Brick grassland parking space identification precision improvement method and system

Abstract

The invention discloses a brick grassland parking space recognition precision improvement method and system. The method comprises the steps of obtaining a fisheye spliced image of a brick grassland parking space; inputting the fisheye spliced image into a trained visual semantic segmentation model to obtain a semantic segmentation result graph; according to the semantic segmentation result map, screening out a parking space segmentation line area shielded by the grassland; repairing and restoring the parking space segmentation line area shielded by the grassland to obtain a repaired fisheye spliced image of the brick grassland parking space; and carrying out parking space identification according to the repaired fisheye spliced image of the brick grassland parking space. According to the method provided by the invention, the shielding influence of the grassland on the brick grassland parking space division line can be removed, the recognition precision of the brick grassland parking space is improved, the applicability of a parking sensing system is improved, and the parking system supports automatic parking of the shielded brick grassland parking space.

Description

Technical Field

[0001] The present invention relates to the technical field of automatic parking, and in particular to a method and system for improving the recognition accuracy of brick-grass parking spaces. Background Art

[0002] At present, the intelligent driving systems of autonomous vehicles are equipped with valet parking, memory parking and automatic parking functions. This function is generally based on the perception results provided by the autonomous vehicle to realize the intelligent parking function. The automatic parking function requires the perception system to input the detection of parking spaces, and perform path planning and parking operations based on the identified parking spaces. However, under special working conditions and some difficult scenarios, the parking space recognition ability and accuracy will be greatly reduced. For example, in summer, some brick grass parking spaces will have lush grass that blocks part of the parking space dividing line, resulting in the perception system being unable to identify the exact parking space position and shape, and ultimately resulting in the automatic parking system being unable to support automatic parking in such parking spaces, affecting the user experience and the applicability of the function.

[0003] At present, the perception system in the parking field mainly recognizes parking spaces based on images from fisheye cameras through deep learning methods. This method requires relatively clear parking space features to accurately identify parking spaces. However, in actual working conditions, for brick-grass parking spaces, it is very easy for the grass to be too high or too messy to block the parking space dividing line. At this time, the perception system may not be able to recognize the parking space, causing the parking space to be ignored or filtered out, and unable to support parking. Summary of the invention

[0004] The technical problem to be solved by the present invention is to provide a method and system for improving the recognition accuracy of brick-grass parking spaces in response to the shortcomings of the prior art. The method can remove the obstruction effect of grass on the brick-grass parking space dividing line, improve the recognition accuracy of brick-grass parking spaces, improve the applicability of the parking perception system, and enable the parking system to support automatic parking in obstructed brick-grass parking spaces.

[0005] To achieve the above object, according to one aspect of the present invention, a method for improving the recognition accuracy of a brick-grass parking space is provided, comprising:

[0006] Get the fisheye stitched image of the brick-grass parking space;

[0007] Inputting the fisheye stitching image into a trained visual semantic segmentation model to obtain a semantic segmentation result image;

[0008] Filter out the parking space dividing line area blocked by the grass according to the semantic segmentation result map;

[0009] Repairing and restoring the parking space dividing line area blocked by the grass to obtain a repaired fisheye stitching image of the brick grass parking space;

[0010] Parking space recognition is performed based on the restored fisheye stitching image of the brick-grass parking space.

[0011] In the above scheme, the method for filtering out the parking space dividing line area blocked by the grass according to the semantic segmentation result map is:

[0012] The category of the position of each pixel in the semantic segmentation result map belongs to one of the background area, the grass area, and the parking space segmentation line area, and its category is recorded as 0, 1, and 2 respectively. The semantic segmentation result is recorded as M se ;

[0013] Send out ray γ along the center of the rear axle of the vehicle, find the category of the pixel point at the corresponding position of ray γ in the semantic segmentation result image, and put the continuous pixel points of category 2 into the point set L i In the semantic segmentation result map, the independent pixel points of category 2 are placed into the point set L i+1 After completion, the ray γ is rotated to complete the judgment of all pixel points in the semantic segmentation result map, and the parking space segmentation line area set L is screened out. split , where L split ={L i}; The parking space dividing line area set L split Contains several point sets L i ;

[0014] For the parking space dividing line area set L split All point sets L i The two straight line boundaries of the parking space dividing line area are obtained by straight line fitting, wherein the expressions of the two straight line boundaries of the parking space dividing line area are I1 and I2 respectively, and the parking space dividing line area set L split All point sets L i The two straight line boundaries corresponding to the parking space dividing line area, namely L i →{I1, I2};

