A semi-automatic method for extracting and semantically recognizing road marking point clouds

Abstract

This invention discloses a semi-automatic extraction and semantic recognition method for road marking point clouds, relating to the field of laser point cloud technology. It employs a Taylor interpolation polynomial model for point clouds, using distance-based intensity correction to ensure that the surface reflection intensity of the same material tends to be similar. The method extracts road traffic marking point clouds based on a dynamic threshold method, separating road markings from the road. Based on the minimum bounding rectangle and shape, it clusters marking point clouds of the same category, calculating the minimum bounding rectangle of each point cloud cluster. Semantic recognition is then performed on the point cloud clusters based on prior knowledge, marking the marking object information. This method can identify markings from point cloud data and classify them into six different marking types. The identified road marking information can provide a basis and support for traffic safety management and provide services for assisted driving and autonomous driving.

Description

Technical Field

[0001] This invention relates to the field of laser point cloud technology, and in particular to a semi-automatic method for extracting and semantically recognizing road marking point clouds. Background Technology

[0002] In recent years, vehicle-mounted LiDAR scanning technology has been widely used in many fields such as urban modeling, high-definition maps, and autonomous driving. The laser scanning data contains extremely rich information, including the coordinates, intensity information, echo information, and color information of discrete three-dimensional points on the target surface. Based on the above information, a variety of meaningful tasks can be carried out, such as road element segmentation, target extraction, and recognition. However, the processing technology of road point cloud data at home and abroad is still immature, and the existing algorithms and theories have certain limitations.

[0003] Road markings not only play a vital role in traffic guidance and control, but are also an important component of high-precision maps. The digitization of road markings can effectively improve road network efficiency, ensure traffic safety, and provide strong support for the development of autonomous driving. They are essential foundational data for many future-oriented applications such as road infrastructure digitization, traffic safety management, and autonomous driving. In the road and surrounding environment information acquired by the vehicle-mounted LiDAR scanning system, because the three-dimensional information such as the distance and density between the road and the markings are similar, the markings cannot be extracted from the road point cloud using the three-dimensional data of the point cloud. Summary of the Invention

[0004] To address the above technical problems, this invention provides a semi-automatic method for extracting and semantically recognizing road marking point clouds, comprising the following steps:

[0005] S1. Input road point cloud;

[0006] S2. A distance-based intensity correction is performed on the point cloud data using a Taylor interpolation polynomial model.

[0007] S3. Extracting road traffic marking point cloud based on dynamic thresholding method: First, the road point cloud is rasterized to generate an intensity feature image of the point cloud; then, the threshold is calculated based on the differences between each part in the point cloud after the region is divided; for the mixed area of ​​road and marking, the OTSU algorithm is used to segment the road markings from the road, thereby obtaining the road traffic marking point cloud.

[0008] S4. Based on Euclidean distance, the gradation point cloud is divided into different point clusters according to the distance between points, and the shape of different point clusters is determined by the minimum bounding rectangle method.

[0009] S5. Based on prior knowledge, classify road marking segments according to their geometric features, perform semantic recognition on point cloud clusters, and mark marking object information.

[0010] The technical solution further defined in this invention is:

[0011] Furthermore, in step S2, a Taylor interpolation polynomial model is used to perform distance-based intensity correction on the point cloud data.

[0012]

[0013] in, The coefficients of the Taylor polynomial are represented by n, where n represents the highest order. Let f(R) denote the i-th derivative of f(R).

[0014] The aforementioned semi-automatic extraction and semantic recognition method for road marking point clouds, using Taylor polynomial coefficients The method of obtaining includes the following steps

[0015] S2.1 Manually extract the target point cloud dataset containing only one material to obtain a point cloud dataset with a data size of n+1;

[0016] S2.2. Combining the vehicle trajectory data retained by the laser scanner during the driving process with the scanner height, the distance from each point in the point cloud dataset to the scanner is obtained;

[0017] S2.3. Obtain the coefficients of the Taylor polynomial by fitting the distances from each point in the point cloud dataset to the scanner. .

