Centerline Detection Method, Device, Electronic Device, Storage Medium and Product

Abstract

The present invention provides a center line detection method, apparatus, electronic device, storage medium and product. The method includes: obtaining an image to be processed; inputting the image to be processed into a pre-trained center point prediction model to obtain a plurality of candidate center points; calculating a confidence value for each of the plurality of candidate center points to obtain a plurality of confidence values; when a first confidence value is greater than a first preset threshold, performing a regression operation on a first candidate center point corresponding to the first confidence value to determine that the first candidate center point is located on the center line of the lane line. The present invention realizes the prediction of candidate center points other than the center point on the center line through a preset center point prediction model, and performs a regression operation on the candidate center points, ensuring the accuracy of center line detection when the offset is too large, ensuring the safety of the autonomous driving vehicle during driving, and improving the passenger experience.

Description

Technical Field

[0001] The present invention relates to the technical field of lane line detection, and particularly to a center line detection method, device, electronic device, storage medium and product. Background Art

[0002] With the continuous development of society and the continuous improvement of living standards, the number of vehicles for convenient travel is increasing, and people's requirements for current intelligent transportation services are also getting higher and higher.

[0003] In the prior art, in vehicle lane line detection, a segmentation network is often used to determine the center line, that is, after the network is segmented, the center line of the lane line is determined according to the corresponding initial point by predicting the line type. This detection method often results in inaccurate position information of the predicted center point due to inaccurate confirmation of the initial point. Moreover, for the case where the offset is too large, the center point cannot be regressed, resulting in inaccurate prediction of the center line, causing the vehicle to deviate during driving, with low driving safety and poor passenger experience. Summary of the Invention

[0004] The present invention provides a center line detection method, device, electronic device, storage medium and product, aiming to solve the technical problems in the prior art that the detection accuracy of the lane line center line is not high and the passenger experience is poor because the center line cannot be regressed by using the method of predicting the line type, and to achieve the purpose of improving the detection accuracy of the lane line center line, ensuring the safety of autonomous vehicle driving, and enhancing the passenger experience.

[0005] In a first aspect, the present invention provides a center line detection method, including:

[0006] Obtain an image to be processed; wherein, the image to be processed is an image collected in real time during driving;

[0007] Input the image to be processed into a pre-trained center point prediction model to obtain a plurality of candidate center points;

[0008] Calculate the confidence of each candidate center point among the plurality of candidate center points to obtain a plurality of confidence values;

[0009] When a first confidence value is greater than a first preset threshold, perform a regression operation on the first candidate center point corresponding to the first confidence value to determine that the first candidate center point is located on the center line of the lane line;

[0010] Wherein, the first confidence value is any one of the plurality of confidence values, and the first candidate center point is any one of the plurality of candidate center points;

[0011] Wherein, the center point prediction model is trained based on sample images and sample label data.

[0012] Further, according to the centerline detection method provided by the present invention, the method further includes:

[0013] When the first confidence value is less than or equal to the first preset threshold, a discard operation is performed on the first candidate center point corresponding to the first confidence value.

[0014] Further, according to the centerline detection method provided by the present invention, the center point prediction model includes a generation layer and a prediction layer, wherein,

[0015] The generation layer is used to perform Gaussian distribution generation processing on the to-be-processed image to obtain the Gaussian distribution of the to-be-processed image, and confirm a plurality of key points from the Gaussian distribution;

[0016] The prediction layer performs prediction processing on the plurality of key points to obtain a plurality of candidate center points.

[0017] Further, according to the centerline detection method provided by the present invention, the step of inputting the to-be-processed image into the generation layer to perform Gaussian distribution generation processing to obtain the Gaussian distribution of the to-be-processed image includes:

[0018] Determine a first center point according to the to-be-processed image;

[0019] Generate a one-dimensional Gaussian distribution according to the position information of the first center point and the width value of the lane line to obtain the Gaussian distribution of the to-be-processed image.

