Lane detection method, device and equipment based on hybrid coordinate key point estimation

Abstract

The present application relates to a lane line detection method based on hybrid coordinate key point estimation. By extracting row and column anchor confidence feature maps, the method can extract coordinate points with higher confidence; by extracting row and column anchor offset feature maps, the offset corresponding to the coordinate point with higher confidence can be confirmed, and then the coordinate point can be corrected based on the offset to eliminate quantization errors, thereby ultimately improving the accuracy of subsequent detections; in addition, the method also combines a column anchor detection method on the basis of a row anchor detection method, and selects a target row anchor lane line or a target column anchor lane line as output by judging the absolute value of the average value between the slope of the target row anchor lane line and the slope of the target column anchor lane line and the size between a second threshold, thereby effectively solving the problem of missed detection caused by the inability of traditional key point estimation methods to effectively characterize lane lines with excessively large inclination angles in the scene, thereby improving the accuracy of the final lane line detection.

Description

Technical Field

[0001] The embodiments of the present application relate to the technical field of lane line detection, and in particular to a lane line detection method, device and equipment based on hybrid coordinate key point estimation. Background Art

[0002] Intelligent assisted driving technology has been widely used in civilian cars and special operation vehicles. Lane lines, as important visual cues for vehicle driving, play a key role in restricting driving range and improving driving safety and stability.

[0003] In today's autonomous driving systems, lane detection is usually achieved by computer vision technology. Traditional lane detection methods often rely on artificially set prior knowledge to design model structures and parameters, such as edge detection or color segmentation. However, these technologies are easily disturbed by road conditions when facing complex road conditions, resulting in limited robustness and accuracy. Therefore, deep learning has gradually replaced these traditional methods. With the help of deep neural networks, the system can automatically learn and adapt to different road scenarios to achieve accurate recognition and positioning of lane lines.

[0004] However, existing deep learning methods still have limitations when faced with lane line inclination angles that are too large in images, and it is difficult to meet the requirements of autonomous driving systems for lane line detection in complex scenarios. Summary of the invention

[0005] The following is a summary of the subject matter described in detail herein. This summary is not intended to limit the scope of the claims.

[0006] The present application aims to at least solve the technical problems existing in the prior art. To this end, the present application proposes a lane line detection method, device and equipment based on hybrid coordinate key point estimation, which can improve the accuracy of the final lane line detection.

[0007] A first aspect of an embodiment of the present application provides a lane line detection method based on hybrid coordinate key point estimation, the method comprising:

[0008] Get lane image;

[0009] Extracting a lane line detection result from the lane image according to the detection model; the detection process of the detection model includes:

[0010] extracting an initial feature map from the lane image;

[0011] Extracting a row anchor confidence feature map and a corresponding row anchor offset feature map, as well as a column anchor confidence feature map and a corresponding column anchor offset feature map from the initial feature map;

[0012] Extracting a plurality of first coordinate points whose confidence is greater than a first threshold from the line anchor confidence feature map, and extracting a plurality of first offsets corresponding to the plurality of first coordinate points from the line anchor offset feature map, performing offset compensation on the plurality of first coordinate points according to the plurality of first offsets, and generating a plurality of line anchor lane lines according to the compensated plurality of first coordinate points;

[0013] Extracting a plurality of second coordinate points whose confidences are greater than the first threshold from the column anchor confidence feature map, and extracting a plurality of second offsets corresponding to the plurality of second coordinate points from the column anchor offset feature map, performing offset compensation on the plurality of second coordinate points according to the plurality of second offsets, and generating a plurality of column anchor lane lines according to the compensated plurality of second coordinate points;

[0014] Calculate the absolute value of the average value between the slope of the target row anchor lane line and the slope of the target column anchor lane line, and when the absolute value is greater than a second threshold, use the target row anchor lane line as the lane line detection result; when the absolute value is less than or equal to the second threshold, use the target column anchor lane line as the lane line detection result; wherein, the target row anchor lane line is any one of the multiple row anchor lane lines, and the target column anchor lane line is one of the multiple column anchor lane lines corresponding to the target row anchor lane line.

[0015] A second aspect of an embodiment of the present application provides a lane line detection device based on hybrid coordinate key point estimation, the device comprising:

[0016] An image acquisition unit, used for acquiring a lane image;

[0017] A lane line detection unit is used to extract a lane line detection result from the lane image according to a detection model; the detection process of the detection model includes:

[0018] extracting an initial feature map from the lane image;

[0019] Extracting a row anchor confidence feature map and a corresponding row anchor offset feature map, as well as a column anchor confidence feature map and a corresponding column anchor offset feature map from the initial feature map;

[0020] Extracting a plurality of first coordinate points whose confidence is greater than a first threshold from the line anchor confidence feature map, and extracting a plurality of first offsets corresponding to the plurality of first coordinate points from the line anchor offset feature map, performing offset compensation on the plurality of first coordinate points according to the plurality of first offsets, and generating a plurality of line anchor lane lines according to the compensated plurality of first coordinate points;

[0021] Extracting a plurality of second coordinate points whose confidences are greater than the first threshold from the column anchor confidence feature map, and extracting a plurality of second offsets corresponding to the plurality of second coordinate points from the column anchor offset feature map, performing offset compensation on the plurality of second coordinate points according to the plurality of second offsets, and generating a plurality of column anchor lane lines according to the compensated plurality of second coordinate points;

[0022] Calculate the absolute value of the average value between the slope of the target row anchor lane line and the slope of the target column anchor lane line, and when the absolute value is greater than a second threshold, use the target row anchor lane line as the lane line detection result; when the absolute value is less than or equal to the second threshold, use the target column anchor lane line as the lane line detection result; wherein, the target row anchor lane line is any one of the multiple row anchor lane lines, and the target column anchor lane line is an image acquisition unit lane line detection unit corresponding to the target row anchor lane line among the multiple column anchor lane lines.

[0023] The electronic device provided in the third aspect of an embodiment of the present application includes at least one control processor and a memory for communicating with the at least one control processor; the memory stores instructions that can be executed by the at least one control processor, and the instructions are executed by the at least one control processor so that the at least one control processor can execute the lane line detection method based on hybrid coordinate key point estimation.

[0024] A fourth aspect of an embodiment of the present application provides a computer-readable storage medium, which stores computer-executable instructions, and the computer-executable instructions are used to enable a computer to execute the lane line detection method based on hybrid coordinate key point estimation.