[0018] In the aforementioned semi-automatic extraction and semantic recognition method for road marking point clouds, step S2 employs an improved ratio method for intensity correction. For all points, the corrected intensity value is calculated using the ratio method described below.

[0019]

[0020] in, This represents the corrected strength value. Indicates the original strength value. This indicates the location of the extreme points of the fitted function. This indicates the threshold value selected for the intensity correction model;

[0021] If the ratio Greater than the threshold If the strength value at that point is overcorrected, an improved ratio method will be used to correct the strength at that point.

[0022] The aforementioned semi-automatic extraction and semantic recognition method for road marking point clouds includes the following steps in step S3: Extracting the road traffic marking point cloud based on a dynamic threshold method.

[0023] S3.1 Input the road point cloud to be segmented, and use the virtual grid method to divide the point cloud. Set the virtual grid size. Create An empty cell array is used to place the original point cloud into the corresponding cell array according to its two-dimensional coordinates. Any point... The number of rows and columns of the virtual grid is as follows:

[0024]

[0025] Where w and l represent the row and column numbers of the virtual grid where the point is located. , Represents the minimum coordinate in the point set, and [.] indicates rounding;

[0026] S3.2 Within each grid cell, the maximum inter-class variance of the intensity values ​​in the point cloud is calculated using the OTSU algorithm principle;

[0027] S3.3. Extract road markings according to different area types. If it is a full road area, discard it; if it is a full road marking area, retain it; if it is a mixed road and road marking area, then use the OTSU algorithm to segment it.

[0028] S3.4 Summarize the indices of all grid points identified as datum point clouds to obtain the datum point clouds in the point cloud dataset.

[0029] The aforementioned semi-automatic extraction and semantic recognition method for road marking point clouds, in step S3.2, calculates the maximum inter-class variance of intensity values ​​in the point cloud using the OTSU algorithm principle, including the following steps.

[0030] S3.2.1. Traverse all point cloud data and perform intensity histogram statistics. Let the total number of point clouds be N, and divide them into M intensity levels, with the number of point clouds at each intensity level being... The probability of the i-th intensity level point cloud appearing is calculated using the following formula.

[0031]

[0032] Let the segmentation threshold be The proportions of the target area and the background area to the whole are as follows:

[0033]

[0034] The average intensity of the target area and the background area is as follows:

[0035]

[0036] The formula for calculating inter-class variance is:

[0037]

[0038] Where g represents the inter-class variance, This indicates the proportion of pixels contained in the target region out of the total number of pixels. Indicates the average gray level. This indicates the proportion of pixels in the background area out of the total number of pixels. Indicates the average gray level;

[0039] S3.2.2 Calculate the segmentation threshold according to the formula for calculating inter-class variance. The inter-class variance;

[0040] S3.2.3. Traverse all point clouds to find the maximum inter-class variance and its corresponding segmentation threshold;

[0041] S3.2.4. Based on the maximum inter-class variance and the segmentation threshold, the grid is divided into the full road area, the full road marking area, and the mixed road marking area.

[0042] The semi-automatic extraction and semantic recognition method for road marking point clouds described above, in step S4, the method for finding the minimum bounding rectangle includes the following steps.

[0043] S4.1 Cluster the extracted line point cloud according to Euclidean distance, and use twice the average distance between points as the threshold, that is, divide point pairs with a distance less than this into the same point cloud cluster;

[0044] S4.2 Mark point clouds with fewer than 20 points in a point cloud cluster as noise and delete them;

[0045] S4.3. Traverse all point cloud clusters and project the point cloud clusters onto a two-dimensional plane. Calculate the planar convex hull of the midpoints of a point cloud cluster;

[0046] S4.4. Use the idea of ​​rotating calipers to find the minimum bounding rectangle;

[0047] S4.5 Repeat steps S4.3 to S4.4 until the minimum bounding rectangle of all point cloud clusters has been found.