[0020] Further, according to the centerline detection method provided by the present invention, before obtaining the to-be-processed image, the method further includes:

[0021] Step S1: Use the to-be-trained center point prediction model to perform Gaussian distribution generation processing on the sample center points included in the sample image to obtain a plurality of key points and the Gaussian distribution of the sample image;

[0022] Step S2: Calculate the loss value according to the plurality of key points and the sample center points, and obtain the prediction result of the sample image according to the calculation result;

[0023] Step S3: According to the prediction result and the sample label data, determine whether the model training termination condition is satisfied. When the model training termination condition is not satisfied, adjust the to-be-trained center point prediction model, and use the adjusted center point prediction model to re-execute Step S1; when the model training termination condition is satisfied, obtain the trained center point prediction model.

[0024] Further, according to the centerline detection method provided by the present invention, calculating a loss value based on the multiple key points and the sample center point, and obtaining a prediction result of the sample image according to the calculation result, including:

[0025] Calculating loss values for the multiple key points based on the FL algorithm to obtain multiple loss values;

[0026] Performing backpropagation training according to the multiple loss values to obtain a prediction result of the sample image.

[0027] In a second aspect, the present invention further provides a centerline detection device, including:

[0028] An acquisition module, configured to acquire an image to be processed; wherein, the image to be processed is an image acquired in real time during driving;

[0029] An input module, configured to input the image to be processed into a pre-trained center point prediction model to obtain multiple candidate center points;

[0030] A calculation module, configured to calculate the confidence of each candidate center point among the multiple candidate center points to obtain multiple confidence values;

[0031] A regression module, configured to perform a regression operation on a first candidate center point corresponding to the first confidence value when the first confidence value is greater than a first preset threshold, and determine that the first candidate center point is located on the center line of the lane line;

[0032] Wherein, the first confidence value is any one of the multiple confidence values, and the first candidate center point is any one of the multiple candidate center points;

[0033] Wherein, the center point prediction model is trained based on sample images and sample label data.

[0034] In a third aspect, the present invention further provides an electronic device, including:

[0035] A processor, a memory, and a bus, wherein,

[0036] The processor and the memory communicate with each other through the bus;

[0037] The memory stores program instructions executable by the processor, and the processor can execute the steps of the centerline detection method described in any one of the above by calling the program instructions.

[0038] In a fourth aspect, the present invention further provides a non-transitory computer-readable storage medium, and the non-transitory computer-readable storage medium stores computer instructions, and the computer instructions cause the computer to execute the steps of the centerline detection method described above.

[0039] In a fifth aspect, the present invention further provides a computer program product, including a computer program, characterized in that when the computer program is executed by a processor, the steps of the center line detection method described in any one of the above are implemented.

[0040] The present invention provides a center line detection method, device, electronic device, storage medium and product. The method includes: obtaining an image to be processed, where the image to be processed is an image collected in real time during driving; inputting the image to be processed into a pre-trained center point prediction model to obtain a plurality of candidate center points; calculating a confidence value for each of the plurality of candidate center points to obtain a plurality of confidence values; when a first confidence value is greater than a first preset threshold, performing a regression operation on the first candidate center point corresponding to the first confidence value to determine that the first candidate center point is located on the center line of the lane line. The present invention realizes the prediction of key points other than the center points on the center line through a preset center point prediction model, obtains a plurality of candidate center points, and performs a regression operation on the candidate center points to ensure the accurate detection of the center line when the offset is too large, ensure the safety of the autonomous driving vehicle during driving, and improve the passenger experience. Description of the Drawings

[0041] In order to more clearly illustrate the technical solutions in the present invention or the prior art, the following will briefly introduce the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings in the following description are some embodiments of the present invention. For those of ordinary skill in the art, other drawings can be obtained based on these drawings without creative efforts.

