Lane line prediction method, device, equipment and product

Abstract

The invention discloses a lane line prediction method, device, equipment and product, and relates to the technical field of lane line detection, and the method comprises the steps: obtaining road real-time data under a low-illumination road condition, and generating a to-be-detected image based on the road real-time data; the to-be-detected image is input to a target detection network model, a target lane line detection result is obtained, and the target detection network model is used for enhancing image performance and optimizing lane line detection precision; and predicting the lane line based on the target lane line detection result. The real-time data can reflect the change of the road in time, so that a more accurate to-be-detected image is generated, the optimized network model can accurately identify the lane line, more reliable data is provided for lane line prediction, and therefore, a target lane line detection result obtained through the target optimization network model is more accurate in detection. According to the invention, the lane line detection precision under the low illumination condition can be ensured, and further, the accuracy of the lane line prediction result is improved on the basis of ensuring the lane line detection precision.

Description

Technical Field

[0001] The present application relates to the field of lane line detection technology, and in particular to lane line prediction methods, devices, equipment and products. Background Art

[0002] With the continuous advancement of digital and intelligent technologies, autonomous driving has become a focus of current attention. Active driving mainly relies on on-board sensors (such as cameras, ultrasonic radars, lidars, etc.) to obtain road condition information from the surrounding environment, thereby achieving tasks such as positioning, obstacle avoidance and path planning. Lane line detection is an important subtask in autonomous driving environment perception. Accurate and timely lane line detection can provide important information support for the regulation and control of autonomous driving.

[0003] At present, with the continuous advancement of deep learning technology, lane detection methods based on this technology have gradually replaced traditional machine learning technology with their excellent detection accuracy and many advantages, and have become the mainstream solution in the field of lane detection. The current data sets of lane detection algorithms mainly focus on driving scenes with good lighting, while the data collected in low-light environments such as night or tunnels is relatively small, which leads to the algorithm facing the challenges of complex feature extraction and reduced detection accuracy under low-light conditions.

[0004] Therefore, how to improve the accuracy of lane line prediction in low-light scenarios is a problem that needs to be solved urgently. Summary of the invention

[0005] The main purpose of this application is to provide a lane line prediction method, device, equipment and product, aiming to solve the technical problem of low accuracy of lane line prediction in low illumination scenes.

[0006] To achieve the above purpose, the present application proposes a lane line prediction method, the method comprising:

[0007] Acquire real-time road data under low-light road conditions, and generate an image to be tested based on the real-time road data;

[0008] Inputting the image to be tested into a target detection network model to obtain a target lane line detection result, wherein the target detection network model is used to enhance image performance and optimize lane line detection accuracy;

[0009] Based on the target lane line detection result, the lane line is predicted.

[0010] In one embodiment, the target detection network model includes a target image enhancement network and a target lane detection network. Before the step of inputting the image data to be tested into the target detection network model to obtain the target lane detection result, the step includes:

[0011] Obtain low-light image datasets and road real-time image datasets;

[0012] Inputting the low-light image dataset into a target image enhancement network to obtain a target enhanced image dataset;

[0013] The target enhanced image data set is input into the target lane detection network to obtain the target lane line detection result.

[0014] In one embodiment, the target image enhancement network includes a first backbone network, a channel attention network, and a first self-supervised loss network, and the step of inputting the low-illumination image dataset into the target image enhancement network to obtain the target enhanced image dataset includes:

[0015] Inputting the low-light image data set into the first backbone network, extracting initial image features through depthwise separable convolution, wherein the depthwise separable convolution is used to reduce the amount of calculation of the model;

[0016] Performing global maximum pooling and global average pooling on the initial image features through the channel attention network to obtain intermediate image features;

[0017] The intermediate image features are input into the shared connection layer, and the degree of attention paid by the model to each channel feature is dynamically adjusted through the S-type function to obtain the target image features;

[0018] Optimizing the image enhancement network through the first self-supervised loss network to obtain a target image enhancement network;

[0019] The target image features are input into a target image enhancement network to obtain a target enhanced image dataset.

[0020] In one embodiment, the target lane detection network includes a second backbone network and a second self-supervised loss network, and the step of inputting the target enhanced image data set into the target lane detection network to obtain a target lane line detection result includes:

[0021] Inputting the target enhanced image data set into the second backbone network, and obtaining an initial lane line detection result by introducing shallow features;

[0022] Optimizing the lane detection network through the second self-supervised loss network to obtain a target lane detection network;

[0023] The initial lane line detection result is input into the target lane line detection network to obtain the target lane line detection result.

