An image enhancement method, system, and device for autonomous driving vision tasks

Abstract

This invention proposes an image enhancement method, system, and device for autonomous driving vision tasks, belonging to the technical field of combining computer vision and autonomous driving. The method includes: acquiring multi-view images of a target to be detected and stitching adjacent view images; applying a first-size grid mask to the stitched image; clearing the mask from overlapping areas of the stitched image to obtain an intermediate image; detecting a preset target region in the intermediate image and applying a second-size grid mask to the preset target region; and replacing the first-size grid mask at the corresponding position with the second-size grid mask to obtain the enhanced image. Based on this method, an image enhancement system for autonomous driving vision tasks is also proposed. This invention employs a multi-view grid mask enhancement method, using a first-size mask and a second-size mask for global subject occlusion and local small target occlusion, respectively, improving the robustness and generalization ability of the visual perception model in autonomous driving systems under complex environments.

Description

Technical Field

[0001] This invention belongs to the technical field of computer vision and autonomous driving, and specifically relates to an image enhancement method, system and device for autonomous driving vision tasks. Background Technology

[0002] In the fields of image processing and computer vision, data augmentation techniques, as an important means to alleviate model overfitting and improve the generalization ability of deep learning models, have been widely applied to tasks such as object detection, image classification, and semantic segmentation. Current mainstream data augmentation methods can be summarized into three categories: Information removal: such as random erasure, cropping, and mesh masking; Spatial transformation: such as random cropping, random scaling, rotation, and mirroring; Color perturbation: such as changing the brightness and contrast of an image.

[0003] In the field of autonomous driving, mesh mask augmentation is a commonly used method. Mesh mask augmentation uses a uniformly arranged grid of squares to cover parts of an image, mimicking occlusion of unknown objects in real-world scenarios and enhancing data diversity. However, the effectiveness of data augmentation is often affected by the mask size. When the mask size is large, the excessive occlusion area turns ground truth data into noise, hindering the training process; when the mask size is small, the changes in the ground truth objects are subtle, amounting to only random noise coverage, resulting in limited data augmentation. Mesh mask augmentation was initially designed for single-object image classification tasks, such as ImageNet and CIFAR10. Its fixed mask size and simplified design make it difficult to adapt to complex environments with multiple scales and multiple objects, such as road scenes in autonomous driving. Large masks may occlude small distant objects, while small masks are insufficient to effectively perturb the main target, limiting the augmentation effect. Figure 1 This is a schematic diagram illustrating the use of mesh mask enhancement for single-target image detection in existing technologies; where... Figure 1 (a) Applying GridMask to different datasets; Figure 1 (b) shows the target detection results for autonomous driving. The mask grid suitable for enhancing the main vehicle (truck) can easily completely obscure distant vehicles and signs. Furthermore, in the surround-view image processing task for autonomous driving, there are overlapping areas between multiple adjacent camera views. If the masking method is not constrained, it may lead to repeated occlusion of key areas, affecting the accuracy of subsequent tasks such as depth estimation and 3D reconstruction. Summary of the Invention

[0004] To address the aforementioned technical problems, this invention proposes an image enhancement method, system, and device for autonomous driving vision tasks. These methods aim to improve the robustness and generalization ability of visual perception models in autonomous driving systems under complex environments.

[0005] To achieve the above objectives, the present invention adopts the following technical solution:

[0006] An image enhancement method for autonomous driving vision tasks includes the following steps:

[0007] Acquire multi-view images of the target to be detected, and stitch adjacent view images together to obtain a stitched image;

[0008] Apply a grid mask of the first size to the stitched image;

[0009] Remove the mask from the overlapping areas of the stitched images to obtain the intermediate image;

[0010] Detect a preset target region in the intermediate image, and apply a second-size grid mask to the preset target region;

[0011] The enhanced image is obtained by replacing the first-size grid mask at the corresponding position with the second-size grid mask.