[0042] Figure 1 It is a flowchart of a center line detection method provided by the present invention;

[0043] Figure 2 It is a structural diagram of a center line detection device provided by the present invention;

[0044] Figure 3 It is a structural diagram of an electronic device provided by the present invention. Detailed Embodiments

[0045] To make the objectives, technical solutions and advantages of the present invention clearer, the following will clearly and completely describe the technical solutions in the present invention with reference to the drawings in the present invention. Obviously, the described embodiments are some, but not all, of the embodiments of the present invention. Based on the embodiments in the present invention, all other embodiments obtained by those of ordinary skill in the art without creative efforts fall within the protection scope of the present invention.

[0046] Figure 1The flowchart of the centerline detection method provided by the present invention is shown as follows. Figure 1 As shown, the centerline detection method provided by the present invention includes the following steps:

[0047] Step 101: Obtain the image to be processed; wherein, the image to be processed is an image collected in real time during driving.

[0048] Step 102: Input the image to be processed into a pre-trained center point prediction model to obtain multiple candidate center points.

[0049] Step 103: Calculate the confidence value of each candidate center point among the multiple candidate center points to obtain multiple confidence values.

[0050] Step 104: When the first confidence value is greater than the first preset threshold, perform a regression operation on the first candidate center point corresponding to the first confidence value to determine that the first candidate center point is located on the centerline of the lane line.

[0051] Wherein, the first confidence value is any one of the multiple confidence values, and the first candidate center point is any one of the multiple candidate center points.

[0052] Wherein, the center point prediction model is trained based on sample images and sample label data.

[0053] Specifically, the confidence interval is an interval estimate of a certain population parameter of a sample, which shows the degree to which the true value of this parameter has a certain probability of falling around the measurement result. The confidence interval gives the range of the credibility of the measured value of the measured parameter, that is, the "certain probability" required above. This probability is called the confidence level.

[0054] In step 101, the image to be processed obtained is an image collected in real time during vehicle driving and is the image to be processed. It should be noted that in this embodiment, the starting point in the centerline of the image to be processed is used as a center point for subsequent processing.

[0055] In step 102, the image to be processed is input into the center point prediction model to obtain multiple candidate center points. Among them, the multiple candidate center points can be center points predicted other than the center point on the centerline. It should be noted that the center point prediction model is pre-trained based on sample images and sample label data.

[0056] In step 103, calculate the confidence of each candidate center point among the multiple candidate center points obtained in step 102 to obtain multiple confidence values.

[0057] In step 104, when the obtained first confidence value is greater than the first preset threshold, a regression operation is performed on the first candidate center point corresponding to the first confidence value to determine that the first candidate center point is located on the center line of the lane line. Suppose the size of the first preset threshold is 0.3. From the feature map of the heatmap of the predicted center point, the first confidence value is calculated to be 0.5, which is greater than the preset threshold 0.3. Then a regression operation is performed on this candidate center point. The offset function is used to predict the size of the compensation for the regression operation. For example, the position of the center point on the center line is (x0, y0), and the position information of the first candidate center point predicted by the center point prediction model is (x1, y0). Then, based on the information of the position of the candidate center point (x1, y0), the offset function is used to obtain the offset (δx, δy) in the offset feature map. After that, this offset is added to x1, and finally the position of the predicted center point passing through the center line is obtained as (x1' = x1 + δx, y1' = y0). For example, the position of the center point on the center is (2, 8), and a point with the position (7, 8) is predicted as a candidate center point. The corresponding offset values δx and δy can be obtained at the position (7, 8) through this offset feature map. Here, they are (-5, 0) respectively. Then, at this time, the coordinates of the predicted point on the center line are (7 - 5 = 2, 8 + 0 = 8), so that the candidate center point is on the center line.

