WO2022237139A1

WO2022237139A1 - Lanesegnet-based lane line detection method and system

Info

Publication number: WO2022237139A1
Application number: PCT/CN2021/135474
Authority: WO
Inventors: 高尚兵; 胡序洋; 汪长春; 陈浩霖; 蔡创新; 相林; 于永涛; 周君; 朱全银; 张正伟; 郝明阳; 张骏强; 李�杰; 李少凡
Original assignee: 淮阴工学院
Priority date: 2021-05-14
Filing date: 2021-12-03
Publication date: 2022-11-17
Also published as: CN113343778A; CN113343778B

Abstract

A LaneSegNet-based lane line detection method and system. According to the method and system, polygon filling is first performed on an image to obtain a ROI, and then an ROI image is input to a trained LaneSegNet network model to obtain a binary image comprising lane lines, pixel point coordinates of each lane line is clustered by using a DBSCAN algorithm, polynomial fitting is performed on same, and the fitted lane lines are displayed on the original image. The constructed LaneSegNet network model comprises a network architecture of an encoding module, a decoding module, an enhanced receptive field module, and enhanced feature modules. The network receptive field is improved by using the parallel atrous convolution modules. Feature information irrelevant to the current task is removed by using the enhanced feature modules. Asymmetric convolutions are used to construct a feature extraction network, so that network parameters are reduced. The method has an accuracy of 98.62%, can be used for detecting lane lines on a highway, and has good robustness and real-time performance.

Description

A lane line detection method and system based on LaneSegNet

technical field

The invention belongs to the technical field of computer vision, and in particular relates to a lane line detection method and system based on LaneSegNet (lane line segmentation network).

Background technique

With the improvement of people's living standards, cars play an increasingly important role in people's lives. However, the increase in car ownership leads to more and more traffic accidents. In order to ensure driving safety, the automatic driving function is increasingly With more and more attention, lane line detection is an important part of the automatic driving function.

The current traditional lane line detection methods, such as using Hough transform to detect lane lines, first use image processing to extract lane line edge features, and then use Hough transform to detect and fit the lines. The above-mentioned lane line detection method can only be used for uniform illumination, single environment, no occlusion, and blurred detection, and its robustness is poor.

With the wide application of deep learning in various fields, the combination of lane line detection and deep learning is getting closer and closer. For example, an end-to-end lane line detection proposed by Davy Neven et al. first detects lane lines through a deep neural network. , and then use the clustering algorithm to cluster the lane lines, and finally use polynomial fitting to get the lane lines. However, the detection process of the above method is relatively complicated and time-consuming, and it is difficult to meet the real-time requirements.

Contents of the invention

Purpose of the invention: To solve the problems of poor robustness, complicated process and long detection time of current lane line detection, the present invention provides a lane line detection method and system based on LaneSegNet.

Technical solution: In order to achieve the above-mentioned purpose of the invention, the present invention adopts the following technical solution:

A lane detection method based on LaneSegNet, comprising the following steps:

(1) Carry out polygon filling to road image, obtain the ROI area image that contains lane line;

(2) Input the ROI area image into the trained LaneSegNet network model to obtain a binary image containing lane lines;

The LaneSegNet network model includes sequentially connected initial modules, three convolutional downsampling modules, enhanced receptive field modules, four convolutional upsampling modules, and two enhanced feature modules; the initial module is used to halve the size of the input image , the convolutional downsampling module is used to extract lane line feature information, the enhanced receptive field module is used to increase the receptive field of the network, the enhanced feature module is used to enhance lane line information, and the convolutional upsampling module is used to restore image size and image features ; wherein the first enhanced feature module is connected to the output of the first convolution downsampling module and the second convolution upsampling module, and the second enhancement feature module is connected to the output of the second downsampling module and the first convolution upsampling module, The second convolution upsampling module is connected to the output end of the second enhancement feature module, and the third convolution upsampling module is connected to the output end of the first enhancement feature module;

(3) use the DBSCAN algorithm to cluster the lane line pixel point coordinates to the binary image that step (2) obtains, divide the lane lines of different classes, use quadratic polynomial to fit respectively for the lane lines of different classes;

(4) Display the fitted lane lines on the original road image to realize the visualization of lane line detection.