[0024] In one embodiment, the first self-supervised function network includes an exposure control loss, and the step of optimizing the image enhancement network through the first self-supervised loss network to obtain a target image enhancement network includes:

[0025] Reducing the exposure control loss by a sliding window method to obtain an improved exposure control loss;

[0026] Based on the improved exposure control loss, the image enhancement network is optimized to obtain a target image enhancement network.

[0027] In one embodiment, the second self-supervised function network includes a matching loss, and the step of optimizing the lane detection network by the second self-supervised loss network to obtain the target lane detection network includes:

[0028] The matching loss is reduced by introducing a lane line variable in a vertical direction, thereby obtaining an improved matching loss;

[0029] The lane detection network is optimized based on the improved matching loss to obtain a target lane detection network.

[0030] In addition, to achieve the above purpose, the present application also proposes a lane line prediction device, the lane line prediction device comprising:

[0031] A data acquisition module, used for collecting real-time road data and generating an image to be tested based on the real-time road data;

[0032] A result acquisition module, used to input the image to be tested into a target detection network model to obtain a target lane line detection result, wherein the target detection network model is used to enhance image performance and optimize lane line detection accuracy;

[0033] The lane prediction module is used to predict the lane line based on the target lane line detection result.

[0034] In addition, to achieve the above-mentioned purpose, the present application also proposes a lane line prediction device, which includes: a memory, a processor, and a computer program stored in the memory and executable on the processor, and the computer program is configured to implement the steps of the lane line prediction method described above.

[0035] In addition, to achieve the above-mentioned purpose, the present application also proposes a storage medium, which is a computer-readable storage medium, and a computer program is stored on the storage medium. When the computer program is executed by a processor, the steps of the lane line prediction method described above are implemented.

[0036] In addition, to achieve the above-mentioned purpose, the present application also provides a computer program product, which includes a computer program, and when the computer program is executed by a processor, it implements the steps of the lane line prediction method as described above.

[0037] One or more technical solutions proposed in this application have at least the following technical effects:

[0038] Real-time road data under low-light conditions is obtained, and images to be tested are generated based on the real-time road data. The images to be tested are input into the target detection network model to obtain the target lane line detection results. The target detection network model is used to enhance image performance and optimize the detection accuracy of lane lines. Based on the target lane line detection results, the lane lines are predicted. Real-time data can reflect the changes in the road in a timely manner, thereby generating more accurate images to be tested. The optimized network model can accurately identify lane lines and provide more reliable data for lane line prediction. Therefore, the target lane line detection results obtained by the target optimization network model can ensure the lane line detection accuracy under low-light conditions. Furthermore, on the basis of ensuring the lane line detection accuracy, the accuracy of the lane line prediction results is improved. BRIEF DESCRIPTION OF THE DRAWINGS

[0039] The accompanying drawings, which are incorporated in and constitute a part of this specification, illustrate embodiments consistent with the present application and, together with the description, serve to explain the principles of the present application.

[0040] In order to more clearly illustrate the embodiments of the present application or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, for ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative labor.

[0041] Figure 1 This is a flow chart of the first embodiment of the lane line prediction method of the present application;

[0042] Figure 2 This is a flow chart of the second embodiment of the lane line prediction method of the present application;

[0043] Figure 3 This is a schematic diagram of the structure of the Zero-DCE network in the embodiment of the present application;

[0044] Figure 4 This is a schematic diagram of the structure of the CLRNet network in the embodiment of the present application;

[0045] Figure 5 A schematic diagram of the CLRNet network structure that introduces a shallow feature layer in an embodiment of the present application;

[0046] Figure 6This is a schematic diagram of the module structure of the lane line prediction device according to an embodiment of the present application;

[0047] Figure 7 Schematic diagram of the device structure of the hardware operating environment involved in the lane line prediction method in the embodiment of the present application.

[0048] The purpose, features and advantages of this application will be further described in conjunction with the embodiments and with reference to the accompanying drawings. DETAILED DESCRIPTION

[0049] It should be understood that the specific embodiments described herein are only used to explain the technical solutions of the present application and are not used to limit the present application.

[0050] In order to better understand the technical solution of the present application, a detailed description will be given below in conjunction with the accompanying drawings and specific implementation methods.

[0051] Since the current lane detection algorithm's data set mainly focuses on driving scenes with good lighting, and the data collected in low-light environments such as night or tunnels is relatively small, the algorithm faces the challenges of complex feature extraction and reduced detection accuracy under low-light conditions. In addition, given the limited resources of on-board computing equipment, the design of the detection algorithm for relatively basic objects such as lane lines with simple features needs to be efficient and save computing resources.