[0012] This invention also proposes an image enhancement system for autonomous driving vision tasks, comprising:

[0013] The stitching module is used to acquire multi-view images of the target to be detected and stitch adjacent view images to obtain a stitched image.

[0014] A first application module is used to apply a grid mask of a first size to the stitched image;

[0015] The clearing module is used to clear the mask of the overlapping area of ​​the stitched image to obtain the intermediate image;

[0016] The second application module is used to detect a preset target region in the intermediate image and apply a second-size grid mask to the preset target region;

[0017] The replacement module is used to replace the first-size grid mask at the corresponding position with a second-size grid mask to obtain an enhanced image.

[0018] This invention also proposes an image enhancement device for autonomous driving vision tasks, comprising:

[0019] Memory, used to store computer programs;

[0020] A processor is configured to implement the steps of the image enhancement method for autonomous driving vision tasks when executing the computer program.

[0021] The effects described in the invention are merely those of the embodiments, and not all the effects of the invention. One of the above technical solutions has the following advantages or beneficial effects:

[0022] This invention proposes an image enhancement method, system, and device for autonomous driving vision tasks, comprising the following steps: acquiring multi-view images of a target to be detected, and stitching adjacent view images to obtain a stitched image; applying a first-size grid mask to the stitched image; clearing the mask in the overlapping area of ​​the stitched image to obtain an intermediate image; detecting a preset target region in the intermediate image, and applying a second-size grid mask to the preset target region; and replacing the first-size grid mask at the corresponding position with the second-size grid mask to obtain an enhanced image. Based on the image enhancement method for autonomous driving vision tasks, an image enhancement system for autonomous driving vision tasks is also proposed. This invention employs a multi-view grid mask enhancement method for multi-view data in autonomous driving, using a first-size mask and a second-size mask for global subject occlusion and local small target occlusion, respectively, thereby improving the robustness and generalization ability of the visual perception model in the autonomous driving system under complex environments. Attached Figure Description

[0023] Figure 1 This is a schematic diagram of the use of mesh mask enhancement for single-target image detection in the prior art.

[0024] Figure 2 This is a flowchart of an image enhancement method for autonomous driving vision tasks proposed in Embodiment 1 of the present invention;

[0025] Figure 3 This is a schematic diagram illustrating an example of multi-view mesh mask enhancement proposed in Embodiment 1 of the present invention;

[0026] Figure 4 This is a schematic diagram of an image enhancement system for autonomous driving vision tasks proposed in Embodiment 2 of the present invention;

[0027] Figure 5 This is a schematic diagram of an image enhancement device for autonomous driving vision tasks proposed in Embodiment 2 of the present invention. Detailed Implementation

[0028] To clearly illustrate the technical features of this solution, the invention will be described in detail below through specific embodiments and in conjunction with the accompanying drawings. The following disclosure provides many different embodiments or examples for implementing different structures of the invention. To simplify the disclosure of the invention, components and arrangements of specific examples are described below. Furthermore, reference numerals and / or letters may be repeated in different examples. This repetition is for simplification and clarity and does not in itself indicate a relationship between the various embodiments and / or arrangements discussed. It should be noted that the components illustrated in the drawings are not necessarily drawn to scale. Descriptions of well-known components, processing techniques, and processes are omitted in this invention to avoid unnecessarily limiting the invention.

[0029] Example 1

[0030] Embodiment 1 of this invention proposes an image enhancement method for autonomous driving vision tasks, addressing the shortcomings of existing technologies where fixed mask sizes and simplified designs make it difficult to adapt to the complex multi-scale, multi-target environments of autonomous driving vision tasks. This invention uses a first-size mask and a second-size mask for global subject occlusion and local small target occlusion, respectively, improving the robustness and generalization ability of the visual perception model in autonomous driving systems under complex environments.