Preferably, the initial module includes a convolutional layer with a convolution kernel size of k×k and a step size of 1, a convolution layer with a convolution kernel size of k×k and a step size of 2, and a maximum pooling layer, a connection layer, in which two convolutional layers are sequentially connected, and between the convolutional layer and the pooling layer are parallel connections, and k is 3 or 5.

Preferably, the unit structures of the three convolution down-sampling modules are the same, including a 1×1 convolution, two k×k convolutions and the first branch of the 1×1 convolution series connection, and three dilation rates respectively The second branch of the hole convolution series connection of 1, 2, and 5, the first branch and the second branch are connected in parallel, the input of this module is added to the output of the two parallel connections, and then divided into The third branch of a 1×1 convolution, two k×k convolutions and 1×1 convolution in series and the fourth branch of the maximum pooling layer, the third branch and the fourth branch are connected in parallel , the outputs of the two branches are added, and k is 3 or 5.

Preferably, the enhanced receptive field module includes three parallel atrous convolution branches, the first atrous convolution branch includes a k×k convolution with an expansion rate of 1, and the second atrous convolution branch includes four Consecutive k×k convolutions with dilation rates of 2, 5, 9, and 13 respectively. The structure of the third hole convolution branch is the same as that of the second hole convolution branch. The first hole convolution branch The output of the atrous convolution branch is the input of the second atrous convolution branch, the output of the second atrous convolution branch is the input of the third atrous convolution branch, and finally the outputs of the three atrous convolution branches are added together , k is 3 or 5.

Preferably, the two enhanced feature modules have the same structure, including the first k×k asymmetric convolution, two parallel global average pooling and global maximum pooling, and the second k×k asymmetric convolution , a 1×1 convolution, a sigmoid activation layer, and a threshold layer, multiply the final threshold with the input, k is 3 or 5; the threshold function is:

Preferably, the four convolution upsampling modules have the same structure, including a 1×1 convolution connected in sequence, a k×k convolution, two parallel transposed convolutions and upsampling, and a 1× 1 convolution; each convolution operation is followed by Batch Normalization and PReLU nonlinear activation function processing.

As preferably, the step of training LaneSegNet network model in described step (2) comprises:

(2.1) Input the ROI area image of the road image and the binary image of the marked lane line into the LaneSegNet network model as training sample data;

(2.2) Calculate the loss of the LaneSegNet network, and continuously optimize the parameters in the network with the minimum loss as the goal;

(2.3) When the loss value is stable within a certain range, the network parameters are saved to obtain the final lane line detection model.

Based on the same inventive concept, the present invention provides a lane detection system based on LaneSegNet, including:

The preprocessing module is used to carry out polygon filling to the road image, and obtains the ROI area image containing the lane line;

Lane line recognition module, for inputting the ROI area image into the LaneSegNet network model trained, obtains the binary image that contains lane line; Described LaneSegNet network model comprises the initial module connected successively, three convolution down-sampling modules, Enhanced receptive field module, four convolutional upsampling modules, and two enhanced feature modules; the initial module is used to halve the size of the input image, the convolutional downsampling module is used to extract lane line feature information, and the enhanced receptive field module is used to Increase the receptive field of the network, the enhanced feature module is used to enhance the lane line information, and the convolution upsampling module is used to restore the image size and image features; the first enhanced feature module is combined with the first convolution downsampling module and the second convolution The output terminal of the upsampling module is connected, the second enhanced feature module is connected with the output terminal of the second downsampling module and the first convolutional upsampling module, the second convolutional upsampling module is connected with the output terminal of the second enhanced feature module, and the third volume The product upsampling module is connected with the first enhanced feature module output;

The lane line fitting module is used to cluster the lane line pixel coordinates using the DBSCAN algorithm on the binary image obtained by the LaneSegNet network model, divide different types of lane lines, and use quadratic polynomials for different types of lane lines. combine;

And, the result output module is used for displaying the fitted lane line on the original road image, so as to realize the visualization of lane line detection.

Based on the same inventive concept, the present invention provides a lane line detection system based on LaneSegNet, including a memory, a processor and a computer program stored on the memory and operable on the processor, the computer program being loaded into the processor When realizing the described lane detection method based on LaneSegNet.

Beneficial effects: Compared with the prior art, the beneficial effects of the present invention are as follows: 1. By using the LaneSegNet model architecture, the structure is simple and the number of parameters is small to realize high-precision extraction of lane lines in road images. 2. The enhanced receptive field module increases the receptive field of the network model to avoid the problem that the lane line spans too large in the image and is difficult to segment. 3. The enhanced feature module removes features irrelevant to the task, making the network more focused on extracting lane line features, and avoiding segmentation difficulties caused by the small proportion of lane line pixels in the image.