[0052] The present application provides a solution to obtain real-time road data under low-light conditions, generate a test image based on the real-time road data, input the test image into a target detection network model, and obtain a target lane line detection result. The target detection network model is used to enhance image performance and optimize lane line detection accuracy. Based on the target lane line detection result, the lane line is predicted. Real-time data can reflect road changes in a timely manner, thereby generating more accurate images to be tested. While reducing the amount of network parameters and calculations, the original image enhancement performance of the model is maintained. Therefore, the target lane line detection result obtained by the target optimization network model can ensure the lane line detection accuracy under low-light conditions, while ensuring that the amount of parameters and calculations are reduced, so that the reasoning speed is improved.

[0053] Based on this, the present application embodiment provides a lane line prediction method, referring to Figure 1 , Figure 1 This is a flow chart of the first embodiment of the lane line prediction method of the present application.

[0054] In this embodiment, the lane line prediction method includes steps S10 to S30:

[0055] Step S10, acquiring real-time road data under low-light road conditions, and generating an image to be tested based on the real-time road data.

[0056] It should be noted that low-light road conditions refer to road conditions observed under poor lighting conditions (e.g., at night or in bad weather). Real-time road data can be data related to road conditions collected in real time, which helps to understand the current traffic conditions. The image to be tested is an image to be processed generated under specific conditions (e.g., low light) and is used for lane line detection.

[0057] Step S20, input the image to be tested into the target detection network model to obtain the target lane line detection result. The target detection network model is used to enhance image performance and optimize the lane line detection accuracy.

[0058] It should be noted that the target detection network model is a neural network model based on deep learning, which is used to identify and locate specific targets (such as lane lines, vehicles, etc.) in images. Exemplary target detection algorithms may include YOLO, SSD, Faster R-CNN, etc. It is understandable that the target lane line detection result can be obtained by processing the target detection network model, and may include information such as the position and shape of the lane line.

[0059] Step S30: predicting the lane line based on the target lane line detection result.

[0060] It should be noted that lane line prediction can be based on the target lane line detection results and use algorithms to infer the future state or shape of the lane line.

[0061] In this embodiment, real-time road data under low-light conditions is obtained, and an image to be tested is generated based on the real-time road data. The image to be tested is input into the target detection network model to obtain the target lane line detection result. The target detection network model is used to enhance image performance and optimize the detection accuracy of the lane line. Based on the target lane line detection result, the lane line is predicted. Real-time data can reflect the changes in the road in a timely manner, thereby generating a more accurate image to be tested. The optimized network model can accurately identify the lane line and provide more reliable data for lane line prediction. Therefore, the target lane line detection result obtained by the target optimization network model can ensure the lane line detection accuracy under low-light conditions. Furthermore, on the basis of ensuring the lane line detection accuracy, the accuracy of the lane line prediction result is improved.

[0062] Reference Figure 2 , Figure 2 This is a flow chart of the second embodiment of the lane line prediction method of the present application. Figure 1 The first embodiment shown proposes a second embodiment of the lane line prediction method of the present application.

[0063] In the second embodiment, the step S20 includes:

[0064] Step S201, obtaining a low-illumination image dataset and a road real-time image dataset.

[0065] For example, the real-time road image dataset can be directly collected by a specially configured data collection vehicle at various times in the real traffic environment (including periods of time when lighting conditions vary significantly). At the same time, open source low-light image datasets can be introduced through open source datasets such as CULane and TuSimple.

[0066] Step S202: input the low-light image data set into the target image enhancement network to obtain a target enhanced image data set.

[0067] Exemplarily, the target image enhancement network can generate clearer and easier-to-analyze images by means of denoising, brightness adjustment, contrast enhancement, etc. The target enhanced image dataset can be a set of images processed by the image enhancement network. Exemplarily, the target image enhancement network can be a Zero-DCE network. It can be understood that the samples from the open source dataset account for more than 80% of the lane line dataset, and most of this data is collected under good lighting conditions, so the proportion of low-light images in the original data is relatively small. It is not appropriate to directly use the lane line dataset as a low-light enhancement dataset to train the Zero-DCE network. Therefore, the number of low-light images and normal-light images in the lane line dataset can be calculated by manual statistics, and low-light images are randomly sampled from the open source low-light image dataset to construct a low-light enhancement dataset, where the ratio between low-light images and normal-light images is 1:1.

[0068] Step S203: input the target enhanced image data set into the target lane detection network to obtain the target lane line detection result.