[0025] This embodiment has at least the following beneficial effects:

[0026] By extracting row and column anchor confidence feature maps, this method can extract coordinate points with higher confidence; by extracting row and column anchor offset feature maps, the offset corresponding to the coordinate point with higher confidence can be confirmed, and then the coordinate point can be corrected based on the offset to eliminate quantization errors, ultimately improving the accuracy of subsequent detection; in addition, this method also combines the column anchor detection method on the basis of the row anchor detection method, and selects the target row anchor lane line or the target column anchor lane line as the output by judging the size between the absolute value of the average value of the slope of the target row anchor lane line and the slope of the target column anchor lane line and the second threshold, which effectively solves the problem of missed detection caused by the traditional key point estimation method due to the inability to effectively represent the lane line with too large an inclination angle in the scene, thereby improving the accuracy of the final detection of the lane line.

[0027] It can be understood that the beneficial effects of the second to fourth aspects compared with the related art are the same as the beneficial effects of the first aspect compared with the related art. Please refer to the relevant description in the first aspect, and no further details will be given here. BRIEF DESCRIPTION OF THE DRAWINGS

[0028] In order to more clearly illustrate the technical solutions in the embodiments of the present application, the drawings required for use in the embodiments or related technical descriptions will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative labor.

[0029] Figure 1 is a flow chart of a lane line detection method based on hybrid coordinate key point estimation provided by the present application;

[0030] Figure 2 It is a detection flow chart of the detection model provided by this application;

[0031] Figure 3 It is a structural diagram of the lane detection model provided by this application;

[0032] Figure 4 It is a structural diagram of the confidence graph convolution module provided by this application;

[0033] Figure 5 It is a structural diagram of the offset graph convolution module or the auxiliary attention graph convolution module provided in this application;

[0034] Figure 6 is a schematic diagram of a hybrid coordinate detection module provided by the present application;

[0035] Figure 7 is a schematic diagram of an embodiment of neighbor point selection in auxiliary attention loss provided by the present application;

[0036] Figure 8 is a comparison chart of the detection results of the lane line detection method provided in the embodiment of the present application and the detection results of other methods;

[0037] Fig. 9 It is a structural diagram of an electronic device provided in an embodiment of the present application. DETAILED DESCRIPTION

[0038] In order to make the purpose, technical solution and advantages of the present application more clearly understood, the present application is further described in detail below in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present application and are not used to limit the present application.

[0039] It should be noted that, although the functional modules are divided in the device schematic diagram and the logical order is shown in the flowchart, in some cases, the steps shown or described may be performed in a different order than the module division in the device or the order in the flowchart. The terms "first", "second", etc. in the specification, claims and the above drawings are used to distinguish similar objects, and are not necessarily used to describe a specific order or sequence.

[0040] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as those commonly understood by those skilled in the art to which this application belongs. The terms used herein are only for the purpose of describing the embodiments of this application and are not intended to limit this application.

[0041] like Figure 1 and Figure 2 One embodiment of the present application provides a lane line detection method based on hybrid coordinate key point estimation, the method comprising:

[0042] Step S110, acquiring a lane image;

[0043] Step S120, extracting lane line detection results from the lane image according to the detection model.

[0044] As shown in the figure, the detection process of the detection model includes:

[0045] Step S1210, extracting an initial feature map from the lane image;

[0046] Step S1220, extracting a row anchor confidence feature map and a corresponding row anchor offset feature map, as well as a column anchor confidence feature map and a corresponding column anchor offset feature map from the initial feature map;

[0047] Step S1230, extracting a plurality of first coordinate points whose confidence is greater than a first threshold from the line anchor confidence feature map, and extracting a plurality of first offsets corresponding to the plurality of first coordinate points from the line anchor offset feature map, performing offset compensation on the plurality of first coordinate points according to the plurality of first offsets, and generating a plurality of line anchor lane lines according to the compensated plurality of first coordinate points;

[0048] Step S1240, extracting a plurality of second coordinate points whose confidences are greater than a first threshold from the column anchor confidence feature map, and extracting a plurality of second offsets corresponding to the plurality of second coordinate points from the column anchor offset feature map, performing offset compensation on the plurality of second coordinate points according to the plurality of second offsets, and generating a plurality of column anchor lane lines according to the compensated plurality of second coordinate points;

[0049] Step S1250, calculate the absolute value of the average value between the slope of the target row anchor lane line and the slope of the target column anchor lane line, and when the absolute value is greater than a second threshold, use the target row anchor lane line as the lane line detection result; when the absolute value is less than or equal to the second threshold, use the target column anchor lane line as the lane line detection result; wherein, the target row anchor lane line is any one of multiple row anchor lane lines, and the target column anchor lane line is one of the multiple column anchor lane lines corresponding to the target row anchor lane line.

[0050] like Figure 3 In this embodiment, the entire model includes: a backbone network (BoneNet), a neck network (Neck), a detection head (Header), a channel instantiation module and a hybrid coordinate detection module; wherein the detection head includes: a confidence graph convolution module, an offset graph convolution module, and an auxiliary attention graph convolution module.

[0051] In step S110, the lane image may be captured by a vehicle-mounted camera, and the image of the road ahead may be captured by the camera of the smart vehicle.

[0052] In step S120, the detection model is used to detect the lane line in the lane image. It is worth noting that the training process of the detection model is described in detail in the subsequent embodiments, and only the use process is described here. Step S120 includes the following process:

[0053] In step S1210 , a feature map is first extracted from the lane image.

[0054] In some embodiments, a convolutional neural network may be used to extract feature maps.

[0055] In some embodiments, the retrieval model includes: a backbone network and a neck network, the backbone network is a ResNet-18 network, and the neck network includes a channel attention module and a deconvolution layer connected in sequence;

[0056] Extracting the initial feature map from the lane image in step S1210 includes:

[0057] Extract convolutional feature maps from lane images based on the backbone network;

[0058] The convolution feature map is enhanced and upsampled according to the neck network to obtain the initial feature map.

[0059] The basic meaning of ResNet18 is that the basic architecture of the network is ResNet, and the depth of the network is 18 layers. However, the network depth here refers to the weight layer of the network, that is, including the pooling activation layer and the linear layer. ResNet-18 has the advantages of high parameter efficiency, strong feature extraction ability and good training convergence.