[0058] In the embodiment of the present invention, by inputting the images collected in real time during the vehicle driving process into a pre-trained center point prediction model, multiple candidate center points are obtained, and the confidence values of the multiple candidate center points are calculated. When the obtained first confidence value is greater than the first preset threshold, a regression operation is performed on the first candidate center point corresponding to the first confidence value to determine that the first candidate center point is located on the center line of the lane line. The present invention realizes the prediction of candidate center points other than the center point on the center line through the preset center point prediction model, and performs a regression operation on the candidate center points to ensure the accurate detection of the center line when the offset is too large, ensure the safety of the autonomous driving vehicle during driving, and improve the passenger experience.

[0059] In an embodiment of the present invention, the method further includes:

[0060] When the first confidence value is less than or equal to the first preset threshold, a discard operation is performed on the first candidate center point corresponding to the first confidence value.

[0061] Specifically, the discard operation means directly abandoning the candidate center point that cannot be regressed without performing a regression operation.

[0062] In the embodiment of the present invention, when the calculated first confidence value is less than or equal to the first preset threshold, it indicates that the candidate center point is not within the range of regression operation processing, and it is directly discarded and cannot be regressed to the center line to become the center point on the center line. For example, if the size of the first preset threshold is 0.3, from the feature map of the predicted heatmap, a first confidence value less than the first preset threshold is obtained. Then, when the predicted position of point A at (7, 8) is a candidate center point, since the first confidence value is less than the first preset threshold, the obtained candidate center point A is directly discarded. It should be noted that the size of the first preset threshold can be set according to actual needs and is not specifically limited herein.

[0063] In the embodiment of the present invention, by discarding the candidate center points greater than the first preset threshold, the accuracy of center line detection is ensured.

[0064] In another embodiment of the present invention, the center point prediction model includes a generation layer and a prediction layer, where

[0065] the generation layer is used to perform Gaussian distribution generation processing on the image to be processed, obtain the Gaussian distribution of the image to be processed, and confirm multiple key points from the Gaussian distribution;

[0066] the prediction layer performs prediction processing on the multiple key points to obtain multiple candidate center points.

[0067] Specifically, the Gaussian distribution is also called the normal distribution, which is a very important probability distribution in the fields of mathematics, physics, engineering, etc.

[0068] In the embodiment of the present invention, the center line part included in the real-time acquired image information is divided into several parts, and then the starting point of each part is used as a center point, and a one-dimensional Gaussian distribution is generated according to the width of the lane line. For example, key point A is obtained, and according to the width of the lane line at its corresponding y value, the left edge x1 and the right edge x2 are determined. Set the length L = (0.5 * (x2 - x1 + 1)), and then generate a list containing values such as x = [-L, -L + 1, …, 0, L - 1, L]. Then, take the In this way, a one-dimensional Gaussian distribution is generated, and multiple key points are obtained from the candidate region formed by x1 and x2.

[0069] In the embodiment of the present invention, then, according to the feature map predicted from the initially obtained key points, the feature map of the difference in the y direction from each key point to the final point, the probability map of each key point, and the feature map of the parameters of the predicted quadratic equation corresponding to each key point are obtained, and multiple candidate center points are obtained from the feature map.

[0070] In the embodiment of the present invention, by inputting the image to be processed into the center point prediction model, generating a Gaussian distribution and predicting the center point, a plurality of candidate center points are obtained. The provided model prediction method is simple and has relatively high detection accuracy.

[0071] In another embodiment of the present invention, the step of inputting the image to be processed into the generation layer to generate a Gaussian distribution of the image to be processed includes:

[0072] Determining a first center point according to the image to be processed;

[0073] Generating a one-dimensional Gaussian distribution according to the position information of the first center point and the width value of the lane line, to obtain the Gaussian distribution of the image to be processed.

[0074] Specifically, the lane line is a guiding lane line, which is a lane marking for guiding the direction, used to indicate that the vehicle should drive in the indicated direction when entering the intersection section. Such markings are generally drawn at traffic intersections with heavy traffic, aiming to clarify the driving direction, keep each lane in order, and relieve traffic pressure.