[0048] In the aforementioned semi-automatic extraction and semantic recognition method for road marking point clouds, step S4.4 involves finding the minimum bounding rectangle as follows: First, connect any two points on the convex hull of the plane as one side of the rectangle. Then, use this side as the x-axis base coordinate and generate the y-axis base coordinate accordingly. Next, rotate all points using this base coordinate to find the minimum and maximum x-coordinates of all points with this side as the reference, i.e., the minimum and maximum y-coordinates. Then, obtain the area value and boundary data of this range. Finally, enumerate each side and save the minimum area and parameters each time to find the rectangle with the smallest area and parameters.

[0049] The aforementioned semi-automatic extraction and semantic recognition method for road marking point clouds includes the following steps in step S5: determining different point cloud clusters based on semantic knowledge of shape.

[0050] S5.1. For a given point cloud cluster and its minimum bounding rectangle, determine the length L of the minimum bounding rectangle and the critical length. If the size relationship satisfies the following formula, the point cloud cluster corresponding to the smallest bounding rectangle enters the solid line decision branch; otherwise, it enters the dashed line and guide arrow decision branch.

[0051]

[0052]

[0053] in, Indicates the standard length of the dashed line. Indicates the standard length of the pointing arrow. Indicates the standard length of text;

[0054] S5.2 For a point cloud cluster entering the solid line decision branch, the width W of its minimum bounding rectangle is equal to the standard width of the solid line. If the following relationship is satisfied, then the point cloud cluster is the solid white line representing the edge of the roadway.

[0055]

[0056] in, Indicates the standard width of the solid line;

[0057] S5.3 For point cloud clusters entering the decision branches of dashed lines and guide arrows, determine the relationship between the width W of the minimum bounding rectangle and the standard width, and the length L of the minimum bounding rectangle and the standard length.

[0058] If the following formula is satisfied, then the point cloud cluster is the white dashed line representing the edge of the roadway.

[0059]

[0060] If the following formula is satisfied, then the point cloud cluster is capable of crossing lane dividing lines.

[0061]

[0062] If the following formula is satisfied, then the point cloud cluster is a guide arrow.

[0063]

[0064] If the following formula is satisfied, then the point cloud cluster is text.

[0065]

[0066] in, and This indicates the standard width and length of the white dashed line representing the edge of the roadway; and This indicates the standard width and length that can be crossed over lane dividers; and This indicates the standard width and length that can be crossed over lane dividers; and Standard width and length for road markings; coefficients , , , Indicates the tolerance factor;

[0067] If none of the above conditions are met, the point cloud cluster will be identified as other.

[0068] In the aforementioned semi-automatic extraction and semantic recognition method for road marking point clouds, step S5.3, , , as well as The values ​​are 0.2, 0.1, 0.1 and 0.2 respectively.

[0069] The beneficial effects of this invention are:

[0070] In this invention, for situations where the reflection intensity values ​​of the same target vary significantly, a Taylor interpolation polynomial model is used for distance-based intensity correction, making the surface reflection intensity of the same material tend to be the same, while the difference in reflection intensity between different materials becomes more obvious. Based on this, this invention uses an improved dynamic thresholding method to achieve road marking point cloud segmentation. In the point cloud after region segmentation, a threshold is calculated based on the differences of each part. For mixed areas of roads and markings, the OTSU algorithm is used for segmentation, thereby obtaining the road traffic marking point cloud. This effectively completes the extraction of markings from point clouds containing a lot of non-infrastructure noise (such as vehicles) and uneven intensity distribution, and is ready for subsequent road marking recognition, namely, identifying dashed markings, solid markings, and arrows. Based on the minimum bounding rectangle and shape of road marking semantic recognition, the road marking point cloud has a high fitting degree with the minimum bounding rectangle. The minimum bounding rectangle can effectively select each independent marking, and the minimum bounding rectangle features of different types of markings are very different, which can be used for subsequent road marking semantic recognition. The identified road marking information can provide a basis and support for traffic safety management and provide services for assisted driving and autonomous driving. Attached Figure Description