[0031] Figure 2 This is a flowchart of an image enhancement method for autonomous driving vision tasks proposed in Embodiment 1 of the present invention;

[0032] In step S210, multi-view images of the target to be detected are acquired;

[0033] In this application, multi-view images are acquired through a panoramic camera, which can be multi-view panoramic images of different numbers. For example, 4 images (front view, left view, right view, rear view) or 6 images (left front view, front view, right front view, left rear view, rear view, right rear view).

[0034] In step S220, adjacent viewpoint images are stitched together to obtain a stitched image;

[0035] Choose any adjacent viewpoints for stitching according to task requirements. For example, select three images from six panoramic views: left front view, front view, and right front view for stitching. It is recommended that the resolution of the stitched image maintain a shorter side resolution of at least 1920 pixels to accommodate the requirements of the first-size mask.

[0036] In step S230, a first-size grid mask is applied to the stitched image;

[0037] The method for determining the first-size mesh mask is as follows: the side length of the first-size mesh mask is a preset ratio of the short side of the stitched image; the density calculation process of the first-size mesh mask is as follows: mask density = (area of ​​first-size mesh mask / image area) × 100%.

[0038] The squares of the first-size mesh mask have a side length that is 1 / 20th of the shorter side of the stitched image. For example, if the shorter side of the stitched image is 1920 pixels, the mask side length is 96 pixels.

[0039] The mask density is set to approximately 20%, meaning the spacing between each mask square is twice the side length, to avoid excessive occlusion. The first-size mesh mask effectively covers large nearby targets (such as trucks and buses), simulating occlusion phenomena in real-world scenes and enhancing the model's ability to focus on global features.

[0040] The calculation process for mask density is as follows:

[0041] Mask density = (Area of ​​first-size network mask / Image area) × 100%

[0042] = [ (side length² × number of rows × number of columns) / (W × H) ] × 100%

[0043] Where: number of rows = ceil( H / (side length + spacing) ); number of columns = ceil( W / (side length + spacing) ); spacing = side length × k ; k is the spacing coefficient, when k=2, the mask density is approximately 20%.

[0044] In step S240, the mask of the overlapping area of ​​the stitched image is removed to obtain the intermediate image;

[0045] The process of clearing the mask in the overlapping area of ​​the stitched image is as follows: Based on the camera calibration parameters, the overlapping area of ​​the multi-view images (such as the boundary between the front and left views) is calculated, and a mask clearing template is generated. The clearing width is half the side length of the mask to ensure that the depth estimation algorithm in subsequent tasks is not affected by occlusion.

[0046] Methods for calculating the overlapping region of multi-view images based on panoramic camera calibration parameters include:

[0047] splicing boundary point set Perform distortion correction; ;in, For the first A matrix of intrinsic parameters for a surround-view camera; The distortion coefficient;

[0048] The pixel coordinates after distortion correction Mapped to the world coordinate system; ; The representative is the world coordinate system; For the first A rotation matrix for a panoramic camera; ; For the first One pan-view camera translation vector;

[0049] Projected onto adjacent cameras The image plane is; ; For the first A matrix of intrinsic parameters for a surround-view camera; For the first A rotation matrix for a panoramic camera; For the first One pan-view camera translation vector;

[0050] Calculate the overlapping region: ; Represents overlapping areas.

[0051] Clear width = max(ceil(mask side length / 2), minimum protected pixels (e.g., 16px))

[0052] Example: If the mask side length is 96px, then the clear width is 48px.

[0053] In step S250, a preset target region is detected in the intermediate image;

[0054] The object detection model is used to identify a preset target region in the image. A wide range of object detection models can be selected, including Faster R-CNN, YOLO, VIT, and the DETR series, to obtain the detection results for the preset target region.

[0055] In step S260, a second-sized mesh mask is applied to the preset target area;

[0056] The second-sized mesh mask is half the square root of the area of ​​the smallest prediction box; for example, if a small target with an area of ​​400 pixels² is detected (such as a traffic sign in the distance), the mask side length is 10 pixels.

[0057] The density of the second-sized mesh mask is twice that of the first-sized mesh mask, approximately 40%, which enhances the occlusion effect on the preset target.