[0060] The neck network is mainly for feature enhancement and upsampling. The main advantage of upsampling is that it can increase the sampling rate or resolution of the signal or image, thereby improving its quality and details, and facilitating subsequent feature processing.

[0061] In step S1220, a row anchor confidence feature map, a row anchor offset feature map, a column anchor offset feature map, and a column anchor offset feature map can be extracted from the initial feature map. The difference between row anchors and column anchors is that they are used to represent different lane line parts. Confidence and offset are important concepts in machine learning and computer vision. Confidence usually refers to the degree of certainty of the model in its prediction results, while offset refers to the deviation between the model output and the actual target during the prediction process.

[0062] In some embodiments, the detection model includes: two confidence map convolution modules and two offset map convolution modules;

[0063] The confidence map convolution module includes a sequentially connected convolution layer, a normalization layer, a Relu activation function, a convolution layer, and a Sigmoid activation function; the offset map convolution module includes a sequentially connected convolution layer, a normalization layer, a Relu activation function, and a convolution layer;

[0064] Extracting a row anchor confidence feature map and a corresponding row anchor offset feature map, as well as a column anchor confidence feature map and a corresponding column anchor offset feature map from the initial feature map in step S1220 includes:

[0065] Extracting a row anchor confidence feature map from the feature map according to one of the two confidence map convolution modules;

[0066] Extracting a column anchor confidence feature map from the feature map according to another confidence map convolution module in the two confidence map convolution modules;

[0067] Extracting a row anchor offset feature map from the feature map according to one of the two offset map convolution modules;

[0068] A column anchor offset feature map is extracted from the feature map according to the other offset map convolution module in the two offset map convolution modules.

[0069] like Figure 3 The detection head includes an offset map convolution module 1 and an offset map convolution module 2. The offset map convolution module 1 processes the row anchor offset feature map, and the offset map convolution module 2 processes the column anchor offset feature map.

[0070] The detection head also includes a confidence map convolution module 1 and a confidence map convolution module 2. The confidence map convolution module 1 processes the row anchor confidence feature map, and the confidence map convolution module 2 processes the column anchor confidence feature map.

[0071] In this embodiment, the confidence map convolution module is used to extract row and column anchor confidence feature maps, and then the coordinate points (i.e., pixels) with higher confidence can be extracted. The use of coordinate points with higher confidence can improve the recognition accuracy; the offset map convolution module can be used to extract row and column anchor offset feature maps, and then the offset corresponding to the coordinate points with higher confidence can be confirmed, and then correction can be made based on the offset to eliminate quantization errors, thereby ultimately improving the accuracy of subsequent detection.

[0072] In step S1230, for the anchor lane line, first, multiple first coordinate points whose confidence is greater than a first threshold are extracted from the anchor confidence feature map; then, multiple first offsets corresponding to the multiple first coordinate points are extracted from the anchor offset feature map; secondly, offset compensation is performed on the multiple first coordinate points according to the multiple first offsets; finally, multiple anchor lane lines are generated according to the compensated multiple first coordinate points.

[0073] In some embodiments, both the row anchor confidence feature map and the row anchor offset feature map are N-channel feature maps;

[0074] The detection module also includes: a channel instantiation module;

[0075] Step S1230 extracts a plurality of first coordinate points whose confidence is greater than a first threshold from the line anchor confidence feature map, extracts a plurality of first offsets corresponding to the plurality of first coordinate points from the line anchor offset feature map, performs offset compensation on the plurality of first coordinate points according to the plurality of first offsets, and generates a plurality of line anchor lane lines according to the compensated plurality of first coordinate points, including:

[0076] Extracting M1 first coordinate points whose confidence is greater than a first threshold from the row anchor confidence feature map of the target single channel according to the channel instantiation module, wherein the row anchor confidence feature map of the target single channel is the row anchor confidence feature map of any single channel in the row anchor confidence feature maps of the N channels;

[0077] Extracting M1 first offsets corresponding to M1 first coordinate points from the row anchor offset feature map of the target single channel according to the channel instantiation module, the row anchor offset feature map of the target single channel is a row anchor offset feature map of the single channel corresponding to the row anchor confidence feature map of the target single channel in the row anchor offset feature map of the N channels;

[0078] According to the channel instantiation module, offset compensation is performed on the corresponding M1 first coordinate points using M1 first offsets to obtain compensated M1 anchor lane points;

[0079] According to the channel instantiation module, the anchor lane line corresponding to the anchor confidence feature map of the target single channel is obtained by performing polynomial fitting on the compensated M1 anchor lane points;

[0080] According to the channel instantiation module, N single-channel anchor confidence feature maps are traversed to obtain N anchor lane lines.

[0081] In this embodiment, the anchor confidence feature map and the anchor offset feature map are both N single-channel feature maps, and the feature map of one channel corresponds to the anchor confidence feature map and anchor offset feature map of one lane line. The feature map of each channel can obtain M1 anchor lane points, and then fit to obtain the anchor lane corresponding to the channel. N anchor lanes can be generated corresponding to N channels; M1 and N are both positive integers.

[0082] In step S1240, for the column anchor lane line, firstly, a plurality of second coordinate points whose confidence is greater than a first threshold are extracted from the column anchor confidence feature map; then, a plurality of second offsets corresponding to the plurality of second coordinate points are extracted from the column anchor offset feature map; then, offset compensation is performed on the plurality of second coordinate points according to the plurality of second offsets; and finally, a plurality of column anchor lane lines are generated according to the plurality of compensated second coordinate points;

[0083] In some embodiments, the column anchor confidence feature map and the column anchor offset feature map are both N-channel feature maps;

[0084] Step S1240 extracts a plurality of second coordinate points whose confidence is greater than a first threshold from the column anchor confidence feature map, extracts a plurality of second offsets corresponding to the plurality of second coordinate points from the column anchor offset feature map, performs offset compensation on the plurality of second coordinate points according to the plurality of second offsets, and generates a plurality of column anchor lane lines according to the compensated plurality of second coordinate points, including:

[0085] Extracting M2 second coordinate points whose confidence is greater than a first threshold from the column anchor confidence feature map of the target single channel according to the channel instantiation module, wherein the column anchor confidence feature map of the target single channel is a column anchor confidence feature map of any single channel in the column anchor confidence feature maps of the N channels;

[0086] Extracting M2 second offsets corresponding to the M2 second coordinate points from the column anchor offset feature map of the target single channel according to the channel instantiation module, the column anchor offset feature map of the target single channel is a column anchor offset feature map of the single channel corresponding to the column anchor confidence feature map of the target single channel in the column anchor offset feature map of the N channels;

[0087] According to the channel instantiation module, the M2 second offsets are used to compensate the corresponding M2 second coordinate points, so as to obtain the compensated M2 column anchor lane points;

[0088] According to the channel instantiation module, the column anchor lane line corresponding to the column anchor confidence feature map of the target single channel is obtained by performing polynomial fitting on the compensated M2 column anchor lane points;

[0089] According to the channel instantiation module, the column anchor confidence feature maps of N single channels are traversed to obtain N column anchor lane lines.