[0075] In the embodiment of the present invention, the first center point A is confirmed from the image to be processed, and then a one-dimensional Gaussian distribution is generated according to the coordinate position of the center point A and the width value of the lane line. The specific operation method is as described in the above embodiment and will not be elaborated here. The method provided in the embodiment of the present invention can obtain the Gaussian distribution of the image to be processed, which is convenient for obtaining a plurality of candidate key points and provides support for subsequent processing.

[0076] In another embodiment of the present invention, before obtaining the image to be processed, the method further includes:

[0077] Step S1: Using the center point prediction model to be trained to generate a Gaussian distribution for the sample center points included in the sample image, to obtain a plurality of key points and the Gaussian distribution of the sample image;

[0078] Step S2: Calculating a loss value according to the plurality of key points and the sample center points, and obtaining a prediction result of the sample image according to the calculation result;

[0079] Step S3: Judging whether the model training termination condition is satisfied according to the prediction result and the sample label data. When the model training termination condition is not satisfied, adjusting the center point prediction model to be trained, and using the adjusted center point prediction model to re-execute Step S1; when the model training termination condition is satisfied, obtaining the trained center point prediction model.

[0080] Specifically, the model training termination condition means that when the accuracy rate of model training reaches the required threshold, the training stops.

[0081] In the embodiment of the present invention, a Gaussian distribution generation process is performed on the sample center points in the sample image by using the center point prediction model to be trained, obtaining a plurality of key points. A Gaussian distribution of the sample image is generated according to the plurality of key points, and then a loss value is calculated for the key points and the sample center points to obtain a prediction result of the sample image, that is, a plurality of candidate center points are obtained. Then, the obtained plurality of candidate center points are compared with the sample label data to determine whether the accuracy rate of the model prediction meets the set requirements. For example, if the set condition for the model to stop training is that the accuracy rate is 90%, when the accuracy rate of the model during training is greater than 90%, the training stops. If this requirement is not met, the parameters of the model need to be adjusted, and the model with the adjusted parameters is used to repeat the training until the requirement is met and then the training stops, obtaining the trained center point prediction model.

[0082] In the embodiment of the present invention, a Gaussian distribution generated by using the initial center point in the sample image is used, and a plurality of key points are obtained from the Gaussian distribution. Among them, all key points with probability values greater than 0 form positive samples in the training sample set, and a loss value calculation process is performed on the obtained plurality of key points. Finally, the final model is obtained by comparing the prediction result with the sample label data. The training method provided by the present invention improves the accuracy rate of the model.

[0083] In another embodiment of the present invention, calculating the loss value according to the plurality of key points and the sample center point, and obtaining the prediction result of the sample image according to the calculation result includes:

[0084] Calculating the loss value for the plurality of key points based on the FL algorithm to obtain a plurality of loss values;

[0085] Performing backpropagation training according to the plurality of loss values to obtain the prediction result of the sample image.

[0086] Specifically, the FL (Focal Loss) algorithm is mainly to solve the problem of serious imbalance in the proportion of positive and negative samples in one-stage object detection, and this loss function reduces the weight of a large number of simple negative samples in training.

[0087] In the embodiment of the present invention, the FL algorithm is used to calculate the loss value for a plurality of key points obtained from the Gaussian distribution based on the center point in the sample image, and backpropagation training is performed according to the obtained plurality of loss values to obtain a plurality of candidate center points. It should be noted that using F loss =-(1 - p) γ*Calculate the loss value using log(p), and predict multiple candidate center points from multiple key points according to the size of the loss value.

[0088] The method provided by the present invention can expand the number of center points in the sample image, ensure the accuracy of model training, and ensure the accuracy of center line detection when the offset is too large.