[0069] It should be noted that the target lane detection network can be used to identify and locate lane lines in an image, and the target lane line detection result is the output result obtained after processing by the lane detection network, and can include information about the lane line. Exemplarily, the target lane detection network can select a CLRNet network, and the Zero-DCE network and the CLRNet network can be connected in series, wherein the output end of the Zero-DCE network is used as the input end of the CLRNet network, so as to obtain a more accurate target lane line detection result.

[0070] In this embodiment, a low-light image dataset and a real-time road image dataset are obtained to diversify the datasets. By inputting the low-light image dataset into a target image enhancement network, the clarity and visibility of the image can be significantly improved. By inputting the obtained target enhanced image dataset into a target lane detection network, the target lane line detection result is obtained. The generated lane line detection result can be fed back to the automatic driving or driving assistance system in real time, so that the vehicle can make timely and accurate decisions in a complex environment, thereby improving driving safety.

[0071] Based on the above Figure 2 The second embodiment shown provides an implementation method of the lane line prediction method of the present application.

[0072] In one embodiment, step S202 includes: inputting the low-illumination image dataset into the first backbone network, extracting initial image features through deep separable convolution, and deep separable convolution is used to reduce the computational complexity of the model; performing global maximum pooling and global average pooling on the initial image features through a channel attention network to obtain intermediate image features; inputting the intermediate image features into a shared connection layer, dynamically adjusting the model's attention to each channel feature through an S-type function, and obtaining target image features; optimizing the image enhancement network through a first self-supervised loss network to obtain a target image enhancement network; inputting the target image features into the target image enhancement network to obtain a target enhanced image dataset.

[0073] For example, the Zero-DCE network contains seven convolutional layers, such as Figure 3 As shown in the figure, in order to reduce the amount of calculation and parameters of the model, on the basis of maintaining the basic network structure, the deep separable convolution is used to replace the traditional convolution layer; in order to enhance the feature expression ability of the model, the channel attention module is introduced into the backbone network of the model, that is, the SE channel attention module is added after each convolution layer; the channel attention mechanism performs global maximum pooling and global average pooling on the input features respectively, and then sends them to the shared connection layer of the two layers respectively, and then adds the outputs, and finally uses the Sigmoid function for activation operation, which can dynamically adjust the model's attention to each channel feature, while enhancing the key features and suppressing redundant feature information. The improved Zero-DCE network maintains the original image enhancement performance of the model while reducing the amount of model parameters and calculation.

[0074] It should be noted that the step of optimizing the image enhancement network through the first self-supervised loss network to obtain the target image enhancement network includes: reducing the exposure control loss through the sliding window method to obtain an improved exposure control loss; optimizing the image enhancement network based on the improved exposure control loss to obtain the target image enhancement network.

[0075] Exemplarily, the first self-supervised loss is the self-supervised loss of the Zero-DCE model, which can be composed of the following four parts: spatial consistency loss L spa , exposure control loss L exp , Color constant loss L col and lighting smoothness loss L tv , expressed as formula (1):

[0076] L1=λ1L spa +λ2L exp +λ3L col +λ4L tv (1)

[0077] In formula (1), λ1, λ2, λ3, and λ4 are L spa , L exp , L col and L tv The weight coefficient of .

[0078] Among them, the exposure control loss L exp The calculation formula is expressed as formula (2):

[0079]

[0080] In formula (2), M indicates that there are M 16×16 non-overlapping pixel blocks in the output low-light enhancement image, k indicates the kth pixel block in the traversal process, and Y k represents the average intensity value of the kth pixel block, and E is a constant between 0 and 1. The selection of E often needs to be based on the experience of the model trainer. During the training process, all samples need to be calculated based on this constant, which brings great difficulties to the training of the model. mean , I mean Represents the average intensity value of the entire low-light enhanced image. In addition, if non-overlapping pixel blocks are used for calculation, a large chromatic aberration effect may occur between the pixel blocks, resulting in a decrease in image quality. To address this problem, this embodiment uses a sliding window method to traverse the output low-light enhanced image, keeping the window size at 16×16, that is, the pixel block size remains unchanged, but the moving step is set to 8. As a result, the overlapping area between each pixel block and the surrounding adjacent pixel blocks accounts for one quarter of the pixel block. The number of pixel blocks obtained by the sliding window method is M N , the improved exposure control loss L expn The calculation formula is expressed as formula (3):

[0081]

[0082] Based on the improved self-supervised loss function, the improved Zero-DCE model is trained using the low-light enhancement dataset, and the model with the best comprehensive performance of peak signal-to-noise ratio (PSNR) and structural similarity (SSIM) on the test set is retained. Through the sliding window method, the model can dynamically adjust the exposure effect in different local areas, thereby effectively reducing the exposure control loss.