[0090] In this embodiment, the column anchor confidence feature map and the column anchor offset feature map are both N single-channel feature maps, and the feature map of one channel corresponds to the column anchor confidence feature map and column anchor offset feature map of one lane line. The feature map of each channel can obtain M1 column anchor lane points, and then perform column fitting to obtain the column anchor lane corresponding to the channel. N column anchor lanes can be generated corresponding to N channels; where M2 is a positive integer.

[0091] In step S1250, when the absolute value of the average value (i.e., the difference between the average slopes) of the slope of the target row anchor lane line and the slope of the target column anchor lane line is greater than a second threshold, the target row anchor lane line is used as the lane line detection result; when the absolute value is less than or equal to the second threshold, the target column anchor lane line is used as the lane line detection result.

[0092] In this step, a second threshold is set first. If the absolute value of the difference between the average slope of the row anchor lane line and the corresponding column anchor lane line is greater than the second threshold, the row anchor lane line or the column anchor lane line is selected.

[0093] In some embodiments, the detection module further includes: the detection model further includes: a hybrid coordinate detection module;

[0094] Step S1250 calculates the absolute value of the average value between the slope of the target row anchor lane line and the slope of the target column anchor lane line, and when the absolute value is greater than a second threshold, uses the target row anchor lane line as the lane line detection result; when the absolute value is less than or equal to the second threshold, uses the target column anchor lane line as the lane line detection result, including:

[0095] Calculate the absolute value of the average between the slope of the target row anchor lane line and the slope of the target column anchor lane line according to the hybrid coordinate detection module;

[0096] The hybrid coordinate detection module determines the size between the absolute value and the second threshold value. If the absolute value is greater than the second threshold value, the target row anchor lane line is used as the target lane line; if the absolute value is less than or equal to the second threshold value, the target column anchor lane line is used as the target lane line;

[0097] According to the hybrid coordinate detection module, N row anchor lane lines and the corresponding N column anchor lane lines are traversed to obtain N target lane lines, and the N target lane lines are used as lane line detection results.

[0098] The lane line detection method based on hybrid coordinate key point estimation provided in this embodiment has at least the following benefits:

[0099] By extracting row and column anchor confidence feature maps, this method can extract coordinate points with higher confidence; by extracting row and column anchor offset feature maps, the offset corresponding to the coordinate point with higher confidence can be confirmed, and then the coordinate point can be corrected based on the offset to eliminate quantization errors, ultimately improving the accuracy of subsequent detection; in addition, this method also combines the column anchor detection method on the basis of the row anchor detection method, and selects the target row anchor lane line or the target column anchor lane line as the output by judging the size between the absolute value of the average value of the slope of the target row anchor lane line and the slope of the target column anchor lane line and the second threshold, which effectively solves the problem of missed detection caused by the traditional key point estimation method due to the inability to effectively represent the lane line with too large an inclination angle in the scene, thereby improving the accuracy of the final detection of the lane line.

[0100] like Figures 3 to 8 , this embodiment provides a lane line detection method based on hybrid coordinate key point estimation, comprising the following steps:

[0101] Step S910, obtaining a data set and a detection label;

[0102] Get the lane line dataset and the corresponding detection labels.

[0103] Step S920, constructing a lane line detection model;

[0104] The lane detection model consists of the backbone network BoneNet, the neck network Neck, the detection head Header, the channel instantiation module and the hybrid coordinate detection module. The specific construction process includes the following steps:

[0105] Step S9210, construct the backbone network BoneNet: ResNet-18 is used as the backbone network, and the input image is processed by the backbone network to obtain the convolution feature map .

[0106] Step S9220, construct the neck network Neck: the network consists of a channel attention module and a deconvolution layer; the convolution feature map obtained in step S9210 Input the channel attention module and deconvolution layer in sequence to get the output of the neck network Neck .

[0107] Step S9230, construct a detection header: the detection head consists of a confidence map convolution module 1, an offset map convolution module 1, a confidence map convolution module 2, an offset map convolution module 2 and an auxiliary attention map convolution module; wherein the confidence map convolution module 1 and the confidence map convolution module 2 have the same structure, both of which are constructed in sequence by a convolution layer, a normalization layer, a Relu activation function, a convolution layer and a Sigmoid activation function; the offset map convolution module 1 and the offset map convolution module 2 have the same structure, both of which are constructed in sequence by a convolution layer, a normalization layer, a Relu activation function and a convolution layer; the auxiliary attention map convolution module is constructed in sequence by a convolution layer, a normalization layer, a Relu activation function and a convolution layer.

[0108] The result obtained in step S9210 Input the confidence map convolution modules 1 and 2 respectively to obtain the row anchor confidence feature map And column anchor confidence feature map ;

[0109] Will Input the offset map convolution modules 1 and 2 respectively to obtain the row anchor offset feature map and column anchor offset feature map ;

[0110] Will Input the auxiliary attention graph convolution module to obtain the auxiliary attention feature map ;in, , , and They are all multi-channel feature maps, and the number of channels is N , is a dual-channel feature map, feature map , , , and All have the same height and width; the number of channels N Preferred option 4.

[0111] Step S9240, constructing a channel instantiation module;

[0112] The obtained in step S9230 and According to the corresponding channels, it is divided into N For single-channel feature maps, each pair of single-channel feature maps corresponds to a lane line anchor confidence feature map and a lane anchor offset feature map; for each pair of single-channel feature maps, first select all lane anchor confidence feature maps with confidence greater than the first threshold. P The coordinate points are taken as the anchor lane points, and all the anchor lane points are combined into an anchor lane point list Then, the offset values ​​corresponding to these row anchor lane points are obtained from the row anchor offset feature map, and these offset values ​​are combined into a row anchor offset list. ;use Offset value compensation in The horizontal coordinate value of the corresponding anchor lane point in , thereby eliminating the quantization error of the horizontal coordinate caused by downsampling, and obtaining a new list of anchor lane points , the compensation formula is as follows:

[0113] ;

[0114] in, , for Middle i The horizontal and vertical coordinate values ​​of the anchor lane points, for Middle and i The offset value corresponding to the anchor lane point, , for Middle i The horizontal and vertical coordinate values ​​of the anchor lane points; Perform polynomial fitting to obtain a detection result of the anchor lane line; traverse using the above method N For a single channel feature map, we get N The anchor lane line detection results are combined into an anchor lane line list ; P A real number, preferably 0.35.