[0089] Figure 2 A center line detection device provided by the present invention, as Figure 2 shown, the center line detection device provided by the present invention includes:

[0090] An acquisition module 201 for acquiring an image to be processed; wherein, the image to be processed is an image collected in real time during driving;

[0091] An input module 202 for inputting the image to be processed into a pre-trained center point prediction model to obtain multiple candidate center points;

[0092] A calculation module 203 for calculating the confidence value of each candidate center point among the multiple candidate center points to obtain multiple confidence values;

[0093] A regression module 204 for performing a regression operation on a first candidate center point corresponding to the first confidence value when the first confidence value is greater than a first preset threshold, and determining that the first candidate center point is located on the center line of the lane line;

[0094] Wherein, the first confidence value is any one of the multiple confidence values, and the first candidate center point is any one of the multiple candidate center points;

[0095] Wherein, the center point prediction model is trained based on sample images and sample label data.

[0096] The center line detection device provided by the present invention realizes the prediction of candidate center points other than the center points on the center line through a preset center point prediction model, and performs a regression operation on the candidate center points, ensuring the accuracy of center line detection when the offset is too large, ensuring the safety of the autonomous driving vehicle during driving, and improving the passenger experience.

[0097] Further, the center line detection device further includes a discard module, and the discard module is used for:

[0098] When the first confidence value is less than or equal to the first preset threshold, performing a discard operation on the first candidate center point corresponding to the first confidence value.

[0099] Further, the input module further includes a generation unit and a prediction unit, wherein,

[0100] The generating unit is used to perform generating processing on the image to be processed according to a Gaussian distribution, obtain the Gaussian distribution of the image to be processed, and confirm a plurality of key points from the Gaussian distribution;

[0101] The predicting unit is used to perform prediction processing on the plurality of key points to obtain a plurality of candidate center points.

[0102] Further, the generating unit is further used for:

[0103] Determine a first center point according to the image to be processed;

[0104] Generate a one-dimensional Gaussian distribution according to the position information of the first center point and the width value of the lane line to obtain the Gaussian distribution of the image to be processed.

[0105] Further, the device further includes a training module, and the training module includes a generating unit, a predicting unit, and a training termination unit:

[0106] The generating unit uses a center point prediction model to be trained to perform Gaussian distribution generating processing on the sample center points included in the sample image, and obtain a plurality of key points and the Gaussian distribution of the sample image;

[0107] The predicting unit calculates a loss value according to the plurality of key points and the sample center points, and obtains a prediction result of the sample image according to the calculation result;

[0108] The training termination unit determines whether the model training termination condition is satisfied according to the prediction result and the sample label data. When the model training termination condition is not satisfied, adjust the center point prediction model to be trained, and use the adjusted center point prediction model to operate again; when the model training termination condition is satisfied, obtain the trained center point prediction model.

[0109] Further, the training module is further used for:

[0110] Calculate loss values for the plurality of key points based on the FL algorithm to obtain a plurality of loss values;

[0111] Perform backpropagation training according to the plurality of loss values to obtain a prediction result of the sample image.

[0112] Since the device described in the embodiments of the present invention has the same principle as the method described in the above embodiments, more detailed explanation content will not be elaborated here.

[0113] Figure 3 It is a schematic diagram of the physical structure of the electronic device provided by the embodiments of the present invention, as Figure 3As shown in the figure, the present invention provides an electronic device, including: a processor 301, a memory 302, and a bus 303;

[0114] Among them, the processor 301 and the memory 302 complete mutual communication through the bus 303;

[0115] The processor 301 is used to call program instructions in the memory 302 to execute the methods provided in the above method embodiments, for example, including: obtaining an image to be processed; wherein, the image to be processed is an image collected in real time during driving; inputting the image to be processed into a pre-trained center point prediction model to obtain a plurality of candidate center points; calculating a confidence value for each candidate center point among the plurality of candidate center points to obtain a plurality of confidence values; when a first confidence value is greater than a first preset threshold, performing a regression operation on the first candidate center point corresponding to the first confidence value to determine that the first candidate center point is located on the center line of the lane line; wherein, the first confidence value is any one of the plurality of confidence values, and the first candidate center point is any one of the plurality of candidate center points; wherein, the center point prediction model is trained based on sample images and sample label data.