[0071] Figure 1 This is a schematic diagram of the overall process of the present invention;

[0072] Figure 2This is a diagram showing the point cloud extraction result of the marking line in an embodiment of the present invention;

[0073] Figure 3 This is the minimum bounding rectangle diagram of the Euclidean clustering of the punctuation point cloud in this embodiment of the invention. Detailed Implementation

[0074] This embodiment provides a semi-automatic method for extracting and semantically recognizing road marking point clouds, such as... Figure 1 As shown, it includes the following steps

[0075] S1. Input road point cloud;

[0076] S2. A Taylor interpolation polynomial model is used to perform distance-based intensity correction on point cloud data, so that the surface reflection intensity of the same material tends to be the same, and the difference in reflection intensity of different materials becomes more obvious.

[0077] A distance-based intensity correction is performed using a Taylor interpolation polynomial model, as shown in the following equation.

[0078]

[0079] in, The coefficients of the Taylor polynomial are represented by n, where n represents the highest order. Let f(R) denote the i-th derivative of f(R).

[0080] For Taylor polynomial coefficients To extract target point cloud datasets containing only one material, it is necessary to manually extract the target point cloud datasets containing only one material. In this embodiment, the ground without markings is regarded as a material with the same laser reflectivity, and the ground point cloud without road markings is manually extracted to obtain a point cloud dataset with a data volume of n+1.

[0081] Then, the distance from each point in the point cloud dataset to the scanner is calculated. The laser scanner retains vehicle trajectory data during its movement; combining this data with the scanner height yields the distance from each point to the scanner. Finally, the coefficients of the Taylor polynomial are obtained by fitting the above data. .

[0082] An improved ratio method is used for strength correction. For all points, the corrected strength value is calculated using the ratio method of the following formula:

[0083]

[0084] in, This represents the corrected strength value. Indicates the original strength value. This indicates the location of the extreme points of the fitted function. This indicates the threshold value selected for the intensity correction model;

[0085] If the ratio Greater than the threshold If the intensity value at a given point is overcorrected, an improved ratio method is used to correct the intensity at that point. This model can ensure the continuity of intensity correction near the intensity threshold point. Regarding the selection of the threshold, too large a threshold will lead to poor correction effect, while too small a threshold will result in undercorrection. In this embodiment, the threshold is selected based on the actual situation of the point cloud intensity values. That is, the corrected intensity value should not exceed half of the original point cloud intensity value.

[0086] S3. Extracting road traffic marking point cloud based on dynamic thresholding method: First, rasterize the road point cloud to generate an intensity feature image of the point cloud; calculate the threshold based on the differences of each part in the point cloud after dividing the region; for the mixed area of ​​road and marking, use the OTSU algorithm to segment and separate the road marking from the road, thereby obtaining the road traffic marking point cloud.

[0087] In step S3, the process of the road traffic marking extraction algorithm based on the dynamic threshold method is as follows:

[0088] S3.1 Input the road point cloud to be segmented, and use the virtual grid method to divide the point cloud. Set the virtual grid size. Create An empty cell array is used to place the original point cloud into the corresponding cell array according to its two-dimensional coordinates. Any point... The number of rows and columns of the virtual grid is as follows:

[0089]

[0090] Where w and l represent the row and column numbers of the virtual grid where the point is located. , [.] represents the minimum coordinate in the point set, and [.] represents rounding.

[0091] S3.2 Within each grid cell, the maximum inter-class variance of intensity values ​​in the point cloud is calculated using the OTSU algorithm, including the following steps.

[0092] S3.2.1. Traverse all point cloud data and perform intensity histogram statistics. Let the total number of point clouds be N, and divide them into M intensity levels, with the number of point clouds at each intensity level being... The probability of the i-th intensity level point cloud appearing is calculated using the following formula.

