Road Point Cloud Extraction Method Based on Scan Lines and Sliding Least Squares Fitting

Abstract

The present invention discloses a method for extracting road point clouds based on scan lines and sliding least squares fitting, comprising the following steps: scan line reconstruction; accurate estimation of the height of the POS system from the road surface; robust extraction of road surface seed points; complete and robust extraction of road surface points; an efficient method for extracting road surface points based on scan lines. The present invention extracts road surface point clouds scan line by scan line, improving the efficiency of road surface point extraction; making full use of prior knowledge, laying a foundation for the robust extraction of road surface points; the robust extraction of road surface seed points, clustering the seed point candidates of the critical rows of the alternative points in the scan lines, effectively improving the robustness of the road surface seed point extraction; using sliding weighted least squares to fit local road points, determining the fitting method and the weight determination method, which can effectively achieve the local optimal fitting of road surface points in the scan lines and achieve the robust and complete extraction of road surface points.

Description

Technical Field

[0001] The present invention relates to the technical field of remote sensing mapping, and specifically to a method for extracting road point clouds based on scan lines and sliding least squares fitting. Background Art

[0002] Vehicle-mounted / UAV laser scanning systems can efficiently obtain the three-dimensional spatial information and attribute information of ground objects during the movement of the platform, and are a means of obtaining three-dimensional spatial data that has developed rapidly and is used more and more widely in recent years. The vehicle-mounted / UAV laser point cloud contains undifferentiated sampling of various ground objects such as the ground and vegetation. Extracting the required specific ground object information from the vehicle-mounted / UAV laser point cloud is a necessary processing step for the efficient application of vehicle-mounted / UAV laser point clouds in various industries.

[0003] Road points are an important part of the vehicle-mounted / UAV laser point cloud and are a reliable data source for applications such as road condition surveys, road geometric parameter extraction, and road surface obstacle detection and recognition. In applications such as road condition surveys, road geometric parameter extraction, and road surface obstacle detection and recognition, in order to improve the efficiency of data processing and the robustness and reliability of algorithms, road surface point clouds are usually first extracted from the vehicle-mounted / UAV laser point cloud.

[0004] Due to the complex and diverse road surface conditions, as well as the discrete sampling, occlusion, and noise of the laser point cloud, extracting road surface point clouds from the vehicle-mounted / UAV laser point cloud is challenging. Yang et al. projected the vehicle-mounted laser point cloud into a feature image, determined the segmentation threshold using the discrete discriminant analysis method, and segmented the part belonging to the road surface from the feature image. This method requires gridification when generating the feature image, and the boundary accuracy of the extracted road surface point cloud is limited. Kumar et al. extracted the boundary line of the road by combining two improved parametric active contour models, and obtained the road surface point cloud through the extracted boundary line. This method has a large amount of calculation and low efficiency. Guan et al. divided the vehicle-mounted laser point cloud into blocks, constructed virtual scan lines for each block of laser point cloud, and detected the road curb through elevation changes in each virtual scan line to obtain road surface points. This method is only suitable for roads with obvious road curbs. Riveiro et al. analyzed each scan line and obtained the elevation peak points indicating the start and end of the road surface in each scan line through PCA (Principal Component Analysis). This method is also only applicable to roads with obvious road curbs. Holgado-Barco et al. aimed at a specific type of road, set the scan angle, and used the points within a specific scan angle range as road surface points. This method requires manual adjustment of the threshold for different road types and is only suitable for roads with a constant road surface width.

[0005] In summary, although many scholars at home and abroad have carried out research on the extraction of road point clouds from vehicle-mounted / drone point clouds and made some progress, they all have their own limitations and lack a solution that is applicable to different road types and is efficient and stable. Summary of the Invention

[0006] The purpose of the present invention is to provide a method for extracting road point clouds by sliding least squares fitting to solve the problems raised in the above background technology.

[0007] To achieve the above purpose, the present invention provides the following technical solution: A kind of,...

[0008] A method for extracting road point clouds based on scan lines and sliding least squares fitting, comprising the following steps:

[0009] S1. Scan line reconstruction: According to the scan angle of each laser point in the scanner coordinate system, the original laser point cloud is reconstructed into different scan lines;

[0010] S2. Accurate estimation of the height of the POS system from the road surface: There is a fixed height difference HD between the POS system and the ground, and then the accurate height is obtained through calculation;

[0011] S3. Robust extraction of road surface seed points: Based on the calculation of the POS data in S2, and then the extraction of seed points is carried out;

[0012] S4. Complete and robust extraction of road points: Based on the seed points obtained in S3, fitting growth is carried out, and then road points are obtained.