[0116] An embodiment of the present invention provides a non-transitory computer-readable storage medium. The non-transitory computer-readable storage medium stores computer instructions, and the computer instructions cause the computer to execute the methods provided in the above method embodiments, for example, including: obtaining an image to be processed; wherein, the image to be processed is an image collected in real time during driving; inputting the image to be processed into a pre-trained center point prediction model to obtain a plurality of candidate center points; calculating a confidence value for each candidate center point among the plurality of candidate center points to obtain a plurality of confidence values; when a first confidence value is greater than a first preset threshold, performing a regression operation on the first candidate center point corresponding to the first confidence value to determine that the first candidate center point is located on the center line of the lane line; wherein, the first confidence value is any one of the plurality of confidence values, and the first candidate center point is any one of the plurality of candidate center points; wherein, the center point prediction model is trained based on sample images and sample label data.

[0117] The present invention also provides a computer program product, which includes a computer program stored on a non-transitory computer-readable storage medium. The computer program includes program instructions. When the program instructions are executed by a computer, the computer can execute the methods provided in the above method embodiments. The method includes: obtaining an image to be processed; wherein, the image to be processed is an image collected in real time during driving; inputting the image to be processed into a pre-trained center point prediction model to obtain a plurality of candidate center points; calculating a confidence value for each of the plurality of candidate center points to obtain a plurality of confidence values; when a first confidence value is greater than a first preset threshold, performing a regression operation on a first candidate center point corresponding to the first confidence value to determine that the first candidate center point is located on the center line of the lane line; wherein, the first confidence value is any one of the plurality of confidence values, and the first candidate center point is any one of the plurality of candidate center points; wherein, the center point prediction model is trained based on sample images and sample label data.

[0118] Those of ordinary skill in the art can understand that all or part of the steps to implement the above method embodiments can be completed by hardware related to program instructions. The foregoing program can be stored in a computer-readable storage medium. When the program is executed, it executes the steps including the above method embodiments; and the foregoing storage medium includes: various media such as ROM, RAM, magnetic disk, or optical disk that can store program codes.

[0119] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit them. Although the present invention has been described in detail with reference to the foregoing embodiments, those of ordinary skill in the art should understand that they can still modify the technical solutions described in the foregoing embodiments or perform equivalent replacements for some of the technical features. However, these modifications or replacements do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the present invention in each embodiment.

Claims

1. A center line detection method, characterized in that, Including: Obtain an image to be processed; wherein, the image to be processed is an image collected in real time during driving; Input the image to be processed into a pre-trained center point prediction model to obtain multiple candidate center points; Calculate the confidence value of each candidate center point among the multiple candidate center points to obtain multiple confidence values; When the first confidence value is greater than the first preset threshold, perform a regression operation on the first candidate center point corresponding to the first confidence value to determine that the first candidate center point is located on the center line of the lane line; Wherein, the first confidence value is any one of the multiple confidence values, and the first candidate center point is any one of the multiple candidate center points; Wherein, the center point prediction model is trained based on sample images and sample label data; Wherein, performing the regression operation on the first candidate center point corresponding to the first confidence value to determine that the first candidate center point is located on the center line of the lane line specifically includes: Obtain the position of the center point on the center line as (x0, y0), and the first candidate center point corresponding to the first confidence value is (x1, y0); use the offset function to obtain the offset in the feature map of offset as ( ) based on the information of the position of the candidate center point (x1, y0). After that, add the offset to x1 to obtain the position of the first candidate center point on the center line of the lane line as ( ).

2. The centerline detection method according to claim 1, characterized in that, The method further includes: When the first confidence value is less than or equal to the first preset threshold, perform a discard operation on the first candidate center point corresponding to the first confidence value.

