map corresponding to the lane line image in the first training set; When the pixel distinction meets the preset value, the trained encoder and decoder meet the requirements.

5. A lane line recognition method according to claim 1, characterized in that: Before inputting the binary lane line edge map and the lane line coordinates into a preset lane line recognition network model, the method further includes: The lane line edge map in the second training set and the lane line coordinates marked in the lane line edge map are used as input, and the number of lane lines corresponding to the lane line edge map in the second training set is used as output to train the preset neural network to obtain a preset lane line recognition network model.

6. A lane line recognition method according to claim 5, characterized in that: The training of the preset neural network specifically includes: Obtaining a distance difference between the predicted lane line key value and the marked lane line key value, and determining a minimum square error corresponding to the distance difference; Obtaining a lane line coordinate value, and determining a minimum predicted distance between the lane line coordinate value and the nearest estimated lane line; According to the minimum square error and the minimum predicted distance, a loss function value of the weakly supervised learning training of the lane line local detection network is obtained.

7. A lane line recognition device, comprising: at least one processor; as well as, a memory communicatively connected to the at least one processor; wherein, The memory stores instructions executable by the at least one processor, the instructions being executed by the at least one processor to enable the at least one processor to: Inputting the lane line image to be measured into a lane edge preprocessing network to obtain lane line features, wherein the lane edge preprocessing network is composed of a corresponding encoder and a decoder; Generate a binary lane line edge map according to the lane line features, so as to obtain lane line coordinates through the binary lane line edge map; Inputting the binary lane line edge map and the lane line coordinates into a preset lane line recognition network model, determining the distance between the lane line coordinates and the estimated lane line through the preset lane line recognition network model, so as to recognize the lane line in the lane line image to be tested according to the distance; The step of generating a binary lane line edge map according to the lane line features specifically includes: Performing feature resolution restoration on the lane line features through the micro-pixel convolution layer of the decoder in the lane edge preprocessing network to obtain a binary lane line edge map; The step of inputting the binary lane line edge map and the lane line coordinates into a preset lane line recognition network model specifically includes: Inputting the binary lane line edge map into a lane line local detection network to obtain a binary lane line edge map after dimensionality reduction; Obtaining the number of lane line edge points corresponding to the binary lane line edge graph after dimensionality reduction; Performing dimension reduction on the lane line coordinate value according to the number of lane line edge points; The reduced-dimensional binary lane line edge map and the reduced-dimensional lane line coordinate values ​​are input into the preset lane line recognition network model.

8. A non-volatile computer storage medium storing computer executable instructions, wherein the computer executable instructions are configured to: The lane line image to be tested is input into the lane edge preprocessing network to obtain the lane line features, where: The lane edge preprocessing network is composed of corresponding encoders and decoders; Generate a binary lane line edge map according to the lane line features, so as to obtain lane line coordinates through the binary lane line edge map; Inputting the binary lane line edge map and the lane line coordinates into a preset lane line recognition network model, determining the distance between the lane line coordinates and the estimated lane line through the preset lane line recognition network model, so as to recognize the lane line in the lane line image to be tested according to the distance; The step of generating a binary lane line edge map according to the lane line features specifically includes: Performing feature resolution restoration on the lane line features through the micro-pixel convolution layer of the decoder in the lane edge preprocessing network to obtain a binary lane line edge map; The step of inputting the binary lane line edge map and the lane line coordinates into a preset lane line recognition network model specifically includes: Inputting the binary lane line edge map into a lane line local detection network to obtain a binary lane line edge map after dimensionality reduction; Obtaining the number of lane line edge points corresponding to the binary lane line edge graph after dimensionality reduction; Performing dimension reduction on the lane line coordinate value according to the number of lane line edge points; The reduced-dimensional binary lane line edge map and the reduced-dimensional lane line coordinate values ​​are input into the preset lane line recognition network model.

Images