Description of drawings

Fig. 1 is the method flowchart of the embodiment of the present invention;

Fig. 2 is the LaneSegNet model network structural diagram in the embodiment of the present invention;

Fig. 3 is the initial module network structure diagram in the embodiment of the present invention;

FIG. 4 is a network structure diagram of an encoding module in an embodiment of the present invention;

FIG. 5 is a network structure diagram of an enhanced receptive field module in an embodiment of the present invention;

FIG. 6 is a network structure diagram of an enhanced feature module in an embodiment of the present invention;

FIG. 7 is a network structure diagram of a decoding module model in an embodiment of the present invention;

Fig. 8 is a schematic diagram of some data sets used in the embodiment of the present invention;

Fig. 9 is a schematic diagram of some data set tags used in the embodiment of the present invention;

Fig. 10 is the data set picture labeling process in the embodiment of the present invention;

Fig. 11 is a segmentation diagram of an example of road dividing line detection in the embodiment of the present invention;

Fig. 12 is a fitting diagram of lane lines in an embodiment of the present invention;

Fig. 13 is the miou diagram in the training process in the embodiment of the present invention;

FIG. 14 is an acc graph during training in an embodiment of the present invention.

Detailed ways

The present invention will be further described below in conjunction with the accompanying drawings and specific embodiments.

As shown in Fig. 1, a kind of lane line detection method based on LaneSegNet disclosed in the embodiment of the present invention first carries out polygon filling to road image, obtains the ROI (region of interest) area containing lane line; Then ROI area image is input into In the trained LaneSegNet network model, a binary image containing lane lines is obtained; then the DBSCAN algorithm is used to cluster the pixel coordinates of lane lines in the binary image, and quadratic polynomials are used to fit different types of lane lines respectively ; Finally, the fitted lane line is displayed on the original road image to realize the visualization of lane line detection.

The specific structure of the data set and the network model used in this embodiment will first be described in detail below.

Preprocess the road video captured by the camera, obtain effective road video data, and extract frames from the acquired image to obtain the road image to be marked, and use the marking tool to mark the lane line in the image to be marked to obtain the lane line target mark road image. Thus, the constructed data set is obtained.

As shown in Fig. 2, the LaneSegNet model constructed by the embodiment of the present invention mainly includes an encoding module, a decoding module, an enhanced receptive field module and an enhanced feature module. The encoding module mainly includes three identical convolutional downsampling modules, which expand the receptive field of the network by using hole convolution; the decoding module mainly includes four identical convolutional upsampling modules, which restore feature information and image size; enhance the receptive field module, to further increase the receptive field of the network; to enhance the feature module, to enhance the information related to the current task, and to discard the information irrelevant to the current task. Input the training data into the LaneSegNet network for training, and obtain a binary image containing lane lines. The model structure specifically includes an initial module connected in sequence, three convolutional downsampling modules, an enhanced receptive field module, four convolutional upsampling modules, and two enhanced feature modules; the initial module is used to halve the size of the input image, and the volume The product downsampling module is used to extract the feature information of the lane line, the enhanced receptive field module is used to increase the receptive field of the network, the enhanced feature module is used to enhance the lane line information, and the convolutional upsampling module is used to restore the image size and image features; The first enhanced feature module is connected to the output of the first convolution downsampling module and the second convolution upsampling module, the second enhancement feature module is connected to the second downsampling module and the output of the first convolution upsampling module, and the second The convolutional upsampling module is connected to the output end of the second enhancement feature module, and the third convolutional upsampling module is connected to the output end of the first enhancement feature module.

As shown in Figure 3, the initial module includes a convolution layer with a convolution kernel size of 3×3 and a step size of 1, a convolution layer with a convolution kernel size of 3×3 and a step size of 2, and a maximum pooling layer , a connection layer, where the two convolutional layers are sequentially connected, and the convolutional layer and the pooling layer are connected in parallel.