[0115] The obtained in step S9230 and According to the corresponding channels, it is divided into N For single-channel feature maps, each pair of single-channel feature maps corresponds to a column anchor confidence feature map and a column anchor offset feature map of a lane line;

[0116] For each pair of single-channel feature maps, first select all the features with confidence greater than the threshold from the column anchor confidence feature map. P The coordinate points are used as column anchor lane points, and all column anchor lane points are combined into a column anchor lane point list , then obtain the offset values ​​corresponding to these column anchor lane points from the column anchor offset feature map, and combine these offset values ​​into a column anchor offset list ;use Offset value compensation in The ordinate value of the corresponding column anchor lane point in , thereby eliminating the quantization error of the ordinate caused by downsampling, and obtaining a new column anchor lane point list , the compensation formula is as follows:

[0117] ;

[0118] in, , for Middle i The horizontal and vertical coordinate values ​​of the anchor lane points, for Middle and i The offset value corresponding to the anchor lane point, , for Middle i The horizontal and vertical coordinate values ​​of the anchor lane points in each column; Perform polynomial fitting to obtain a lane line detection result; traverse using the above method N For a single channel feature map, we get N Column anchor lane line detection results, these detection results are combined into a column anchor lane line list .

[0119] Step S9250, constructing a hybrid coordinate detection module;

[0120] The step S9240 obtained and The corresponding row anchor lane line and column anchor lane line are selected, and the selection rule is: when the absolute value of the average slope of the row anchor lane line and the corresponding column anchor lane line is greater than the set second threshold T When Lane line in, otherwise select The lane lines in N The lane line detection results composed of lane lines are used to complete the construction of the lane line detection model. T A real number, preferably 0.45.

[0121] Step S930, constructing a loss function;

[0122] Construct the following hybrid loss function :

[0123] ;

[0124] in is the key point detection loss function, To assist in focusing on the loss function.

[0125] The specific construction process includes the following steps:

[0126] Step S9310, constructing a key point detection loss function :

[0127] ;

[0128] in, is the row anchor confidence loss, which represents the loss between the row anchor confidence prediction result and the row anchor confidence label; is the column anchor confidence loss, which represents the loss between the column anchor confidence prediction result and the column anchor confidence label; is the row anchor offset loss, which represents the loss between the row anchor offset prediction result and the row anchor offset label; is the column anchor offset loss, which represents the loss between the column anchor offset prediction result and the column anchor offset label.

[0129] Row Anchor Confidence Loss The build is as follows:

[0130] ;

[0131] in, and They are row anchor confidence feature maps The height and width of Represents the row anchor confidence feature map On point The value at Represents the row anchor confidence label On point The value at α and β represents an adjustable hyperparameter; It means taking the natural logarithm; α Preferred 2, β Preferred option 4.

[0132] Column Anchor Confidence Loss The build is as follows:

[0133] ;

[0134] in, Represents the column anchor confidence feature map On point The value at Represents column anchor confidence labels On point The value at .

[0135] Row anchor offset loss The build is as follows:

[0136] ;

[0137] in, Represents the row anchor offset feature map On point The value at Indicates the row anchor offset label On point The value at Indicates taking the absolute value.

[0138] Column Anchor Offset Loss The build is as follows:

[0139] ;

[0140] in, Represents the column anchor offset feature map On point The value at Indicates the column anchor offset label On point The value at .

[0141] Step S9320, construct auxiliary attention loss function :

[0142] ;

[0143] in, , Respectively represented in the auxiliary attention feature map midpoint The distance prediction value obtained by row sampling and column sampling of neighboring points is , Respectively represent the coordinate points in the preset auxiliary attention tag ( ) is used to obtain the true distance value obtained by row sampling and column sampling of neighboring points. Indicates the coordinate point in the auxiliary attention tag ( ), where the neighbor point is the lane line point in the input image that is closest to the current lane line point and located above the current lane line point.

[0144] Step S940: training a lane line detection model:

[0145] The lane detection model constructed in step S920 is trained using the data set obtained in step S910; the hybrid loss function constructed in step S930 is used Calculate the error between the prediction result output by the model and the label; use the Adam algorithm to update the model parameters during the training process, and use L-2 regularization as a constraint until the loss no longer decreases, and obtain a trained lane line detection model.

[0146] Step S950: Lane line detection:

[0147] After the test image is normalized, it is input into the trained lane line detection model, and the model output result is the final lane line detection result.

[0148] The following experimental examples are provided:

[0149] The above method is used to conduct lane detection experiments on the Tusimple dataset. The experimental operating system is Ubuntu 18.04, the PyTorch1.2.0 framework based on CUDA10.0 and cuDNN7.6.0, and the training and testing are performed on a personal computer equipped with an Intel Corei5-12600KF CPU (3.70GHz) and NVIDIA GeForce RTX 3090 (24GB) hardware.

[0150] This example uses five indicators, F1, ACC, FP, FN, and FPS, to conduct experimental comparisons between the four test methods, SAD, UFAST-v1, UFAST-v2, and BezierLaneNet, and the method of this embodiment on the Tusimple dataset.

[0151] The comparison results are shown in Table 1. It can be found that compared with other methods, this embodiment can obtain more accurate detection results on the Tusimple dataset and achieve the best evaluation indicators F1 and FP. For details, please refer to the following content:

[0152] Table 1

[0153]

[0154] Figure 8 The following is a comparison chart of the lane line detection results of this embodiment with those of BezierLaneNet and UFAST-v2. The results show that the model designed in this embodiment can handle various challenging scenarios well. Compared with other methods, the lane line detection results obtained by this method are more accurate.

[0155] Compared with the prior art, this embodiment has the following advantages:

[0156] This embodiment combines the column anchor detection method with the row anchor detection method, which effectively solves the missed detection problem caused by the traditional key point estimation method that cannot effectively represent the lane lines with too large inclination angles in the scene.