[0015] Calculate the area L of different parking space dividing lines i and L i+1 The intersection of the two straight line boundaries {I1, I2}, when the parking space dividing line area L i and L i+1 The two straight line boundaries {I1, I2} have intersection points at the same time, which separates the area surrounded by the straight line boundaries of the two parking space division areas and the parking space division line area L i and L i+1 Merge into a new parking space line area L i , delete the parking space dividing line area L i+1 , that is, the parking space dividing line area set L split Updates have been made;

[0016] The semantic segmentation result M se The grass area of ​​category 1 and the parking space dividing line area set L split Find the intersection and calculate the parking space dividing line area set L blocked by the grass roi , the calculation method is as follows:

[0017]

[0018] Among them, L idx is the area occupied by a point set in the parking space dividing line area set, For L idx The category value of the pixel points in the intersection area.

[0019] In the above scheme, the method for repairing and restoring the parking space dividing line area blocked by the grass to obtain the brick grass parking space repaired image is:

[0020] According to the parking space dividing line area set L blocked by the grass roi and the parking space dividing line area set L split , let {L roi_i ∈L roi , L split_i ∈L split} is a matching pair, storing the parking space dividing line area set L blocked by the grass roi and the parking space dividing line area set L split The point set of

[0021] From L split_i Click to delete L roi_i The pixel points in the point set are used to filter out the original parking space dividing line area point set Right now:

[0022]

[0023] For the original parking space dividing line area point set The middle distance is the boundary line of the parking space area blocked by the grass i The distance of pixels greater than the distance threshold is repaired. line i is the boundary line fitted by the pixel points in the parking space line area blocked by grass, and α is the preset distance threshold; the specific method of the repair is: let the fisheye stitching image of the brick grass parking space be IMG, calculate the original parking space segmentation line area point set corresponding to the fisheye stitching image IMG of the brick grass parking space The mean value C of the pixel values ​​greater than the distance threshold α rgb_i ,Right now:

[0024]

[0025] Using the mean value C of the pixel values rgb_i Re-focus on L roi_i The area of ​​the fisheye stitching image IMG corresponding to the brick grass parking space is filled with RGB values, that is, IMg(p ij )=C rgb_i ,p ij ∈L roi_i , and obtain the restored fisheye stitching image of the brick lawn parking space The general image restoration method is to perform pixel restoration through nearest neighbor interpolation. However, in the usage scenario of the present invention, on sunny days, the parking lines near the grass-blocking area will be shadowed by grass and trees. In order to avoid the repaired parking line area being affected by the shadow, the method provided by the present invention is used for image restoration.

[0026] In the above scheme, the method for identifying the parking space according to the restored fisheye stitching image of the brick-grass parking space is:

[0027] The fisheye stitching image of the restored brick lawn parking space Input into the parking space recognition algorithm for parking space recognition.

[0028] In the above scheme, the training method of the visual semantic segmentation model is:

[0029] Parking scene pictures with grass occlusion are collected to form a model training data set, and the backbone network of the simplified semantic segmentation model is trained on the model training data set to obtain pre-training weights; the semantic segmentation model loaded with the pre-training weights is incrementally trained to obtain a trained visual semantic segmentation model.

[0030] In the above scheme, the method for obtaining the simplified semantic segmentation model is:

[0031] The detection head of the visual semantic segmentation model is simplified, and a simplified semantic segmentation model is designed that only outputs three categories: grass area, parking space dividing line area and background area. Simplifying the detection head of the visual semantic segmentation model can reduce the computational time of semantic segmentation.

[0032] In the above scheme, the brick-turf parking space is a parking space with two straight-line boundaries, grass grows on the brick-turf parking space to form a lawn, and the lawn partially blocks the two straight-line boundaries.

[0033] In the above scheme, the visual semantic segmentation model includes SegNet, U-Net, and DeepLab.

[0034] In the above scheme, the parking space recognition algorithm includes Mobilenetv2, repvgg, and resnet.