[0058] In step S270, the first-size mesh mask at the corresponding position is replaced by the second-size mesh mask; based on the coordinate information of the target detection box, the second-size mesh mask is accurately covered at the position of the first-size mesh mask to ensure the accuracy of local occlusion and the coherence of the global mask.

[0059] In step S280, the final mask image is obtained. The data-augmented image is used for subsequent tasks, and the replaced mask template is saved for reuse in subsequent training batches, avoiding the waste of computing power caused by repeated detection.

[0060] Figure 3This is a schematic diagram illustrating an example of multi-view grid mask enhancement proposed in Embodiment 1 of the present invention. The multi-view grid mask enhancement method takes processing a three-view image from the front view as an example. The multi-view images are stitched together, and the stitched result is then arranged into large-size masks using the same grid mask enhancement method. The large-size squares are set to 1 / 20 of the shorter side of the overall image. The purpose of the view stitching is to maintain the stability of the panoramic view mask. Furthermore, to avoid large-scale occlusion in the cross-regions of the multi-view images from adversely affecting depth estimation, the cross-regions of the view views in each image are masked out. Next, target recognition is performed on the input image using target detection models such as YOLO. Predicted boxes with dimensions smaller than 1 / 10 of the image's dimensions are filtered out as small target prediction boxes. Small-size masks are then arranged within the range of these small target prediction boxes, with the small-size squares set to 1 / 2 of the square root of the smallest prediction box area. To avoid wasting computational resources due to repeated target detection, the results are saved after the first detection, and the predicted boxes are directly called during subsequent training when using this data augmentation method. Finally, the large-size mask in the original position is replaced with a small-size mask within the range of the small target prediction box to obtain the final data augmentation result.

[0061] The Multi-view GridMask (MGM) proposed in this invention is more suitable for complex autonomous driving scenarios and surrounding images compared to the classic GridMask used for image classification tasks. Complex autonomous driving scenarios involve a wide range of object scale variations, and MGM dynamically adjusts the grid size, not adhering to a fixed size. Furthermore, there is depth information correlation between adjacent surrounding images; the grid masks for each image are generated uniformly, preserving complete information about the key camera view intersection range, i.e., edge dilution.

[0062] Embodiment 1 of this invention proposes an image enhancement method for vision tasks in autonomous driving. It adopts a multi-view grid mask enhancement method for multi-view data of autonomous driving, and uses a first-size mask and a second-size mask to perform global subject occlusion and local small target occlusion, respectively, thereby improving the robustness and generalization ability of the visual perception model in autonomous driving system in complex environments.

[0063] Example 2

[0064] Based on the image enhancement method for autonomous driving vision tasks proposed in Embodiment 1 of the present invention, Embodiment 2 of the present invention also proposes an image enhancement system for autonomous driving vision tasks. Figure 4 This is a schematic diagram of an image enhancement system for autonomous driving vision tasks proposed in Embodiment 2 of the present invention. The system includes: a stitching module, a first application module, a clearing module, a second application module, and a replacement module.

[0065] The stitching module is used to acquire multi-view images of the target to be detected and stitch adjacent view images to obtain a stitched image.

[0066] A first application module is used to apply a grid mask of a first size to the stitched image;

[0067] The clearing module is used to clear the mask of the overlapping area of ​​the stitched image to obtain the intermediate image;

[0068] The second application module is used to detect a preset target region in the intermediate image and apply a second-size grid mask to the preset target region;

[0069] The replacement module is used to replace the first-size grid mask at the corresponding position with a second-size grid mask to obtain an enhanced image.

[0070] In the stitching module of this application: the method for acquiring multi-view images of the target to be detected is as follows: a surround-view camera is used to acquire multi-view images of the target to be detected; the multi-view images include: left front view image, front view image, right front view image, left rear view image, rear view image, and right rear view image. Any adjacent view images can be flexibly selected for stitching according to task requirements.