[0157] The channel instantiation module constructed in this embodiment efficiently instantiates lane lines in the channel dimension through channel indexes, thereby avoiding a large number of post-processing operations in traditional lane line detection methods.

[0158] This embodiment designs an auxiliary attention loss, which enhances the network's learning of the position information between lane points, thereby achieving the detection of lane lines whose visual features are not obvious at the far end.

[0159] An embodiment of the present application provides a lane line detection device based on hybrid coordinate key point estimation, the device comprising:

[0160] An image acquisition unit, used for acquiring a lane image;

[0161] The lane line detection unit is used to extract the lane line detection result from the lane image according to the detection model; the detection process of the detection model includes:

[0162] Extracting an initial feature map from the lane image;

[0163] Extracting a row anchor confidence feature map and a corresponding row anchor offset feature map, as well as a column anchor confidence feature map and a corresponding column anchor offset feature map from the initial feature map;

[0164] Extracting a plurality of first coordinate points whose confidence is greater than a first threshold from the line anchor confidence feature map, and extracting a plurality of first offsets corresponding to the plurality of first coordinate points from the line anchor offset feature map, performing offset compensation on the plurality of first coordinate points according to the plurality of first offsets, and generating a plurality of line anchor lane lines according to the compensated plurality of first coordinate points;

[0165] Extracting a plurality of second coordinate points whose confidences are greater than a first threshold from the column anchor confidence feature map, and extracting a plurality of second offsets corresponding to the plurality of second coordinate points from the column anchor offset feature map, performing offset compensation on the plurality of second coordinate points according to the plurality of second offsets, and generating a plurality of column anchor lane lines according to the compensated plurality of second coordinate points;

[0166] The absolute value of the average value between the slope of the target row anchor lane line and the slope of the target column anchor lane line is calculated, and when the absolute value is greater than a second threshold, the target row anchor lane line is used as the lane line detection result; when the absolute value is less than or equal to the second threshold, the target column anchor lane line is used as the lane line detection result; wherein, the target row anchor lane line is any one of the multiple row anchor lane lines, and the target column anchor lane line is one of the multiple column anchor lane lines corresponding to the target row anchor lane line.

[0167] It should be noted that the lane line detection device based on hybrid coordinate key point estimation provided in this embodiment and the lane line detection method based on hybrid coordinate key point estimation mentioned above are based on the same inventive concept. Therefore, the relevant content of the lane line detection method based on hybrid coordinate key point estimation mentioned above is also applicable to the content of the lane line detection device based on hybrid coordinate key point estimation. Therefore, it will not be repeated here.

[0168] like Fig. 9 The present application also provides an electronic device, which includes:

[0169] at least one memory;

[0170] at least one processor;

[0171] at least one program;

[0172] The program is stored in the memory, and the processor executes at least one program to implement the lane line detection method based on hybrid coordinate key point estimation described above in the present disclosure.

[0173] The electronic device may be any intelligent terminal including a mobile phone, a tablet computer, a personal digital assistant (PDA), a vehicle-mounted computer, etc.

[0174] The electronic device according to the embodiment of the present application is described in detail below.

[0175] The processor 1600 may be implemented by a general-purpose central processing unit (CPU), a microprocessor, an application-specific integrated circuit (ASIC), or one or more integrated circuits, and is used to execute relevant programs to implement the technical solutions provided in the embodiments of the present invention.

[0176] The memory 1700 may be implemented in the form of a read-only memory (ROM), a static storage device, a dynamic storage device, or a random access memory (RAM). The memory 1700 may store an operating system and other application programs. When the technical solution provided in the embodiments of this specification is implemented by software or firmware, the relevant program code is stored in the memory 1700, and the processor 1600 calls and executes the lane line detection method based on hybrid coordinate key point estimation in the embodiment of the present invention.

[0177] Input / output interface 1800, used to implement information input and output;

[0178] Communication interface 1900, used to realize communication interaction between the device and other devices, which can be realized through wired mode (such as USB, network cable, etc.) or wireless mode (such as mobile network, WIFI, Bluetooth, etc.);

[0179] Bus 2000 , which transmits information between various components of the device (e.g., processor 1600 , memory 1700 , input / output interface 1800 , and communication interface 1900 );

[0180] The processor 1600 , the memory 1700 , the input / output interface 1800 , and the communication interface 1900 are connected to each other in communication within the device via the bus 2000 .

[0181] An embodiment of the present invention further provides a storage medium, which is a computer-readable storage medium. The computer-readable storage medium stores computer-executable instructions, and the computer-executable instructions are used to enable a computer to execute the above-mentioned lane line detection method based on hybrid coordinate key point estimation.

[0182] The memory, as a non-transient computer-readable storage medium, can be used to store non-transient software programs and non-transient computer executable programs. In addition, the memory may include a high-speed random access memory, and may also include a non-transient memory, such as at least one disk storage device, a flash memory device, or other non-transient solid-state storage device. In some embodiments, the memory may optionally include a memory remotely disposed relative to the processor, and these remote memories may be connected to the processor via a network. Examples of the above-mentioned network include, but are not limited to, the Internet, an intranet, a local area network, a mobile communication network, and combinations thereof.

[0183] The embodiments described in the present invention are intended to more clearly illustrate the technical solutions of the embodiments of the present invention and do not constitute a limitation on the technical solutions provided by the embodiments of the present invention. Those skilled in the art will appreciate that with the evolution of technology and the emergence of new application scenarios, the technical solutions provided by the embodiments of the present invention are also applicable to similar technical problems.

[0184] Those skilled in the art will appreciate that the technical solutions shown in the figures do not limit the embodiments of the present invention and may include more or fewer steps than shown in the figures, or a combination of certain steps, or different steps.

[0185] The device embodiments described above are merely illustrative, and the units described as separate components may or may not be physically separated, that is, they may be located in one place or distributed on multiple network units. Some or all of the modules may be selected according to actual needs to achieve the purpose of the solution of this embodiment.

[0186] Those skilled in the art will appreciate that all or some of the steps in the methods disclosed above, and the functional modules / units in the systems and devices may be implemented as software, firmware, hardware, or a suitable combination thereof.