[0035] According to another aspect of the present invention, a system for improving the recognition accuracy of a brick-grass parking space is provided, comprising:

[0036] An acquisition module, used to acquire a fisheye stitching image of a brick-grass parking space;

[0037] A segmentation module, used for inputting the fisheye stitching image into a trained visual semantic segmentation model to obtain a semantic segmentation result map;

[0038] A screening module, used for screening out the parking space dividing line area blocked by the grass according to the semantic segmentation result map;

[0039] A repair module, used to repair and restore the parking space dividing line area blocked by the grass to obtain a repaired fisheye stitching image of the brick grass parking space;

[0040] The recognition module is used to recognize the parking space according to the restored fisheye stitching image of the brick-grass parking space.

[0041] In general, the above technical solutions conceived by the present invention can achieve the following beneficial effects compared with the prior art:

[0042] (1) The present invention provides a method for improving the accuracy of parking space recognition on a brick lawn. The method solves the problem of reduced parking space recognition accuracy caused by partial obstruction of parking space lines by grass and trees by screening and repairing the parking space dividing line areas in the fisheye mosaic image.

[0043] (2) The present invention provides a method for improving the accuracy of parking space recognition on a brick-grass lawn. By simplifying and pre-training the detection head of the visual semantic segmentation model, a better segmentation effect is obtained, thereby improving the detection rate and accuracy of the model and algorithm used for parking space recognition.

[0044] (3) The method for improving the recognition accuracy of brick-grass parking spaces provided by the present invention is highly universal and applicable to most types of vehicles. It can be operated modularly without changing the original parking perception algorithm, thereby improving the applicability of the function and customer experience. BRIEF DESCRIPTION OF THE DRAWINGS

[0045] Various other advantages and benefits will become apparent to those of ordinary skill in the art by reading the detailed description of the preferred embodiments below. The accompanying drawings are only for the purpose of illustrating the preferred embodiments and are not to be considered as limiting the present invention. Also, throughout the accompanying drawings, the same reference figures are used to represent the same components. In the drawings:

[0046] Figure 1The present invention is a flowchart of a method for improving the accuracy of brick-grass parking space recognition according to an embodiment of the present invention.

[0047] Figure 2 This is a schematic diagram of the process of performing semantic segmentation on a fisheye stitched image using a trained visual semantic segmentation model in an embodiment of the present invention.

[0048] Figure 3 It is a schematic diagram of a process of filtering out a parking space dividing line area blocked by grass according to a semantic segmentation result map in an embodiment of the present invention. DETAILED DESCRIPTION

[0049] In order to make the purpose, technical solutions and advantages of the present invention more clearly understood, the present invention is further described in detail below in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not intended to limit the present invention. In addition, the technical features involved in the various embodiments of the present invention described below can be combined with each other as long as they do not conflict with each other.

[0050] It should be understood that the size of the serial numbers of the steps in the embodiments does not mean the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of the present application.

[0051] Example 1

[0052] The present application provides a method for improving the accuracy of brick-grass parking space recognition. Figure 1 ,include:

[0053] S1, obtain the fisheye stitching image of the brick grass parking space.

[0054] In an embodiment of the present application, the vehicle has a scene recognition function, which can detect whether the parking space to be parked is a brick grass parking space. When the parking space to be parked is a brick grass parking space, a fisheye stitching image of the brick grass parking space is obtained for the next step of the present method.

[0055] It can be understood that in the embodiment of the present application, the brick grass parking space is a parking space with two straight line boundaries. When a vehicle is parked normally in the brick grass parking space, the two straight line boundaries are parallel to the side body lines of the vehicle; grass grows on the brick grass parking space to form a lawn, and the lawn partially blocks the two straight line boundaries, and any straight line boundary of the brick grass parking space is not completely blocked by the grass.

[0056] S2, input the fisheye stitching image into the trained visual semantic segmentation model to obtain the semantic segmentation result map.

[0057] Specifically, see Figure 2; The visual semantic segmentation models used in the embodiments of the present application include but are not limited to SegNet, U-Net, and DeepLab.

[0058] S3, filter out the parking space dividing line area blocked by the grass according to the semantic segmentation result map.