The unit structure of the three convolutional downsampling modules is the same, as shown in Figure 4, all of which include a 1×1 convolution, two 3×3 convolutions and the first branch of the 1×1 convolution concatenation, and three expansion rates The second branch of the hole convolution series connection of 1, 2, and 5 respectively, the first branch and the second branch are connected in parallel, the input of this module is added to the output of the two parallel connections, and then divided into A third branch connected in series by a 1×1 convolution, two 3×3 convolutions, and a 1×1 convolution, and a fourth branch that sets the maximum pooling layer, and the third branch and the fourth branch are parallel Connect, and the output of the two branches is added.

As shown in Figure 5, the enhanced receptive field module includes three parallel hole convolution branches, the first hole convolution branch includes a 3×3 convolution with an expansion rate of 1, and the second hole convolution branch includes four Consecutive 3×3 convolution, expansion ratios are 2, 5, 9, 13 respectively, the structure of the third dilated convolution branch is the same as that of the second dilated convolution branch, and the structure of the first dilated convolution branch The output is the input of the second hole convolution branch, the output of the second hole convolution branch is the input of the third hole convolution branch, and finally the outputs of the three hole convolution branches are added together.

The structure of the two enhanced feature modules is the same, as shown in Figure 6, including the first 3×3 asymmetric convolution, two parallel global average pooling and global maximum pooling, and the second 3×3 asymmetric convolution product, a 1×1 convolution, a sigmoid activation layer, and a threshold layer, which multiply the final threshold with the input; the threshold function is:

The structure of the four convolutional upsampling modules is the same, as shown in Figure 7, including a 1×1 convolution connected in sequence, a 3×3 convolution, two parallel Conv2DTranspose and UpSampling2D, and a 1×1 convolution Product; each convolution operation is followed by Batch Normalization and PReLU nonlinear activation function processing.

The training steps of the LaneSegNet network model are as follows: first extract the region of interest from the road image, and set six coordinate points: r1 (0,270), r2 (0, h), r3 (w, h), r4 (w, 470), r5 (670,150), r6(570,150), where w is the width of the input image, h is the height of the input image, and the outside of the area surrounded by the coordinate points is set to 0 to obtain the area of interest containing the lane line, and then the area of interest The image and the corresponding binary image are input into the LaneSegNet network model as training sample data, and then the loss of the LaneSegNet network is calculated, and the parameters in the network are continuously optimized with the minimum loss as the goal. When the loss value is stable within a certain range, the network parameters are saved to obtain the final lane line detection model. In this embodiment, the constructed data set is the video data taken by vehicles driving on the highway. Since the original video data contains information irrelevant to the training task, it is necessary to crop and denoise the original video. After processing, the video frame is drawn to obtain enough training data. The obtained data is shown in Figure 8.

Use the Labelme tool to label the lane lines in the obtained road image. After labeling a picture, a .json file will be generated in the original folder, and the corresponding binary image will be generated according to the .json file, as shown in Figure 9. In this embodiment, a total of 11807 road images are labeled, and the labeling process is shown in FIG. 10 .

In this embodiment, the data folder is Datasets, which includes two subfolders, namely Images and Labels, wherein Images saves the training images, and Labels saves the binary images corresponding to the training images, wherein the ratio of the training set to the verification set It is 7:3, the training set image path is saved in train.txt, and the verification set image path is saved in val.txt, which stores the relative paths of training images and labels.

Training of the LaneSegNet network model: Input the marked lane line data set into the LaneSegNet network for training, set the corresponding parameters, and perform model training to obtain the trained LaneSegNet network model. The specific steps are as follows:

1) Set parameters, including learning rate, Epochs size, batch size, etc. Among them, the initial learning rate is 1e-3, and finally drops to 2.5e-5. Among them, batch-size is 2 and Epochs is 100.

2) Train the data, use the parameters set in 1) to train and predict the image, and the prediction effect is shown in Figure 11.

After obtaining a binary image containing lane lines, use DBSCAN to cluster the segmentation results and use a quadratic polynomial to fit the clustering results. The fitting effect is shown in Figure 12.

Obtain the coordinates of the lane line pixel points from the output data of the LaneSegNet network, use the DBSCAN algorithm to cluster the lane line pixel point coordinates, thereby dividing different types of lane lines, and use quadratic polynomials for different types of lane lines to simulate Combined, use the fitted quadratic polynomial to draw lane lines in the original image to realize the visualization of lane line detection.