[0013] Preferably, the specific operation method in step S1 is:

[0014] Based on the vehicle-mounted laser point cloud or the drone laser point cloud being continuously stored in time sequence, the scan angle of each point in the point cloud in the scanner coordinate system can be obtained. When the scan angles of two consecutive points are converted from 360° to 0°, it means the end of the old scan line and the start of the new scan line. Based on this rule, the above laser point cloud can be divided into scan lines one by one, thus realizing the reconstruction of scan lines.

[0015] Preferably, the specific operation method in step S2 is:

[0016] For a record (t0, x0, y0, z0) in the POS data, first obtain the point set within the time range [t0 - δt, t0 + δt] from the original data where δt is a relatively small time threshold. Then obtain the point set from the point set in which the distance (x0, y0) is less than d0 Then the height of the POS system from the road surface can be stably estimated by the following formula:

[0017]

[0018] Wherein, N is the point set the number of points in, z i is the elevation of the i-th point, d hd is the height of the estimated POS system from the ground.

[0019] Preferably, the specific operation method of step S3 is as follows:

[0020] For each point P t (t, x t , y t , z t ) in each scan line, the POS data at the corresponding time t can be interpolated from the original POS data, and O t (t, x, y, z) represents the POS data at time t. Since the POS system is rigidly fixed on the roof platform, although the road surface is not completely flat and the height difference between the POS system and the road surface is not a fixed value, the height difference between the POS system and the road surface is relatively fixed. The height difference between P t and O t and the d hd obtained in 4.2 can be used to obtain some ground points as seed points;

[0021] Based on this, the points satisfying Equation (2) can be used as alternative points for road surface seed points:

[0022] |z t - z - d hd | < ρ hd (2)

[0023] In addition to road surface points, the points satisfying Equation (2) may also include some non-road surface points such as vegetation points outside the road. In order to stably obtain road surface seed points, the alternative seed points obtained through Equation (2) are clustered according to the adjacency of the points in the scan line, and the class with the largest number of points is selected as the final road surface seed point.

[0024] Preferably, the specific operation method of step S4 is as follows:

[0025] Since the road surface is not a strictly flat plane, it is impossible to fit the road surface point cloud with a straight line in the scan line. However, for a local part of the road surface, a straight line can be used to fit in the scan line. For the straight line fitted to the local road surface points, it is possible to judge whether the points in the local range belong to the road surface points by the distance from the points to the fitted straight line;

[0026] To achieve the best fit of the fitted line to the local road surface, based on the existing road surface points, sliding weighted least squares is used for road surface fitting. In sliding weighted least squares fitting, the influence of a point on the fitted line depends on the weight of the point. If the weight is very small, the influence of the point on the fitted line is very small. If the weight is very large, the influence of the point on the fitted line is very large. For points far from the current point, smaller weights are assigned, and for points closer to the current point, larger weights are assigned. In this way, the locally best fit of the fitted line can be achieved;

[0027] To achieve the local road surface line fitting based on sliding weighted least squares, the points on each scan line are transformed into the scan line coordinate system. The scan line coordinate system takes the midpoint of the road surface point seeds extracted in S3 as the coordinate origin, takes the direction along the scan line as the X-axis, and takes the vertical upward direction perpendicular to the X-axis as the Y-axis. After transformation into the scan line coordinate system, the coordinates of the road surface seed points extracted in S3 in the scan line coordinate system are represented by r i (x i ,y i ). The next point to be determined whether it is a road surface point is p, and the point closest to p is r1. Then, the line that best fits the local road surface at p can be obtained by minimizing Equation (4):

[0028] l(x) = ax + b (3)

[0029]

[0030] In Equation (4), x i 、y i are the coordinates of the current road surface seed point in the scan line coordinate system, w i is the weight of the point r i (x i ,y i ). The weight w i is obtained through the exponential weight function:

[0031]

[0032] In Equation (6), d i is the distance from the point r i to the point r1. The weight calculated according to Equation (5) can ensure that the weight of the point r1 closest to p is 1, while the weights of other points are less than 1, and the farther the point is from p, the smaller its weight. h is a parameter that controls the influence of points within a certain distance range on the fitted line. Since the value of e -5 is 0.0067, this weight is very small and has a negligible influence on the fitted line. Therefore, the relationship between the maximum effective distance D and h is:

[0033]

[0034] Compared with the prior art, the beneficial effects of the present invention are as follows: The method for extracting road point clouds by sliding least squares fitting has the following advantages:

[0035] 1) An efficient method for extracting road surface points based on scan lines. The present invention extracts road surface point clouds line by line, effectively reducing the amount of point cloud data processed simultaneously and improving the efficiency of road surface point extraction;

[0036] 2) Make full use of prior knowledge and automatically estimate this prior knowledge based on data, laying a foundation for the robust extraction of road surface points;

[0037] 3) Robust extraction of road surface seed points. Cluster the candidate seed points of the critical rows of the candidate points in the scan line, and select the class with the largest number of points as the final road surface seed points, effectively improving the robustness of road surface seed point extraction;