[0059] In the embodiments of this application, please refer to Figure 3 ,The method of filtering out the parking space segmentation line area blocked by the grass according to the semantic segmentation result map is:

[0060] The category of each pixel position in the semantic segmentation result map belongs to one of the background area, grass area, and parking space segmentation line area, and its category is recorded as 0, 1, and 2 respectively. The semantic segmentation result in the semantic segmentation result map is recorded as M se ;

[0061] Send out ray γ along the center of the rear axle of the vehicle, find the category of the pixel points at the corresponding position of ray γ in the semantic segmentation result map, and put the continuous pixel points of category 2 into the point set L i In the semantic segmentation result map, the independent pixel points of category 2 are put into the point set L i+1 After that, the ray γ is rotated to complete the judgment of all pixels in the semantic segmentation result map, and the parking space segmentation line area set L is screened out. split , where L split ={L i}; that is, the parking space dividing line area set L split Contains several point sets L i ;

[0062] For the parking space dividing line area set L split All point sets L i The two straight line boundaries of the parking space dividing line area are obtained by straight line fitting, where the expressions of the two straight line boundaries of the parking space dividing line area are I1 and I2 respectively, and the parking space dividing line area set L split All point sets L i The two straight line boundaries corresponding to the parking space dividing line area, namely L i →{I1, I2};

[0063] Calculate the area L of different parking space dividing lines i and L i+1 The intersection of the two straight line boundaries {I1, I2}, when the parking space dividing line area L i and L i+1 The two straight line boundaries {I1, I2} have intersection points at the same time, which separates the area surrounded by the straight line boundaries of the two parking space division areas and the parking space division line area L i and L i+1 Merge into a new parking space line area L i, delete the parking space dividing line area L i+1 , that is, the parking space dividing line area set L split Updates have been made;

[0064] The semantic segmentation result M se The set L of grass areas and parking space dividing line areas of category 1 in split Find the intersection and calculate the parking space dividing line area set L blocked by the grass roi , the calculation method is as follows:

[0065]

[0066] Among them, L idx is the area occupied by a point set in the parking space dividing line area set, For L idx The category value of the pixel points in the intersection area.

[0067] Through this step, the parking space dividing line area blocked by the grass is screened out.

[0068] S4, repairing and restoring the parking space dividing line area blocked by the grass to obtain a repaired fisheye stitching image of the brick grass parking space.

[0069] In the embodiment of the present application, specifically, the method for repairing and restoring the parking space dividing line area blocked by the grass to obtain the brick grass parking space repair image is:

[0070] According to the parking space dividing line area set L blocked by the grass roi and the parking space dividing line area set L split , let {L roi_i ∈L roi , L split_i ∈L split} is a matching pair, storing the set of parking space dividing line areas blocked by grass L roi and the parking space dividing line area set L split The point set of

[0071] From L split_i Click to delete L roi_i The pixel points in the point set are used to filter out the original parking space dividing line area point set Right now:

[0072]

[0073] For the original parking space dividing line area point set The boundary line of the parking space area blocked by grass in the middle distance i The distance of pixels greater than the distance threshold is repaired. linei is a boundary line fitted by pixel points in the parking space line area blocked by grass, α is a preset distance threshold, and in this embodiment, α is 0.6 m.

[0074] The specific method of restoration is: let the fisheye stitching image of the brick grass parking space be IMG, calculate the original parking space dividing line area point set corresponding to the fisheye stitching image IMG of the brick grass parking space The mean value C of the pixel values ​​greater than the distance threshold α rgb_i ,Right now:

[0075]

[0076] Using the mean pixel value C rgb_i Re-focus on L roi_i The area of ​​the fisheye stitching image IMG corresponding to the brick grass parking space is filled with RGB values, that is, IMG(p ij )=C rgb_i ,p ij ∈L roi_i , and obtain the restored fisheye stitching image of the brick lawn parking space In the usage scenario of the present invention, on sunny days, the parking space lines close to the grass-blocking area will be shadowed by grass and trees. In order to avoid the repaired parking space line area being affected by the shadow, this method is used for image restoration instead of the nearest neighbor interpolation method used in general image restoration.

[0077] S5, performing parking space recognition based on the restored fisheye stitching image of the brick-grass parking space.

[0078] In the embodiment of the present application, specifically, the method for identifying a parking space based on the restored fisheye stitching image of the brick-grass parking space is:

[0079] The restored fisheye stitching image of the brick lawn parking space Input into the parking space recognition algorithm for parking space recognition.

[0080] Specifically, the parking space recognition algorithm used in the embodiment of the present application is one of Mobilenetv2, repvgg, and resnet.

[0081] An embodiment of the present application provides a method for improving the accuracy of parking space recognition on a brick-turf lawn. The method solves the problem of reduced parking space recognition accuracy caused by partial obstruction of parking space lines by grass and trees by screening and repairing the parking space dividing line areas in the fisheye stitching image. The method has high universality and is applicable to most vehicle models. It can be operated in a modular manner without changing the original parking perception algorithm, thereby improving the applicability of the function and customer experience.