3. The centerline detection method according to claim 1 or 2, characterized in that, The center point prediction model includes a generation layer and a prediction layer, wherein, The generation layer is used to perform Gaussian distribution generation processing on the image to be processed to obtain the Gaussian distribution of the image to be processed, and confirm multiple key points from the Gaussian distribution; The prediction layer performs prediction processing on the multiple key points to obtain multiple candidate center points.

4. The centerline detection method according to claim 3, wherein Inputting the image to be processed into the generation layer for Gaussian distribution generation processing to obtain the Gaussian distribution of the image to be processed includes: Determine a first center point according to the image to be processed; Generate a one-dimensional Gaussian distribution according to the position information of the first center point and the width value of the lane line to obtain the Gaussian distribution of the image to be processed.

5. The centerline detection method according to claim 1 or 2, characterized in that Before obtaining the image to be processed, the method further includes: Step S1: Use the center point prediction model to be trained to perform Gaussian distribution generation processing on the sample center points included in the sample image to obtain multiple key points and the Gaussian distribution of the sample image; Step S2: Calculate the loss value according to the multiple key points and the sample center points, and obtain the prediction result of the sample image according to the calculation result; Step S3: According to the prediction result and the sample label data, determine whether the model training termination condition is satisfied. When the model training termination condition is not satisfied, adjust the center point prediction model to be trained, and use the adjusted center point prediction model to re-execute Step S1; when the model training termination condition is satisfied, obtain the trained center point prediction model.

6. The centerline detection method according to claim 5, characterized in that, The calculating the loss value according to the multiple key points and the sample center points, and obtaining the prediction result of the sample image according to the calculation result includes: Calculate the loss value of the multiple key points based on the FL algorithm to obtain multiple loss values; Perform backpropagation training according to the multiple loss values to obtain the prediction result of the sample image; Wherein, the function expression for calculating the loss value of the multiple key points based on the FL algorithm is: ； p is the probability of predicting a sample as a positive sample, γ is a tuning parameter in the focal loss function, used to adjust the weights of easy and hard samples.

7. A center line detection device, characterized in that, Including: An acquisition module for acquiring an image to be processed, where the image to be processed is an image collected in real time during driving; An input module for inputting the image to be processed into a pre-trained center point prediction model to obtain a plurality of candidate center points; A calculation module for calculating the confidence of each candidate center point among the plurality of candidate center points to obtain a plurality of confidence values; A regression module for performing a regression operation on a first candidate center point corresponding to the first confidence value when the first confidence value is greater than a first preset threshold to determine that the first candidate center point is located on the center line of the lane line; wherein the first confidence value is any one of the plurality of confidence values, and the first candidate center point is any one of the plurality of candidate center points; wherein the center point prediction model is trained based on sample images and sample label data; wherein the performing a regression operation on the first candidate center point corresponding to the first confidence value to determine that the first candidate center point is located on the center line of the lane line specifically includes: Obtain the position of the center point on the center line as (x0, y0), and the first candidate center point corresponding to the first confidence value is (x1, y0); use the offset function to obtain the offset in the feature map of offset as ( ) based on the information of the position of the candidate center point (x1, y0). After that, add the offset to x1 to obtain the position of the first candidate center point on the center line of the lane line as ( ).

8. An electronic device, characterized in that, including: a processor, a memory, and a bus, wherein, the processor and the memory complete communication with each other through the bus; the memory stores program instructions executable by the processor, and the processor can execute the steps of the center line detection method according to any one of claims 1 to 6 by calling the program instructions.

9. A non-transitory computer-readable storage medium, characterized in that, The non-transitory computer-readable storage medium stores computer instructions, and the computer instructions cause the computer to execute the steps of the center line detection method according to any one of claims 1 to 6.

10. A computer program product, comprising a computer program, characterized in that, The computer program, when executed by the processor, implements the steps of the center line detection method according to any one of claims 1 to 6.

Images