[0083] In this embodiment, by using deep separable convolution, the model becomes lighter and can achieve more efficient processing on devices with limited hardware resources. The improved Zero-DCE network not only reduces the amount of calculation through deep separable convolution, but also improves the feature expression ability by introducing the channel attention module. This enables the model to maintain excellent image enhancement performance while reducing the computational complexity. Through the SE channel attention module, the features are globally max-pooled and globally average-pooled after each convolution layer, so that the model can dynamically adjust the attention to different channel features. Through the adjustment of the S-type activation function, the model can enhance key features while suppressing redundant or unimportant feature information, thereby improving the overall feature expression ability. This is crucial to improving the effect of image enhancement, especially under complex lighting conditions, the retention and enhancement of key details can significantly improve image quality. By optimizing the image enhancement network by self-supervised learning, effective enhancement strategies can be automatically learned from the data without relying on a large amount of labeled data. The sliding window method further reduces the exposure control loss and improves the global image exposure effect by optimizing the local area.

[0084] Based on the above Figure 2 The second embodiment shown provides another implementation of the lane line prediction method of the present application.

[0085] In this embodiment, step S203 includes: inputting the target enhanced image data set into the second backbone network, and obtaining the initial lane line detection result by introducing shallow features; optimizing the lane detection network through the second self-supervised loss network to obtain the target lane detection network; inputting the initial lane line detection result into the target lane detection network to obtain the target lane line detection result.

[0086] For example, Figure 4 As shown in Figure 1, the backbone network of the CLRNet network is ResNet18, which contains a total of 5 convolutional layers. In its feature fusion stage, it draws on the top-down structure of the feature pyramid and updates the prediction results in combination with the feature map, where P t (t=0,1,2) are lane prior parameters, including starting point coordinates x, y and angle θ. In order to enhance the network’s positioning and detection accuracy of lane lines, this implementation introduces a shallow feature layer to enrich the feature scale ( Figure 5 ).

[0087] For example, the loss function of the CLRNet network consists of two parts, namely the regression loss L Focal and matching loss L LIoU , where the calculation formula of LIoU is expressed as formula (4):

[0088]

[0089] Where N means dividing the image vertically (i.e. from top to bottom) into N horizontal slices. denote the x-coordinates of the predicted lane and the real lane at the i-th horizontal slice, respectively, and r is a preset constant, usually set to half the lane width. LIoU It can be expressed as formula (5):

[0090] LIoU Loss = 1-LIoU (5)

[0091] It is understandable that only the horizontal x-coordinate is used in the matching process. This method may not be accurate enough for matching lane lines at inclinations because it ignores the changes in the vertical direction of the lane lines, resulting in poor detection performance in inclined or curved parts of the lane lines. In order to improve the matching accuracy in these cases, it is necessary to comprehensively consider the overall geometry and position of the lane lines in the image, including their extension and changes in the vertical direction. In order to solve the problem of poor matching effect at inclined lanes mentioned above, an improved matching loss is proposed.

[0092] It should be noted that the step of optimizing the lane detection network through the second self-supervised loss network to obtain the target lane detection network includes: reducing the matching loss by introducing a lane line variable in the vertical direction to obtain an improved matching loss; optimizing the lane detection network based on the improved matching loss to obtain the target lane detection network.

[0093] Exemplary, improved matching loss L LWIoU can be expressed as formula (6):

[0094] L LWIoU =1-LWIoU (6)

[0095] In formula (6), the new matching method corresponds to LWIoU, which can be expressed as formula (7):

[0096]

[0097] In formula (7), and Indicates the lane width obtained by fitting. For example, the calculation method is expressed as formula (8):

[0098]

[0099] In formula (8), represents the y coordinate of the predicted lane at the i-th horizontal slice, w lane is the lane width obtained through the prior knowledge of the data set and generated through the clustering algorithm. At this time, the loss function of the CLRNet network can be expressed as formula (9):

[0100] L2=β1L Focal +β2L LWIoU (9)

[0101] In formula (9), β1 and β2 are L Focal , L LWIoU The weight coefficient of .

[0102] In this embodiment, by inputting the target enhanced image dataset into the second backbone network and combining it with shallow features, the edges and local features in the image can be better captured. Shallow features usually contain fine-grained information, which helps to locate the preliminary lane lines and improve the accuracy of detection. The lane detection network is optimized through the second self-supervised loss network, reducing the dependence on large-scale manually annotated datasets. Self-supervised learning uses the internal features or structures of the data to optimize the model, which can improve the generalization ability of the model. By introducing lane line variables in the vertical direction, the longitudinal changes of the lane lines in the image can be more accurately captured, reducing the matching error during detection. Through the fusion of shallow features and deep features, the lane detection network can better cope with various road conditions and environmental interference (such as lighting, occlusion, etc.), thereby improving the stability and robustness of detection.