[0082] Example 2

[0083] On the one hand, an embodiment of the present application provides a training method for a visual semantic segmentation model.

[0084] In the embodiments of this application, please refer to Figure 2 , the training method of the visual semantic segmentation model is:

[0085] Parking scene pictures with grass occlusion are collected to form a model training data set. The backbone network of the simplified semantic segmentation model is trained on the model training data set to obtain pre-training weights. The semantic segmentation model loaded with the pre-training weights is incrementally trained to obtain a trained visual semantic segmentation model.

[0086] Among them, the method for obtaining the simplified semantic segmentation model is:

[0087] The detection head of the visual semantic segmentation model is simplified, and a simplified semantic segmentation model is designed to output only three categories: grass area, parking space dividing line area and background area. In the embodiment of the present application, the simplification of the detection head of the visual semantic segmentation model can reduce the computational time of semantic segmentation.

[0088] Another aspect of the present application embodiment provides a system for improving the accuracy of brick-grass parking space recognition, including:

[0089] An acquisition module, used to acquire a fisheye stitching image of a brick-grass parking space;

[0090] The segmentation module is used to input the fisheye stitching image into the trained visual semantic segmentation model to obtain the semantic segmentation result map;

[0091] A screening module is used to screen out the parking space dividing line area blocked by the grass according to the semantic segmentation result map;

[0092] A repair module is used to repair and restore the parking space dividing line area blocked by the grass to obtain a repaired fisheye stitching image of the brick grass parking space;

[0093] The recognition module is used to recognize the parking space according to the restored fisheye stitching image of the brick-grass parking space.

[0094] The embodiments of the present application provide a method and system for improving the accuracy of brick-grass parking space recognition. The method simplifies and pre-trains the detection head of the visual semantic segmentation model to achieve better segmentation effects, thereby improving the detection rate and accuracy of the model and algorithm used for parking space recognition.

[0095] It should be pointed out that, according to the needs of implementation, the various steps described in this application can be divided into more steps, or two or more steps or partial operations of the steps can be combined into a new step to achieve the purpose of the present invention.

[0096] It will be easily understood by those skilled in the art that the above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions and improvements made within the spirit and principles of the present invention should be included in the protection scope of the present invention.

Claims

1. A method for improving the recognition accuracy of brick-grass parking spaces, characterized in that: include: Get the fisheye stitched image of the brick-grass parking space; Inputting the fisheye stitching image into a trained visual semantic segmentation model to obtain a semantic segmentation result image; Filter out the parking space dividing line area blocked by the grass according to the semantic segmentation result map; Repairing and restoring the parking space dividing line area blocked by the grass to obtain a repaired fisheye stitching image of the brick grass parking space; Parking space recognition is performed based on the restored fisheye stitching image of the brick-grass parking space.

2. The method for improving the recognition accuracy of brick-grass parking spaces according to claim 1, characterized in that: The training method of the visual semantic segmentation model is: Collecting pictures of parking spaces blocked by grass to form a model training data set, and training a backbone network of a simplified semantic segmentation model on the model training data set to obtain pre-training weights; Incremental training is performed on the semantic segmentation model after the pre-training weights are loaded to obtain a trained visual semantic segmentation model.