[0071] In the first application module, the method for determining the first-sized mesh mask is as follows:

[0072] The side length of the first-size grid mask is a preset ratio of the shorter side of the stitched image;

[0073] The process of calculating the density of the first-size mesh mask is as follows: Mask density = (area of ​​the first-size mesh mask / image area) × 100%.

[0074] In the clearing module, the process of clearing the mask of the overlapping area of ​​the stitched images to obtain the intermediate image includes:

[0075] The overlapping area of ​​multi-view images is calculated based on the calibration parameters of the panoramic camera, and a mask removal template is generated; the removal width is determined; the width is a preset ratio of the side length of the first-size mesh mask.

[0076] Methods for calculating the overlapping region of multi-view images based on panoramic camera calibration parameters include:

[0077] splicing boundary point set Perform distortion correction; ;in, For the first A matrix of intrinsic parameters for a surround-view camera; The distortion coefficient;

[0078] The pixel coordinates after distortion correction Mapped to the world coordinate system; ; The representative is the world coordinate system; For the first A rotation matrix for a panoramic camera; ; For the first One pan-view camera translation vector;

[0079] Projected onto adjacent cameras The image plane is; ; For the first A matrix of intrinsic parameters for a surround-view camera; For the first A rotation matrix for a panoramic camera; For the first One pan-view camera translation vector;

[0080] Calculate the overlapping region: ; Represents overlapping areas.

[0081] Clear width = max(ceil(mask side length / 2), minimum protected pixels (e.g., 16px))

[0082] In the second application module, the process of detecting the preset target region in the intermediate image includes: using a target detection algorithm to identify the preset target region from the intermediate image.

[0083] The method for determining the second-size mesh mask is as follows: the second-size mesh mask is half the square root of the area of ​​the minimum prediction box, and the density of the second-size mesh mask is twice the density of the first-size mesh mask.

[0084] In the replacement module, the method for replacing the first-size grid mask at the corresponding position with the second-size grid mask to obtain the enhanced image is as follows: based on the coordinate information of the target detection box, the second-size grid mask is overlaid at the position of the first-size grid mask.

[0085] Embodiment 2 of this invention proposes an image enhancement system for vision tasks in autonomous driving. It adopts a multi-view grid mask enhancement method for multi-view data of autonomous driving, and uses a first-size mask and a second-size mask to perform global subject occlusion and local small target occlusion, respectively, thereby improving the robustness and generalization ability of the visual perception model in autonomous driving system in complex environments.

[0086] The description of the relevant parts of the image enhancement system for autonomous driving vision tasks provided in Embodiment 2 of this application can be found in the detailed description of the corresponding parts of the image enhancement method for autonomous driving vision tasks provided in Embodiment 1 of this application, and will not be repeated here.

[0087] Example 3

[0088] This invention also proposes an image enhancement device for vision tasks in autonomous driving. Figure 5 This is a schematic diagram of an image enhancement device for autonomous driving vision tasks according to Embodiment 2 of the present invention, comprising:

[0089] Memory, used to store computer programs;

[0090] When a processor executes the computer program, the method steps are as follows:

[0091] In step S210, multi-view images of the target to be detected are acquired;

[0092] In step S220, adjacent viewpoint images are stitched together to obtain a stitched image;

[0093] In step S230, a first-size grid mask is applied to the stitched image;

[0094] In step S240, the mask of the overlapping area of ​​the stitched image is removed to obtain the intermediate image;

[0095] In step S250, a preset target region is detected in the intermediate image;

[0096] In step S260, a second-sized mesh mask is applied to the preset target area;

[0097] In step S270, the first-size mesh mask at the corresponding position is replaced with the second-size mesh mask;

[0098] In step S280, the final mask image is obtained. The data-augmented image is used for subsequent tasks, and the replaced mask template is saved for reuse in subsequent training batches, avoiding the waste of computing power caused by repeated detection.