The experimental environment used in the embodiment of the present invention is as follows:

operating system:

Windows 10 64 bit

Hardware environment:

Inter Core i5-10400F@2.90GHZ six-core

16GB DDR4 2600MHZ RAM

Nvidia GTX 2060 SUPER with 6GB DRAM

WDS 500G with SSD

Software Environment:

Deep learning framework Keras (2.2.5)

Operating environmentPython 3.6

JetBrains PyCharm 2020.2 x64

CUDA 10.2

MIOU is an evaluation index for semantic segmentation and an important criterion for measuring model performance. miou is the ratio of the intersection and union of the two sets of real values and predicted values. The miou of the LaneSegNet network model proposed in the present invention is shown in Figure 13. It can be seen from the figure that a LaneSegNet-based network model proposed by the present invention The miou value of the lane line detection method is 78.18%.

The method provided by the present invention is to detect lane lines based on the network model of LaneSegNet, and the detection accuracy of this method has achieved a good effect. As shown in Figure 14, the lane line detection based on the method of the present invention has an accuracy of 98.62%. .

Based on the same inventive concept, a lane line detection system based on LaneSegNet disclosed in the embodiment of the present invention includes a preprocessing module, which is used to perform polygon filling on the road image, and obtain an ROI region image containing lane lines; a lane line recognition module, It is used to input the ROI area image into the trained LaneSegNet network model to obtain a binary image containing lane lines; the LaneSegNet network model includes sequentially connected initial modules, three convolutional downsampling modules, enhanced receptive field modules, Four convolutional upsampling modules, and two enhanced feature modules; the initial module is used to halve the size of the input image, the convolutional downsampling module is used to extract lane line feature information, and the enhanced receptive field module is used to increase the perception of the network In the field, the enhanced feature module is used to enhance the lane line information, and the convolution upsampling module is used to restore the image size and image features; wherein the first enhanced feature module is connected with the first convolution downsampling module and the second convolution upsampling module output connected, the second enhanced feature module is connected with the second downsampling module and the output of the first convolutional upsampling module, the second convolutional upsampling module is connected with the output of the second enhanced feature module, and the third convolutional upsampling module is connected with The output end of the first enhanced feature module is connected; the lane line fitting module is used to use the DBSCAN algorithm to cluster the lane line pixel point coordinates for the binary image obtained by the LaneSegNet network model, and divide different types of lane lines, for different types of lane lines The lane lines are respectively fitted by quadratic polynomials; and, the result output module is used to display the fitted lane lines on the original road image, so as to realize the visualization of lane line detection.

For specific details, refer to the above-mentioned method for implementation, and details are not repeated here.

Based on the same inventive concept, a LaneSegNet-based lane line detection system provided in an embodiment of the present invention includes a memory, a processor, and a computer program stored in the memory and operable on the processor, the computer program being loaded into the processing The above-mentioned LaneSegNet-based lane line detection method is implemented when the device is used.

Claims

A lane line detection method based on LaneSegNet, is characterized in that, comprises the following steps:

(1) Carry out polygon filling to road image, obtain the ROI area image that contains lane line;

(2) Input the ROI area image into the trained LaneSegNet network model to obtain a binary image containing lane lines; the LaneSegNet network model includes sequentially connected initial modules, three convolution down-sampling modules, and enhanced receptive field modules , four convolutional upsampling modules, and two enhanced feature modules; the initial module is used to halve the size of the input image, the convolutional downsampling module is used to extract lane line feature information, and the enhanced receptive field module is used to increase the network. The receptive field, the enhanced feature module is used to enhance the lane line information, and the convolutional upsampling module is used to restore the image size and image features; the first enhanced feature module is output with the first convolutional downsampling module and the second convolutional upsampling module Terminal connection, the second enhanced feature module is connected with the output of the second downsampling module and the first convolutional upsampling module, the second convolutional upsampling module is connected with the output end of the second enhanced feature module, and the third convolutional upsampling module connected to the output end of the first enhanced feature module;

(3) use the DBSCAN algorithm to cluster the lane line pixel point coordinates to the binary image that step (2) obtains, divide the lane lines of different classes, use quadratic polynomial to fit respectively for the lane lines of different classes;