3. The method for improving the recognition accuracy of brick-grass parking spaces according to claim 1, characterized in that: The method for filtering out the parking space dividing line area blocked by the grass according to the semantic segmentation result map is: The category of the position of each pixel in the semantic segmentation result map belongs to one of the background area, the grass area, and the parking space segmentation line area, and its category is recorded as 0, 1, and 2 respectively. The semantic segmentation result is recorded as M se ; Send out ray γ along the center of the rear axle of the vehicle, find the category of the pixel point at the corresponding position of ray γ in the semantic segmentation result image, and put the continuous pixel points of category 2 into the point set L i In the semantic segmentation result map, the independent pixel points of category 2 are placed into the point set L i+1 After completion, the ray γ is rotated to complete the judgment of all pixel points in the semantic segmentation result map, and the parking space segmentation line area set L is screened out. split , where L split ={L i }; The parking space dividing line area set L split Contains several point sets L i ; For the parking space dividing line area set L split All point sets L i The two straight line boundaries of the parking space dividing line area are obtained by straight line fitting, wherein the expressions of the two straight line boundaries of the parking space dividing line area are I1 and I2 respectively, and the parking space dividing line area set L split All point sets L i The two straight line boundaries corresponding to the parking space dividing line area, namely L i →{I1, I2}; Calculate the area L of different parking space dividing lines i and L i+1 The intersection of the two straight line boundaries {I1, I2}, when the parking space dividing line area L i and L i+1 The two straight line boundaries {I1, I2} have intersection points at the same time, which separates the area surrounded by the straight line boundaries of the two parking space division areas and the parking space division line area L i and L i+1 Merge into a new parking space line area L i , delete the parking space dividing line area L i+1 , that is, the parking space dividing line area set L split Updates have been made; The semantic segmentation result M se The grass area of ​​category 1 and the parking space dividing line area set L split Find the intersection and calculate the parking space dividing line area set L blocked by the grass roi , the calculation method is as follows: Among them, L idx is the area occupied by a point set in the parking space dividing line area set, For L idx The category value of the pixel points in the intersection area.

4. The method for improving the recognition accuracy of brick-grass parking spaces according to claim 3 is characterized in that: The method for repairing and restoring the parking space dividing line area blocked by the grass to obtain a brick grass parking space repair image is: According to the parking space dividing line area set L blocked by the grass roi and the parking space dividing line area set L split , let {L roi_i ∈L roi , L split_i ∈L split } is a matching pair, storing the parking space dividing line area set L blocked by the grass roi and the parking space dividing line area set L split The point set of From L split_i Click to delete L roi_i The pixel points in the point set are used to filter out the original parking space dividing line area point set Right now: For the original parking space dividing line area point set The middle distance is the boundary line of the parking space area blocked by the grass i The distance of pixels greater than the distance threshold is repaired. line i is the boundary line fitted by the pixel points in the parking space line area blocked by grass, and α is the preset distance threshold; the specific method of the repair is: let the fisheye stitching image of the brick grass parking space be IMG, calculate the original parking space segmentation line area point set corresponding to the fisheye stitching image IMG of the brick grass parking space The mean value C of the pixel values ​​greater than the distance threshold α rgb_i ,Right now: Using the mean value C of the pixel values rgb_i Re-focus on L roi_i The area of ​​the fisheye stitching image IMG corresponding to the brick grass parking space is filled with RGB values, that is, IMG(p ij )=C rgb_i ,p ij ∈L roi_i , and obtain the restored fisheye stitching image of the brick lawn parking space 5. The method for improving the recognition accuracy of brick-grass parking spaces according to claim 4 is characterized in that: The method for identifying parking spaces based on the restored fisheye stitching image of the brick-grass parking space is: The fisheye stitching image of the restored brick lawn parking space Input into the parking space recognition algorithm for parking space recognition.

6. The method for improving the recognition accuracy of brick-grass parking spaces according to claim 2, characterized in that: The method to obtain the simplified semantic segmentation model is: The detection head of the visual semantic segmentation model is simplified, and a simplified semantic segmentation model is designed that only outputs three categories: grass area, parking space dividing line area and background area.

7. The method for improving the recognition accuracy of brick-grass parking spaces according to claim 1, characterized in that: The brick-turf parking space is a parking space with two straight-line boundaries. Grass grows on the brick-turf parking space to form a lawn, and the lawn partially blocks the two straight-line boundaries.

8. The method for improving the recognition accuracy of brick-grass parking spaces according to claim 1, characterized in that: The visual semantic segmentation model includes SegNet, U-Net, and DeepLab.

9. The method for improving the recognition accuracy of brick-grass parking spaces according to claim 1, characterized in that: The parking space recognition algorithms include Mobilenetv2, repvgg, and resnet.

10. A brick-grass parking space recognition accuracy improvement system, characterized in that: include: An acquisition module, used to acquire a fisheye stitching image of a brick-grass parking space; A segmentation module, used for inputting the fisheye stitching image into a trained visual semantic segmentation model to obtain a semantic segmentation result map; A screening module, used for screening out the parking space dividing line area blocked by the grass according to the semantic segmentation result map; A repair module, used to repair and restore the parking space dividing line area blocked by the grass to obtain a repaired fisheye stitching image of the brick grass parking space; The recognition module is used to recognize the parking space according to the restored fisheye stitching image of the brick-grass parking space.

Images