[0099] Embodiment 3 of the present invention proposes an image enhancement device for vision tasks in autonomous driving. It adopts a multi-view grid mask enhancement method for multi-view data of autonomous driving, and uses a first-size mask and a second-size mask to perform global subject occlusion and local small target occlusion respectively, thereby improving the robustness and generalization ability of the visual perception model in autonomous driving system in complex environments.

[0100] It should be noted that the present invention also provides an electronic device, including: a communication interface capable of interacting with other devices such as network devices; and a processor connected to the communication interface to enable information interaction with other devices, used to execute an image enhancement method for autonomous driving vision tasks provided by one or more of the above technical solutions when running a computer program, wherein the computer program is stored in a memory. In practical applications, the various components of the electronic device are coupled together through a bus system. It is understood that the bus system is used to realize the connection and communication between these components. In addition to a data bus, the bus system also includes a power bus, a control bus, and a status signal bus. The memory in the embodiments of this application is used to store various types of data to support the operation of the electronic device. Examples of this data include any computer program used to operate on the electronic device. It is understood that the memory can be volatile memory or non-volatile memory, or both. Non-volatile memory can be read-only memory (ROM), programmable read-only memory (PROM), erasable programmable read-only memory (EPROM), electrically erasable programmable read-only memory (EEPROM), ferromagnetic random access memory (FRAM), flash memory, magnetic surface memory, optical disc, or compact disc read-only memory (CD-ROM); magnetic surface memory can be disk storage or magnetic tape storage. Volatile memory can be random access memory (RAM), which is used as an external cache.By way of example, but not limitation, many forms of RAM are available, such as Static Random Access Memory (SRAM), Synchronous Static Random Access Memory (SSRAM), Dynamic Random Access Memory (DRAM), Synchronous Dynamic Random Access Memory (SDRAM), Double Data Rate Synchronous Dynamic Random Access Memory (DDRSDRAM), Enhanced Synchronous Dynamic Random Access Memory (ESDRAM), SyncLink Dynamic Random Access Memory (SLDRAM), and Direct Rambus Random Access Memory (DRRAM). The memories described in the embodiments of this application are intended to include, but are not limited to, these and any other suitable types of memory. The methods disclosed in the embodiments of this application can be applied to or implemented by a processor. The processor may be an integrated circuit chip with signal processing capabilities. In implementation, each step of the above method can be completed by integrated logic circuits in the processor hardware or by instructions in software form. The processor can be a general-purpose processor, a DSP (Digital Signal Processing, i.e., a chip capable of implementing digital signal processing technology), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. The processor can implement or execute the methods, steps, and logic block diagrams disclosed in the embodiments of this application. A general-purpose processor can be a microprocessor or any conventional processor, etc. The steps of the methods disclosed in the embodiments of this application can be directly manifested as execution by a hardware decoding processor, or execution by a combination of hardware and software modules in the decoding processor. The software modules can be located in a storage medium, which is located in memory. The processor reads the program from the memory and, in conjunction with its hardware, completes the steps of the aforementioned method. When the processor executes the program, it implements the corresponding processes in the various methods of the embodiments of this application; for simplicity, these will not be elaborated further here.

[0101] The description of the relevant parts of the image enhancement device for autonomous driving vision tasks provided in Embodiment 3 of this application can be found in the detailed description of the corresponding parts of the image enhancement method for autonomous driving vision tasks provided in Embodiment 1 of this application, and will not be repeated here.

[0102] It should be noted that, in this document, relational terms such as "first" and "second" are used merely to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that the elements inherent in a process, method, article, or apparatus that includes a list of elements are included. Without further limitations, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes said element. Additionally, portions of the technical solutions provided in the embodiments of this application that are consistent with the implementation principles of corresponding technical solutions in the prior art have not been described in detail to avoid excessive elaboration.

[0103] While specific embodiments of the present invention have been described above in conjunction with the accompanying drawings, this is not intended to limit the scope of protection of the present invention. Those skilled in the art can make other modifications or variations based on the above description. It is neither necessary nor possible to exhaustively describe all embodiments here. Various modifications or variations that can be made by those skilled in the art without creative effort based on the technical solutions of the present invention are still within the scope of protection of the present invention.