[0093]

[0094] Let the segmentation threshold be The proportions of the target area and the background area to the whole are as follows:

[0095]

[0096] The average intensity of the target area and the background area is as follows:

[0097]

[0098] The formula for calculating inter-class variance is:

[0099]

[0100] Where g represents the inter-class variance, This indicates the proportion of pixels contained in the target region out of the total number of pixels. Indicates the average gray level. This indicates the proportion of pixels in the background area out of the total number of pixels. Indicates the average gray level;

[0101] S3.2.2 Calculate the segmentation threshold according to the formula for calculating inter-class variance. The inter-class variance;

[0102] S3.2.3. Traverse all point clouds to find the maximum inter-class variance and its corresponding segmentation threshold;

[0103] S3.2.4. Based on the maximum inter-class variance and the segmentation threshold, the grid is divided into the full road area, the full road marking area, and the mixed road marking area.

[0104] S3.3. Extract road markings according to different area types. If it is a full road area, discard it; if it is a full road marking area, retain it; if it is a mixed road and road marking area, then use the OTSU algorithm for segmentation.

[0105] S3.4 Summarize the indices of all grid points identified as datum point clouds to obtain the datum point clouds in the point cloud dataset;

[0106] The effect of extracting the gradation point cloud using the dynamic threshold method in this embodiment is as follows: Figure 2 As shown, Figure 2 (a) is a schematic diagram of the point cloud extraction of the front road surface. Figure 2 Figure (b) shows the extracted road marking point cloud. As can be seen from the figure, almost all non-road marking point clouds in the point cloud dataset are filtered out. The extracted road marking point cloud has clear boundaries and complete shape. Therefore, the results show that the improved dynamic thresholding method used in this embodiment can effectively complete the extraction of road markings from point clouds containing a lot of non-infrastructure noise (such as vehicles) and uneven intensity distribution. It has the conditions to carry out subsequent road marking, namely, to identify dashed lines, solid lines and arrows in the road markings.

[0107] S4. Based on Euclidean distance, the gradation point cloud is divided into different point clusters according to the distance between points, and the shape of different point clusters is determined by the minimum bounding rectangle method.

[0108] After processing using the dynamic thresholding method described in step S3, the road marking point cloud is extracted from the road. However, it consists of discrete point clouds without object meaning. Therefore, in this embodiment, the road marking point cloud is first divided into different point clusters based on Euclidean distance, and then the shape of different point clusters is determined. The minimum bounding rectangle method is used, and the specific process for obtaining the minimum bounding rectangle is as follows:

[0109] S4.1 Cluster the extracted line point cloud according to Euclidean distance, and use twice the average distance between points as the threshold, that is, divide point pairs with a distance less than this into the same point cloud cluster;

[0110] S4.2 Mark point clouds with fewer than 20 points in a point cloud cluster as noise and delete them;

[0111] S4.3, Traverse all point cloud clusters, using point cloud clusters as the basis for... For example, Projected onto a two-dimensional plane ,calculate The convex hull of the plane at the midpoint;

[0112] S4.4. Find the minimum bounding rectangle using the idea of ​​rotating calipers: First, connect any two points on the convex hull of the plane as one side of the rectangle. Then, use this side as the x-axis base coordinate and generate the y-axis base coordinate accordingly. Next, rotate all points using this base coordinate to find the minimum and maximum x-coordinates of all points with this side as the reference, which are the minimum and maximum y-coordinates. Then, obtain the area value and boundary data of this range. Finally, enumerate each side and save the minimum area and parameters each time to find the rectangle with the smallest area and parameters.

[0113] S4.5 Repeat steps S4.3 to S4.4 until the minimum bounding rectangle of all point cloud clusters has been found.