[0103] In order to further reduce the number of CLRNet network parameters and the amount of calculation, this application proposes an implementation method that can introduce structural reparameterization technology in the inference stage of the model. According to the linear additivity of convolution, its multi-branch structure is equivalently converted to a single-path structure, and its adjacent convolution operations and BN operations are merged into a new convolution operation. Finally, the structurally reparameterized CLRNet network and the Zero-DCE network are connected in series to predict the lane line image input in a low-light environment.

[0104] In this embodiment, by fusing the models, the number of parameters and the amount of calculation can be reduced while ensuring the prediction accuracy of the model remains unchanged, thereby improving the reasoning speed. While ensuring the accuracy of lane line detection, computing resources are saved, and further, the rapid response of lane line prediction is ensured, thereby improving driving safety.

[0105] It should be noted that the above examples are only used to understand the present application and do not constitute a limitation on the lane line prediction method of the present application. More simple transformations based on this technical concept are all within the scope of protection of the present application.

[0106] This application also provides a lane line prediction device, please refer to Figure 6 , the lane line prediction device comprises:

[0107] A data acquisition module 10 is used to collect real-time road data and generate an image to be tested based on the real-time road data;

[0108] A result acquisition module 20, used to input the image to be tested into a target detection network model to obtain a target lane line detection result, wherein the target detection network model is used to enhance image performance and optimize lane line detection accuracy;

[0109] The lane prediction module 30 is used to predict the lane line based on the target lane line detection result.

[0110] The lane line prediction device provided by the present application adopts the lane line prediction method in the above embodiment, which can solve the technical problem of low accuracy of lane line prediction in low illumination scenes. Compared with the prior art, the beneficial effects of the lane line prediction device provided by the present application are the same as the beneficial effects of the lane line prediction method provided by the above embodiment, and the other technical features in the lane line prediction device are the same as the features disclosed in the above embodiment method, which will not be repeated here.

[0111] The present application provides a lane line prediction device, which includes: at least one processor; and a memory communicatively connected to the at least one processor; wherein the memory stores instructions executable by the at least one processor, and the instructions are executed by the at least one processor so that the at least one processor can execute the lane line prediction method in the above-mentioned embodiment one.

[0112] Reference below Figure 7 , which shows a schematic diagram of the structure of a lane line prediction device suitable for implementing the embodiment of the present application. The lane line prediction device in the embodiment of the present application may include but is not limited to mobile terminals such as mobile phones, laptop computers, digital broadcast receivers, PDAs (Personal Digital Assistants), PADs (Portable Application Descriptions), PMPs (Portable Media Players), vehicle-mounted terminals (such as vehicle-mounted navigation terminals), etc., and fixed terminals such as digital TVs, desktop computers, etc. Figure 7The lane line prediction device shown is merely an example and should not bring any limitation to the functions and scope of use of the embodiments of the present application.

[0113] like Figure 7 As shown, the lane line prediction device may include a processing device 1001 (e.g., a central processing unit, a graphics processor, etc.), which can perform various appropriate actions and processes according to a program stored in a read-only memory (ROM: Read Only Memory) 1002 or a program loaded from a storage device 1003 to a random access memory (RAM: Random Access Memory) 1004. In RAM1004, various programs and data required for the operation of the lane line prediction device are also stored. The processing device 1001, ROM1002, and RAM1004 are connected to each other through a bus 1005. An input / output (I / O) interface 1006 is also connected to the bus. Generally, the following systems can be connected to the I / O interface 1006: an input device 1007 including, for example, a touch screen, a touchpad, a keyboard, a mouse, an image sensor, a microphone, an accelerometer, a gyroscope, etc.; an output device 1008 including, for example, a liquid crystal display (LCD: Liquid Crystal Display), a speaker, a vibrator, etc.; a storage device 1003 including, for example, a magnetic tape, a hard disk, etc.; and a communication device 1009. The communication device 1009 may allow the lane line prediction device to communicate with other devices wirelessly or by wire to exchange data. Figure 7 The lane line prediction device with various systems is shown, but it should be understood that it is not required to implement or have all the systems shown. More or fewer systems may be implemented or have instead.

[0114] In particular, according to the embodiments disclosed in the present application, the process described above with reference to the flowchart can be implemented as a computer software program. For example, the embodiments disclosed in the present application include a computer program product, which includes a computer program carried on a computer-readable medium, and the computer program includes a program code for executing the method shown in the flowchart. In such an embodiment, the computer program can be downloaded and installed from a network through a communication device, or installed from a storage device 1003, or installed from a ROM 1002. When the computer program is executed by the processing device 1001, the above-mentioned functions defined in the method of the embodiment disclosed in the present application are executed.