[0187] The terms "first", "second", "third", "fourth", etc. (if any) in the specification of the present application and the above-mentioned drawings are used to distinguish similar objects, and are not necessarily used to describe a specific order or sequence. It should be understood that the data used in this way can be interchangeable where appropriate, so that the embodiments of the present application described herein can be implemented in an order other than those illustrated or described herein. In addition, the terms "including" and "having" and any of their variations are intended to cover non-exclusive inclusions, for example, a process, method, system, product or device comprising a series of steps or units is not necessarily limited to those steps or units clearly listed, but may include other steps or units that are not clearly listed or inherent to these processes, methods, products or devices.

[0188] In the several embodiments provided in the present application, it should be understood that the disclosed devices and methods can be implemented in other ways. For example, the device embodiments described above are only schematic. For example, the division of units is only a logical function division. There may be other division methods in actual implementation, such as multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the mutual coupling or direct coupling or communication connection shown or discussed can be through some interfaces, indirect coupling or communication connection of devices or units, which can be electrical, mechanical or other forms.

[0189] The units described as separate components may or may not be physically separated, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed on multiple network units. Some or all of the units may be selected according to actual needs to achieve the purpose of the solution of this embodiment.

[0190] In addition, each functional unit in each embodiment of the present application may be integrated into one processing unit, or each unit may exist physically separately, or two or more units may be integrated into one unit. The above-mentioned integrated unit may be implemented in the form of hardware or in the form of software functional units.

[0191] If the integrated unit is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present application is essentially or the part that contributes to the prior art or all or part of the technical solution can be embodied in the form of a software product, and the computer software product is stored in a storage medium, including multiple instructions to enable an electronic device (which can be a personal computer, server, or network device, etc.) to perform all or part of the steps of each embodiment method of the present application. The aforementioned storage medium includes: U disk, mobile hard disk, read-only memory (Read-Only Memory, ROM), random access memory (Random Access Memory, RAM), disk or optical disk and other media that can store programs.

[0192] The above is a specific description of the preferred implementation of the embodiments of the present application, but the embodiments of the present application are not limited to the above-mentioned implementation methods. Technical personnel familiar with the field can also make various equivalent modifications or substitutions without violating the spirit of the embodiments of the present application. These equivalent modifications or substitutions are all included in the scope defined by the claims of the embodiments of the present application.

Claims

1. A lane line detection method based on hybrid coordinate key point estimation, characterized in that: The method comprises: Get lane image; Extracting a lane line detection result from the lane image according to the detection model; the detection process of the detection model includes: extracting an initial feature map from the lane image; Extracting a row anchor confidence feature map and a corresponding row anchor offset feature map, as well as a column anchor confidence feature map and a corresponding column anchor offset feature map from the initial feature map; Extracting a plurality of first coordinate points whose confidence is greater than a first threshold from the line anchor confidence feature map, and extracting a plurality of first offsets corresponding to the plurality of first coordinate points from the line anchor offset feature map, performing offset compensation on the plurality of first coordinate points according to the plurality of first offsets, and generating a plurality of line anchor lane lines according to the compensated plurality of first coordinate points; Extracting a plurality of second coordinate points whose confidences are greater than the first threshold from the column anchor confidence feature map, and extracting a plurality of second offsets corresponding to the plurality of second coordinate points from the column anchor offset feature map, performing offset compensation on the plurality of second coordinate points according to the plurality of second offsets, and generating a plurality of column anchor lane lines according to the compensated plurality of second coordinate points; Calculate the absolute value of the average value between the slope of the target row anchor lane line and the slope of the target column anchor lane line, and when the absolute value is greater than a second threshold, use the target row anchor lane line as the lane line detection result; when the absolute value is less than or equal to the second threshold, use the target column anchor lane line as the lane line detection result; wherein, the target row anchor lane line is any one of the multiple row anchor lane lines, and the target column anchor lane line is one of the multiple column anchor lane lines corresponding to the target row anchor lane line.

2. The lane line detection method based on hybrid coordinate key point estimation according to claim 1, characterized in that: The detection model includes a backbone network and a neck network, the backbone network is a ResNet-18 network, and the neck network includes a channel attention module and a deconvolution layer connected in sequence; The extracting an initial feature map from the lane image comprises: Extracting a convolutional feature map from the lane image according to the backbone network; The convolution feature map is enhanced and up-sampled according to the neck network to obtain an initial feature map.

3. The lane line detection method based on hybrid coordinate key point estimation according to claim 1, characterized in that: The detection model includes: two confidence graph convolution modules and two offset graph convolution modules; The confidence map convolution module includes a convolution layer, a normalization layer, a Relu activation function, a convolution layer and a Sigmoid activation function connected in sequence; the offset map convolution module includes a convolution layer, a normalization layer, a Relu activation function and a convolution layer connected in sequence; The extracting a row anchor confidence feature map and a corresponding row anchor offset feature map, and a column anchor confidence feature map and a corresponding column anchor offset feature map from the initial feature map includes: Extracting a row anchor confidence feature map from the feature map according to one of the two confidence map convolution modules; Extracting a column anchor confidence feature map from the feature map according to another of the two confidence map convolution modules; Extracting a row anchor offset feature map from the feature map according to one of the two offset map convolution modules; A column anchor offset feature map is extracted from the feature map according to the other offset map convolution module of the two offset map convolution modules.

4. The lane line detection method based on hybrid coordinate key point estimation according to claim 3, characterized in that: The row anchor confidence feature map and the row anchor offset feature map are both N-channel feature maps; The detection model also includes: a channel instantiation module; The method comprises extracting a plurality of first coordinate points whose confidence is greater than a first threshold from the line anchor confidence feature map, extracting a plurality of first offsets corresponding to the plurality of first coordinate points from the line anchor offset feature map, performing offset compensation on the plurality of first coordinate points according to the plurality of first offsets, and generating a plurality of line anchor lane lines according to the compensated plurality of first coordinate points, including: Extracting M1 first coordinate points whose confidence is greater than a first threshold from the row anchor confidence feature map of the target single channel according to the channel instantiation module, wherein the row anchor confidence feature map of the target single channel is a row anchor confidence feature map of any single channel in the row anchor confidence feature maps of N channels; Extracting M1 first offsets corresponding to the M1 first coordinate points from the row anchor offset feature map of the target single channel according to the channel instantiation module, wherein the row anchor offset feature map of the target single channel is a row anchor offset feature map of the single channel corresponding to the row anchor confidence feature map of the target single channel in the row anchor offset feature maps of the N channels; According to the channel instantiation module, offset compensation is performed on the corresponding M1 first coordinate points using the M1 first offsets to obtain compensated M1 anchor lane points; According to the channel instantiation module, a polynomial fitting is performed on the compensated M1 anchor lane points to obtain an anchor lane line corresponding to the anchor confidence feature map of the target single channel; According to the channel instantiation module, N single-channel anchor confidence feature maps are traversed to obtain N anchor lane lines; M1 and N are both positive integers.