[0114] The result of obtaining the minimum bounding rectangle of the point cloud cluster based on the above process is as follows: Figure 3 As shown, Figure 3 (a) in the image represents the extracted road markings. Figure 3 (b) in the figure represents the minimum bounding rectangle corresponding to the point cloud of the road marking. As can be seen from the figure, the road marking point cloud has a high degree of fit with the minimum bounding rectangle. The minimum bounding rectangle can effectively select each independent marking. Moreover, the minimum bounding rectangle features of different types of markings are very different, which can be used for subsequent semantic recognition of road markings.

[0115] Based on the width W and length L of each minimum bounding rectangle, and combined with the characteristic differences of dashed lines, solid lines and arrows in the road markings, semantic recognition is performed on different road markings.

[0116] S5. Based on prior knowledge, classify road marking segments according to their geometric features, perform semantic recognition on point cloud clusters, and mark marking object information.

[0117] The method for determining different point cloud clusters based on shape semantic knowledge is as follows:

[0118] S5.1. For a given point cloud cluster and its minimum bounding rectangle, determine the length L of the minimum bounding rectangle and the critical length. If the size relationship satisfies the following formula, the point cloud cluster corresponding to the smallest bounding rectangle enters the solid line decision branch; otherwise, it enters the dashed line and guide arrow decision branch.

[0119]

[0120]

[0121] in, This indicates the standard length of the dashed line, which includes two categories in the specification: 200cm and 600cm. This indicates the standard length of the pointing arrow; the specification includes three categories: 650cm, 585cm, and 600cm. The standard length of text is indicated by three categories: 900cm, 600cm, and 300cm. Therefore, if the design speed of this road section is 100km / h or higher, Take 900cm, otherwise, Take 600cm;

[0122] S5.2 For a point cloud cluster entering the solid line decision branch, the width W of its minimum bounding rectangle is equal to the standard width of the solid line. If the following relationship is satisfied, then the point cloud cluster is the solid white line representing the edge of the roadway.

[0123]

[0124] in, This indicates the standard width of the solid line, which includes two categories in the specification: 15cm and 20cm.

[0125] S5.3 For point cloud clusters entering the decision branches of dashed lines and guide arrows, determine the relationship between the width W of the minimum bounding rectangle and the standard width, and the length L of the minimum bounding rectangle and the standard length.

[0126] If the following formula is satisfied, then the point cloud cluster is the white dashed line representing the edge of the roadway.

[0127]

[0128] If the following formula is satisfied, then the point cloud cluster is capable of crossing lane dividing lines.

[0129]

[0130] If the following formula is satisfied, then the point cloud cluster is a guide arrow.

[0131]

[0132] If the following formula is satisfied, then the point cloud cluster is text.

[0133]

[0134] in, and The standard width and length of the white dashed line indicating the edge of the roadway are specified in the standard, which includes two types: 15cm×400cm and 20cm×400cm. and This indicates the standard width and length that can be crossed over lane dividing lines. The specifications include four categories: 10cm×600cm, 15cm×600cm, 10cm×200cm, and 15cm×200cm. and The standard width and length that can be crossed for lane dividing lines are indicated. The regulations stipulate that the straight left or right arrow is 185cm×650cm, the left turn or right turn arrow is 133cm×585cm, the straight arrow is 90cm×600cm and 15cm×200cm, and the U-turn arrow is 133cm×585cm. and The standard width and length for road signage are specified in the regulations, with characters measuring 100cm x 240cm; coefficients... , , , This represents the tolerance factor, which is closely related to the density of the point cloud cluster and the wear and tear of the point cloud. , , , The values ​​are 0.2, 0.1, 0.1, and 0.2 respectively.

[0135] If none of the above conditions are met, the point cloud cluster will be identified as other.

[0136] In addition to the embodiments described above, the present invention may have other implementations. All technical solutions formed by equivalent substitution or equivalent transformation fall within the protection scope claimed by the present invention.