(4) Display the fitted lane lines on the original road image to realize the visualization of lane line detection.
A kind of lane line detection method based on LaneSegNet according to claim 1, it is characterized in that, described initial module comprises a convolution kernel size is k * k, the convolution layer that stride is 1, a convolution kernel size k×k convolutional layer with a step size of 2, a maximum pooling layer, and a connection layer, where the two convolutional layers are sequentially connected, and the convolutional layer and the pooling layer are connected in parallel, k for 3 or 5.
A lane line detection method based on LaneSegNet according to claim 1, wherein the unit structures of the three convolution downsampling modules are the same, including a 1×1 convolution, two k×k convolutions and The first branch of the 1 × 1 convolution series, the second branch of the three hole convolution series with expansion rates of 1, 2, and 5 respectively, the first branch and the second branch are connected in parallel, the module The input of the input is added to the output of two parallel connections, and then divided into a third branch consisting of a 1×1 convolution, two k×k convolutions and 1×1 convolutions connected in series and setting the maximum pooling The fourth branch of the layer, the third branch and the fourth branch are connected in parallel, and the outputs of the two branches are added, and k is 3 or 5.
A lane line detection method based on LaneSegNet according to claim 1, wherein the enhanced receptive field module includes three parallel atrous convolution branches, and the first atrous convolution branch includes a dilation rate 1 k×k convolution, the second dilated convolution branch includes four consecutive k×k convolutions, the expansion rates are 2, 5, 9, 13 respectively, the structure of the third dilated convolution branch is the same as The second hole convolution branch is the same, the output of the first hole convolution branch is the input of the second hole convolution branch, and the output of the second hole convolution branch is the third hole convolution branch The input of the channel, and finally the output of the three dilated convolution branches is added, and k is 3 or 5.
A lane line detection method based on LaneSegNet according to claim 1, wherein the two enhanced feature modules have the same structure, including the first k×k asymmetric convolution, two parallel global average pools Globalization and global maximum pooling, the second k×k asymmetric convolution, a 1×1 convolution, a sigmoid activation layer, a threshold layer, and multiply the final threshold with the input, k is 3 or 5; where the threshold function is:
A lane line detection method based on LaneSegNet according to claim 1, wherein the four convolution upsampling modules have the same structure, and all include a 1×1 convolution connected in sequence, a k×k Convolution, two parallel transposed convolutions and upsampling, a 1×1 convolution; each convolution operation is followed by Batch Normalization and PReLU nonlinear activation function processing.
A kind of lane line detection method based on LaneSegNet according to claim 1, is characterized in that, the step of training LaneSegNet network model in described step (2) comprises:

(2.1) Input the ROI area image of the road image and the binary image of the marked lane line into the LaneSegNet network model as training sample data;

(2.2) Calculate the loss of the LaneSegNet network, and continuously optimize the parameters in the network with the minimum loss as the goal;

(2.3) When the loss value is stable within a certain range, the network parameters are saved to obtain the final lane line detection model.
A lane detection system based on LaneSegNet, characterized in that, comprising:

The preprocessing module is used to carry out polygon filling to the road image, and obtains the ROI area image containing the lane line;

Lane line recognition module, for inputting the ROI area image into the LaneSegNet network model trained, obtains the binary image that contains lane line; Described LaneSegNet network model comprises the initial module connected successively, three convolution down-sampling modules, Enhanced receptive field module, four convolutional upsampling modules, and two enhanced feature modules; the initial module is used to halve the size of the input image, the convolutional downsampling module is used to extract lane line feature information, and the enhanced receptive field module is used to Increase the receptive field of the network, the enhanced feature module is used to enhance the lane line information, and the convolution upsampling module is used to restore the image size and image features; the first enhanced feature module is combined with the first convolution downsampling module and the second convolution The output terminal of the upsampling module is connected, the second enhanced feature module is connected with the output terminal of the second downsampling module and the first convolutional upsampling module, the second convolutional upsampling module is connected with the output terminal of the second enhanced feature module, and the third volume The product upsampling module is connected with the first enhanced feature module output;

The lane line fitting module is used to cluster the lane line pixel coordinates using the DBSCAN algorithm on the binary image obtained by the LaneSegNet network model, divide different types of lane lines, and use quadratic polynomials for different types of lane lines. combine;

And, the result output module is used for displaying the fitted lane line on the original road image, so as to realize the visualization of lane line detection.
A lane line detection system based on LaneSegNet, comprising a memory, a processor, and a computer program stored on the memory and operable on the processor, characterized in that, when the computer program is loaded into the processor, the system according to claim 1 is implemented. The lane line detection method based on LaneSegNet described in any one of -7.