[0115] The lane line prediction device provided by the present application adopts the lane line prediction method in the above embodiment, which can solve the technical problem of low accuracy of lane line prediction in low illumination scenes. Compared with the prior art, the beneficial effects of the lane line prediction device provided by the present application are the same as the beneficial effects of the lane line prediction method provided by the above embodiment, and the other technical features in the lane line prediction device are the same as the features disclosed in the method of the previous embodiment, which will not be repeated here.

[0116] It should be understood that the various parts disclosed in this application can be implemented by hardware, software, firmware or a combination thereof. In the description of the above embodiments, specific features, structures, materials or characteristics can be combined in any one or more embodiments or examples in a suitable manner.

[0117] The above is only a specific implementation of the present application, but the protection scope of the present application is not limited thereto. Any person skilled in the art who is familiar with the present technical field can easily think of changes or substitutions within the technical scope disclosed in the present application, which should be included in the protection scope of the present application. Therefore, the protection scope of the present application should be based on the protection scope of the claims.

[0118] The present application provides a computer-readable storage medium having computer-readable program instructions (ie, computer programs) stored thereon, and the computer-readable program instructions are used to execute the lane line prediction method in the above-mentioned embodiment.

[0119] The computer-readable storage medium provided in the present application may be, for example, a USB flash drive, but is not limited to electrical, magnetic, optical, electromagnetic, infrared, or semiconductor systems, systems or devices, or any combination of the above. More specific examples of computer-readable storage media may include, but are not limited to: an electrical connection with one or more wires, a portable computer disk, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), an optical fiber, a portable compact disk read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination of the above. In this embodiment, the computer-readable storage medium may be any tangible medium containing or storing a program that can be used by or in combination with an instruction execution system, system or device. The program code contained on the computer-readable storage medium may be transmitted using any appropriate medium, including but not limited to: wires, optical cables, RF (Radio Frequency), etc., or any suitable combination of the above.

[0120] The computer-readable storage medium may be included in the lane line prediction device; or may exist independently without being assembled into the lane line prediction device.

[0121] The above-mentioned computer-readable storage medium carries one or more programs. When the above-mentioned one or more programs are executed by the lane line prediction device, the lane line prediction device: obtains real-time road data under low-light road conditions, and generates a test image based on the real-time road data; inputs the test image into the target detection network model to obtain a target lane line detection result, and the target detection network model is used to enhance image performance and optimize lane line detection accuracy; based on the target lane line detection result, predicts the lane line.

[0122] Computer program code for performing the operations of the present application may be written in one or more programming languages ​​or a combination thereof, including object-oriented programming languages ​​such as Java, Smalltalk, C++, and conventional procedural programming languages ​​such as "C" or similar programming languages. The program code may be executed entirely on the user's computer, partially on the user's computer, as a separate software package, partially on the user's computer and partially on a remote computer, or entirely on a remote computer or server. In the case of a remote computer, the remote computer may be connected to the user's computer through any type of network, including a local area network (LAN) or a wide area network (WAN), or may be connected to an external computer (e.g., via the Internet using an Internet service provider).

[0123] The flow chart and block diagram in the accompanying drawings illustrate the possible architecture, function and operation of the system, method and computer program product according to various embodiments of the present application. In this regard, each square box in the flow chart or block diagram can represent a module, a program segment or a part of a code, and the module, the program segment or a part of the code contains one or more executable instructions for realizing the specified logical function. It should also be noted that in some alternative implementations, the functions marked in the square box can also occur in a sequence different from that marked in the accompanying drawings. For example, two square boxes represented in succession can actually be executed substantially in parallel, and they can sometimes be executed in the opposite order, depending on the functions involved. It should also be noted that each square box in the block diagram and / or flow chart, and the combination of the square boxes in the block diagram and / or flow chart can be implemented with a dedicated hardware-based system that performs a specified function or operation, or can be implemented with a combination of dedicated hardware and computer instructions.

[0124] The modules involved in the embodiments described in this application may be implemented by software or hardware, wherein the name of the module does not constitute a limitation on the unit itself in some cases.

[0125] The readable storage medium provided in the present application is a computer-readable storage medium, which stores computer-readable program instructions (i.e., computer programs) for executing the above lane line prediction method, and can solve the technical problem of low accuracy of lane line prediction in low-light scenes. Compared with the prior art, the beneficial effects of the computer-readable storage medium provided in the present application are the same as the beneficial effects of the lane line prediction method provided in the above embodiment, and will not be repeated here.