Claims

1. A semi-automatic method for extracting and semantically recognizing road marking point clouds, characterized in that: Includes the following steps S1. Input road point cloud; S2. A distance-based intensity correction is performed on the point cloud data using a Taylor interpolation polynomial model. S3. Extracting road traffic marking point cloud based on dynamic thresholding method: First, the road point cloud is rasterized to generate an intensity feature image of the point cloud; then, the threshold is calculated based on the differences between each part in the point cloud after the region is divided; for the mixed area of ​​road and marking, the OTSU algorithm is used to segment the road markings from the road, thereby obtaining the road traffic marking point cloud. S4. Based on Euclidean distance, the gradation point cloud is divided into different point clusters according to the distance between points, and the shape of different point clusters is determined by the minimum bounding rectangle method. S5. Based on prior knowledge, classify road marking segments according to their geometric features, perform semantic recognition on point cloud clusters, and label marking object information; determine different point cloud clusters based on shape semantic knowledge, including the following steps. S5.

1. For a given point cloud cluster and its minimum bounding rectangle, determine the length of the minimum bounding rectangle. With critical length If the size relationship satisfies the following formula, the point cloud cluster corresponding to the smallest bounding rectangle enters the solid line decision branch; otherwise, it enters the dashed line and guide arrow decision branch. in, Indicates the standard length of the dashed line. Indicates the standard length of the pointing arrow. Indicates the standard length of text; S5.2 For a point cloud cluster that enters the solid line decision branch, the width of its minimum bounding rectangle is... Standard width of solid line If the following relationship is satisfied, then the point cloud cluster is the solid white line representing the edge of the roadway. in, Indicates the standard width of the solid line; S5.3 For point cloud clusters entering the decision branches of dashed lines and guide arrows, determine the width of the minimum bounding rectangle. With standard width and length Relationship with standard length If the following formula is satisfied, then the point cloud cluster is the white dashed line representing the edge of the roadway. If the following formula is satisfied, then the point cloud cluster is capable of crossing lane dividing lines. If the following formula is satisfied, then the point cloud cluster is a guide arrow. If the following formula is satisfied, then the point cloud cluster is text. in, and This indicates the standard width and length of the white dashed line representing the edge of the roadway; and This indicates the standard width and length that can be crossed over lane dividers; and This indicates the standard width and length that can be crossed over lane dividers; and Standard width and length for road markings; coefficients , , , Indicates the tolerance factor; If none of the above conditions are met, the point cloud cluster will be identified as other.

2. The semi-automatic extraction and semantic recognition method for road marking point clouds according to claim 1, characterized in that: In step S2, a Taylor interpolation polynomial model is used to perform distance-based intensity correction on the point cloud data. in, Denotes the Taylor polynomial coefficients. Indicates the highest order. express of The first derivative.

3. The semi-automatic extraction and semantic recognition method for road marking point clouds according to claim 2, characterized in that: Taylor polynomial coefficients The method of obtaining includes the following steps S2.1 Manually extract the target point cloud dataset containing only one material, resulting in a data volume of [data volume not specified]. Point cloud dataset; S2.

2. Combining the vehicle trajectory data retained by the laser scanner during the driving process with the scanner height, the distance from each point in the point cloud dataset to the scanner is obtained; S2.

3. Obtain the coefficients of the Taylor polynomial by fitting the distances from each point in the point cloud dataset to the scanner. .

4. The semi-automatic extraction and semantic recognition method for road marking point clouds according to claim 2, characterized in that: In step S2, an improved ratio method is used for strength correction. For all points, the corrected strength value is calculated using the ratio method of the following formula. in, This represents the corrected strength value. Indicates the original strength value. This indicates the location of the extreme points of the fitted function. This indicates the threshold value selected for the intensity correction model; If the ratio Greater than the threshold If the strength value at that point is overcorrected, an improved ratio method will be used to correct the strength at that point.