5. The lane line detection method based on hybrid coordinate key point estimation according to claim 4, characterized in that: The column anchor confidence feature map and the column anchor offset feature map are both N-channel feature maps; The step of extracting a plurality of second coordinate points whose confidence is greater than the first threshold from the column anchor confidence feature map, extracting a plurality of second offsets corresponding to the plurality of second coordinate points from the column anchor offset feature map, performing offset compensation on the plurality of second coordinate points according to the plurality of second offsets, and generating a plurality of column anchor lane lines according to the compensated plurality of second coordinate points includes: Extracting M2 second coordinate points whose confidence is greater than a first threshold from the column anchor confidence feature map of the target single channel according to the channel instantiation module, wherein the column anchor confidence feature map of the target single channel is a column anchor confidence feature map of any single channel in the column anchor confidence feature maps of N channels; Extracting M2 second offsets corresponding to the M2 second coordinate points from the column anchor offset feature map of the target single channel according to the channel instantiation module, wherein the column anchor offset feature map of the target single channel is a column anchor offset feature map of the single channel corresponding to the column anchor confidence feature map of the target single channel in the column anchor offset feature map of the N channels; According to the channel instantiation module, using the M2 second offsets, offset compensation is performed on the corresponding M2 second coordinate points to obtain compensated M2 column anchor lane points; According to the channel instantiation module, by performing polynomial fitting on the compensated M2 column anchor lane points, the column anchor lane line corresponding to the column anchor confidence feature map of the target single channel is obtained; According to the channel instantiation module, N single-channel column anchor confidence feature maps are traversed to obtain N column anchor lane lines; M2 is a positive integer.

6. The lane line detection method based on hybrid coordinate key point estimation according to claim 5, characterized in that: The detection model also includes: a hybrid coordinate detection module; The calculating the absolute value of the average value between the slope of the target row anchor lane line and the slope of the target column anchor lane line, and taking the target row anchor lane line as the lane line detection result when the absolute value is greater than a second threshold; and taking the target column anchor lane line as the lane line detection result when the absolute value is less than or equal to the second threshold, includes: Calculating an absolute value of an average value between a slope of a target row anchor lane line and a slope of a target column anchor lane line according to the hybrid coordinate detection module; Determine the size between the absolute value and the second threshold value according to the hybrid coordinate detection module, and if the absolute value is greater than the second threshold value, use the target row anchor lane line as the target lane line; if the absolute value is less than or equal to the second threshold value, use the target column anchor lane line as the target lane line; The N row anchor lane lines and the corresponding N column anchor lane lines are traversed according to the hybrid coordinate detection module to obtain N target lane lines, and the N target lane lines are used as the lane line detection results.

7. The lane line detection method based on hybrid coordinate key point estimation according to claim 1, characterized in that: The detection model includes an auxiliary attention graph convolution module, and the auxiliary attention graph convolution module includes a convolution layer, a normalization layer, a Relu activation function and a convolution layer connected in sequence; After extracting the row anchor confidence feature map and the corresponding row anchor offset feature map, and the column anchor confidence feature map and the corresponding column anchor offset feature map from the initial feature map, the method further includes: Extracting an auxiliary attention feature map from the initial feature map according to the auxiliary attention map convolution module; The loss function of the detection model is the sum of the key point detection loss function and the auxiliary attention loss function; The key point detection loss function is the sum of the row anchor confidence loss function, the column anchor confidence loss function, the row anchor offset loss function and the column anchor offset loss function; The auxiliary attention loss function for: ； in, , Respectively represent the height and width of the auxiliary attention feature map, , Respectively represent the coordinate points in the auxiliary attention feature map ( ) is used to obtain the predicted distance value by row sampling and column sampling of the neighboring points. , Respectively represent the coordinate points in the preset auxiliary attention tag ( ) is used to obtain the true distance value obtained by row sampling and column sampling of neighboring points. Indicates the coordinate point in the auxiliary attention tag ( ) at .

8. A lane line detection device based on hybrid coordinate key point estimation, characterized in that: The device comprises: An image acquisition unit, used for acquiring a lane image; A lane line detection unit is used to extract a lane line detection result from the lane image according to a detection model; the detection process of the detection model includes: extracting an initial feature map from the lane image; Extracting a row anchor confidence feature map and a corresponding row anchor offset feature map, as well as a column anchor confidence feature map and a corresponding column anchor offset feature map from the initial feature map; Extracting a plurality of first coordinate points whose confidence is greater than a first threshold from the line anchor confidence feature map, and extracting a plurality of first offsets corresponding to the plurality of first coordinate points from the line anchor offset feature map, performing offset compensation on the plurality of first coordinate points according to the plurality of first offsets, and generating a plurality of line anchor lane lines according to the compensated plurality of first coordinate points; Extracting a plurality of second coordinate points whose confidences are greater than the first threshold from the column anchor confidence feature map, and extracting a plurality of second offsets corresponding to the plurality of second coordinate points from the column anchor offset feature map, performing offset compensation on the plurality of second coordinate points according to the plurality of second offsets, and generating a plurality of column anchor lane lines according to the compensated plurality of second coordinate points; Calculate the absolute value of the average value between the slope of the target row anchor lane line and the slope of the target column anchor lane line, and when the absolute value is greater than a second threshold, use the target row anchor lane line as the lane line detection result; when the absolute value is less than or equal to the second threshold, use the target column anchor lane line as the lane line detection result; wherein, the target row anchor lane line is any one of the multiple row anchor lane lines, and the target column anchor lane line is one of the multiple column anchor lane lines corresponding to the target row anchor lane line.

9. An electronic device, characterized in that: It includes at least one control processor and a memory for communicating with the at least one control processor; the memory stores instructions that can be executed by the at least one control processor, and the instructions are executed by the at least one control processor so that the at least one control processor can execute the lane line detection method based on hybrid coordinate key point estimation as described in any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores computer-executable instructions, and the computer-executable instructions are used to enable a computer to execute the lane line detection method based on hybrid coordinate key point estimation as described in any one of claims 1 to 7.

Images