Claims

1. An image enhancement method for vision tasks in autonomous driving, characterized in that, Includes the following steps: Acquire multi-view images of the target to be detected, and stitch adjacent view images together to obtain a stitched image; Apply a grid mask of the first size to the stitched image; The method for determining the first size grid mask is as follows: the side length of the first size grid mask is a preset ratio of the short side of the stitched image; The process for calculating the density of the first-size mesh mask is as follows: Mask density = (Area of ​​the first-size mesh mask / Image area) × 100%; Remove the mask from the overlapping areas of the stitched images to obtain the intermediate image; Detect a preset target region in the intermediate image, and apply a second-size grid mask to the preset target region; the second-size grid mask is determined by the following method: the second-size grid mask is half the square root of the area of ​​the minimum prediction box, and the density of the second-size grid mask is twice the density of the first-size grid mask; The enhanced image is obtained by replacing the first-size grid mask at the corresponding position with the second-size grid mask.

2. The image enhancement method for autonomous driving vision tasks according to claim 1, characterized in that, A method for acquiring multi-view images of a target to be detected: using a surround-view camera to acquire multi-view images of the target to be detected; the multi-view images include: left front view image, front view image, right front view image, left rear view image, rear view image and right rear view image.

3. The image enhancement method for autonomous driving vision tasks according to claim 1, characterized in that, The process of removing the mask from the overlapping areas of the stitched images to obtain the intermediate image includes: The overlapping region of multi-view images is calculated based on the calibration parameters of the panoramic camera, and a mask removal template is generated. Determine the clearing width; the width is a preset ratio of the side length of the first-size mesh mask.

4. The image enhancement method for autonomous driving vision tasks according to claim 3, characterized in that, Methods for calculating the overlapping region of multi-view images based on panoramic camera calibration parameters include: splicing boundary point set Perform distortion correction; ;in, For the first A matrix of intrinsic parameters for a surround-view camera; The distortion coefficient; The pixel coordinates after distortion correction Mapped to the world coordinate system; ; The representative is the world coordinate system; For the first A rotation matrix for a panoramic camera; ; For the first One pan-view camera translation vector; Projected onto adjacent cameras The image plane is; ; For the first A matrix of intrinsic parameters for a surround-view camera; For the first A rotation matrix for a panoramic camera; For the first One pan-view camera translation vector; Calculate the overlapping region: ; Represents overlapping areas.

5. The image enhancement method for autonomous driving vision tasks according to claim 1, characterized in that, The process of detecting a preset target region in an intermediate image includes: using a target detection algorithm to identify the preset target region from the intermediate image.

6. The image enhancement method for autonomous driving vision tasks according to claim 1, characterized in that, The method for obtaining an enhanced image by replacing the first-size grid mask at the corresponding position with the second-size grid mask is as follows: based on the coordinate information of the target detection box, the second-size grid mask is covered at the position of the first-size grid mask.

7. An image enhancement system for autonomous driving vision tasks, used to perform the image enhancement method for autonomous driving vision tasks as described in any one of claims 1 to 6, characterized in that, include: The stitching module is used to acquire multi-view images of the target to be detected and stitch adjacent view images to obtain a stitched image. A first application module is used to apply a grid mask of a first size to the stitched image; The clearing module is used to clear the mask of the overlapping area of ​​the stitched image to obtain the intermediate image; The second application module is used to detect a preset target region in the intermediate image and apply a second-size grid mask to the preset target region; The replacement module is used to replace the first-size grid mask at the corresponding position with a second-size grid mask to obtain an enhanced image.

8. An image enhancement device for autonomous driving vision tasks, characterized in that, include: Memory, used to store computer programs; A processor, configured to implement the steps of an image enhancement method for an autonomous driving vision task as described in any one of claims 1 to 6 when executing the computer program.

Images