5. The semi-automatic extraction and semantic recognition method for road marking point clouds according to claim 1, characterized in that: In step S3, the extraction of road traffic marking point cloud based on the dynamic threshold method includes the following steps: S3.1 Input the road point cloud to be segmented, and use the virtual grid method to divide the point cloud. Set the virtual grid size. , create An empty cell array is used to place the original point cloud into the corresponding cell array according to its two-dimensional coordinates. Any point... The number of rows and columns of the virtual grid is as follows: in, , This indicates the row and column number of the virtual grid where the point is located. , Represents the minimum coordinates in the point set. Indicates rounding down; S3.2 Within each grid cell, the maximum inter-class variance of the intensity values ​​in the point cloud is calculated using the OTSU algorithm principle; S3.

3. Extract road markings according to different area types. If it is a full road area, discard it; if it is a full road marking area, retain it; if it is a mixed road and road marking area, then use the OTSU algorithm to segment it. S3.4 Summarize the indexes of all grid points identified as line point clouds.

6. The semi-automatic extraction and semantic recognition method for road marking point clouds according to claim 5, characterized in that: In step S3.2, calculating the maximum inter-class variance of intensity values ​​in the point cloud using the OTSU algorithm includes the following steps: S3.2.

1. Traverse all point cloud data and perform intensity histogram statistics. Let the total number of point clouds be... According to strength, they are divided into There are 1 intensity level, and the number of point clouds at each intensity level is 1. The probability of the i-th intensity level point cloud appearing is calculated using the following formula. Let the segmentation threshold be The proportions of the target area and the background area to the whole are: The average intensity of the target area and the background area is as follows: The formula for calculating inter-class variance is: in, Represents the variance between classes. This indicates the proportion of pixels contained in the target region out of the total number of pixels. Indicates the average gray level. This indicates the proportion of pixels in the background area out of the total number of pixels. Indicates the average gray level; S3.2.2 Calculate the segmentation threshold according to the formula for calculating inter-class variance. The inter-class variance; S3.2.

3. Traverse all point clouds to find the maximum inter-class variance and its corresponding segmentation threshold; S3.2.

4. Based on the maximum inter-class variance and the segmentation threshold, the grid is divided into the full road area, the full road marking area, and the mixed road marking area.

7. The semi-automatic extraction and semantic recognition method for road marking point clouds according to claim 1, characterized in that: In step S4, the method for minimizing the bounding rectangle includes the following steps: S4.1 Cluster the extracted line point cloud according to Euclidean distance, and use twice the average distance between points as the threshold, that is, divide point pairs with a distance less than this into the same point cloud cluster; S4.2 Mark point clouds with fewer than 20 points in a point cloud cluster as noise and delete them; S4.

3. Traverse all point cloud clusters and project the point cloud clusters onto a two-dimensional plane. Calculate the planar convex hull of the midpoints of a point cloud cluster; S4.

4. Use the idea of ​​rotating calipers to find the minimum bounding rectangle; S4.5 Repeat steps S4.3 to S4.4 until the minimum bounding rectangle of all point cloud clusters has been found.

8. The semi-automatic extraction and semantic recognition method for road marking point clouds according to claim 7, characterized in that: In step S4.4, the method for finding the minimum bounding rectangle is as follows: First, connect any two points on the convex hull of the plane as one side of the rectangle. Then, use this side as the x-axis base coordinate and generate the y-axis base coordinate accordingly. Next, rotate all points with this base coordinate to find the minimum and maximum x-coordinates of all points with this side as the reference, that is, the minimum and maximum y-coordinates. Then, obtain the area value and its boundary data when the x-coordinates are within the range of minimum to maximum values. Finally, enumerate each side and save the minimum area and parameters each time to find the rectangle with the smallest area and parameters.

9. A semi-automatic method for extracting and semantically recognizing road marking point clouds according to claim 1, characterized in that: In step S5.3, , , as well as The values ​​are 0.2, 0.1, 0.1 and 0.2 respectively.

Images