[0126] The present application also provides a computer program product, including a computer program, which implements the steps of the lane line prediction method as described above when executed by a processor.

[0127] The computer program product provided by the present application can solve the technical problem of low accuracy of lane line prediction in low illumination scenes. Compared with the prior art, the beneficial effects of the computer program product provided by the present application are the same as the beneficial effects of the lane line prediction method provided by the above embodiment, which will not be repeated here.

[0128] The above descriptions are only some embodiments of the present application, and are not intended to limit the patent scope of the present application. All equivalent structural changes made using the contents of the present application specification and drawings under the technical concept of the present application, or direct / indirect applications in other related technical fields are included in the patent protection scope of the present application.

Claims

1. A lane line prediction method, characterized in that: The method includes: Acquire real-time road data under low-light road conditions, and generate an image to be tested based on the real-time road data; Inputting the image to be tested into a target detection network model to obtain a target lane line detection result, wherein the target detection network model is used to enhance image performance and optimize lane line detection accuracy; Based on the target lane line detection result, the lane line is predicted.

2. The method according to claim 1, characterized in that The target detection network model includes a target image enhancement network and a target lane detection network. The step of inputting the image data to be tested into the target detection network model to obtain a target lane detection result includes: Obtain low-light image datasets and road real-time image datasets; Inputting the low-light image dataset into a target image enhancement network to obtain a target enhanced image dataset; The target enhanced image data set is input into the target lane detection network to obtain the target lane line detection result.

3. The method according to claim 2, characterized in that The target image enhancement network includes a first backbone network, a channel attention network, and a first self-supervised loss network. The step of inputting the low-illumination image dataset into the target image enhancement network to obtain the target enhanced image dataset includes: Inputting the low-light image data set into the first backbone network, extracting initial image features through depthwise separable convolution, wherein the depthwise separable convolution is used to reduce the amount of calculation of the model; Performing global maximum pooling and global average pooling on the initial image features through the channel attention network to obtain intermediate image features; The intermediate image features are input into the shared connection layer, and the degree of attention paid by the model to each channel feature is dynamically adjusted through the S-type function to obtain the target image features; Optimizing the image enhancement network through the first self-supervised loss network to obtain a target image enhancement network; The target image features are input into a target image enhancement network to obtain a target enhanced image dataset.

4. The method according to claim 2, characterized in that The target lane detection network includes a second backbone network and a second self-supervised loss network. The step of inputting the target enhanced image data set into the target lane detection network to obtain a target lane line detection result includes: Inputting the target enhanced image data set into the second backbone network, and obtaining an initial lane line detection result by introducing shallow features; Optimizing the lane detection network through the second self-supervised loss network to obtain a target lane detection network; The initial lane line detection result is input into the target lane line detection network to obtain the target lane line detection result.

5. The method according to claim 3, characterized in that The first self-supervised function network includes an exposure control loss, and the step of optimizing the image enhancement network through the first self-supervised loss network to obtain a target image enhancement network includes: Reducing the exposure control loss by a sliding window method to obtain an improved exposure control loss; Based on the improved exposure control loss, the image enhancement network is optimized to obtain a target image enhancement network.

6. The method according to claim 4, characterized in that The second self-supervised function network includes a matching loss, and the step of optimizing the lane detection network by the second self-supervised loss network to obtain a target lane detection network includes: The matching loss is reduced by introducing a lane line variable in a vertical direction, thereby obtaining an improved matching loss; The lane detection network is optimized based on the improved matching loss to obtain a target lane detection network.

7. A lane line prediction device, characterized in that: The device comprises: A data acquisition module, used for collecting real-time road data and generating an image to be tested based on the real-time road data; A result acquisition module, used to input the image to be tested into a target detection network model to obtain a target lane line detection result, wherein the target detection network model is used to enhance image performance and optimize lane line detection accuracy; The lane prediction module is used to predict the lane line based on the target lane line detection result.

8. A lane line prediction device, characterized in that: The device comprises: a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the computer program is configured to implement the steps of the lane line prediction method according to any one of claims 1 to 6.

9. A storage medium, characterized in that: The storage medium is a computer-readable storage medium, and a computer program is stored on the storage medium. When the computer program is executed by a processor, the steps of the lane line prediction method according to any one of claims 1 to 6 are implemented.

10. A computer program product, characterized in that The computer program product comprises a computer program, and when the computer program is executed by a processor, the steps of the lane line prediction method according to any one of claims 1 to 6 are implemented.

Images