[0038] 4) A method for extracting complete road surface point clouds based on sliding weighted least squares fitting. Use sliding weighted least squares to fit local road points, determine the fitting method and the weight determination method. This method can effectively achieve the local optimal fitting of road surface points in the scan line and realize the robust extraction of complete road surface points. BRIEF DESCRIPTION OF THE DRAWINGS

[0039] Figure 1 is a flow chart of the present invention;

[0040] Figure 2 is a point cloud diagram rendered by scan line after scan line reconstruction of the present invention;

[0041] Figure 3 is a schematic diagram of the fixed height difference between the POS system and the road surface of the present invention;

[0042] Figure 4 is a local schematic diagram of using a straight line to fit road points in the scan line of the present invention;

[0043] Figure 5 is a scan line coordinate system diagram of the present invention. DETAILED DESCRIPTION OF THE INVENTION

[0044] Next, the technical solutions in the embodiments of the present invention will be clearly and completely described in conjunction with the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative efforts shall fall within the protection scope of the present invention.

[0045] Please refer to Figures 1-5, the present invention provides a technical solution: a method for extracting road point cloud based on scan lines and sliding least squares fitting, comprising the following steps:

[0046] S1. Scan line reconstruction: According to the scan angle of each laser point in the scanner coordinate system, the original laser point cloud is reconstructed into different scan lines;

[0047] S2. Accurate estimation of the height of the POS system from the road surface: There is a fixed height difference HD between the POS system and the ground, and then the accurate height is obtained through calculation;

[0048] S3. Robust extraction of road surface seed points: Based on the calculation of the POS data in S2, and then the extraction of seed points is carried out;

[0049] S4. Complete and robust extraction of road points: Based on the seed points obtained in S3, fitting growth is carried out, and then road points are obtained.

[0050] Specifically, the specific operation method in step S1 is:

[0051] Based on the on-vehicle laser point cloud or the UAV laser point cloud stored continuously in time sequence, the scan angle of each point in the point cloud can be obtained in the scanner coordinate system. When the scan angles of two consecutive points are converted from 360° to 0°, it means the end of the old scan line and the start of a new scan line. Based on this rule, the above laser point cloud can be divided into scan lines one by one, so as to realize the reconstruction of scan lines, as Figure 2 shown.

[0052] Specifically, the specific operation method in step S2 is:

[0053] As Figure 3 shown, for a record (t0, x0, y0, z0) in the POS data, first obtain the point set within the time range [t0 - δt, t0 + δt] from the original data where δt is a relatively small time threshold. Then obtain the point set with a distance (x0, y0) less than d0 from the point set Then the height of the POS system from the road surface can be stably estimated by the following formula:

[0054]

[0055] In the formula, N is the number of points in the point set z i is the elevation of the i-th point, and d hd is the estimated height of the POS system from the ground.

[0056] Specifically, the specific operation method in step S3 is:

[0057] For each point P in each scan line t (t, x t , y t , z t ), the corresponding POS data at time t can be interpolated from the original POS data. Let O t (t, x, y, z) represent the POS data at time t. Since the POS system is rigidly fixed on the roof platform, although the road surface is not exactly a plane and the height difference between the POS system and the road surface is not a constant value, the height difference between the POS system and the road surface is relatively fixed. The height difference between P t and O t and the size relationship with d hd obtained in 4.2 can be used to obtain some ground points as seed points;

[0058] Based on this, the points satisfying Equation (2) can be used as alternative points for road surface seed points:

[0059] |z t - z - d hd | < ρ hd (2)

[0060] The points satisfying Equation (2) may include not only road surface points but also some non-road surface points such as vegetation points outside the road. To robustly obtain road surface seed points, the alternative points obtained through Equation (2) are clustered according to the adjacency of the points in the scan line, and the class with the largest number of points is selected as the final road surface seed points.

[0061] Specifically, the specific operation method of step S4 is as follows:

[0062] Since the road surface is not a plane in the strict sense, it is impossible to fit the road surface point cloud with a straight line in the scan line. However, for a local part of the road surface, a straight line can be used for fitting in the scan line. As Figure 4 stated, for the straight line fitted to the local road surface points, it is possible to determine whether the points within the local range belong to the road surface points by the distance from the points to the fitted straight line;

[0063] To achieve the best fit of the fitted straight line to the local road surface, based on the existing road surface points, sliding weighted least squares is used for road surface fitting. In sliding weighted least squares fitting, the influence of a point on the fitted straight line depends on the weight of the point. If the weight is very small, the influence of the point on the fitted straight line is very small. If the weight is very large, the influence of the point on the fitted straight line is very large. A smaller weight is assigned to the points far from the current point, and a larger weight is assigned to the points closer to the current point. In this way, the fitted straight line can be locally optimally fitted;

[0064] To achieve local road surface straight line fitting based on sliding weighted least squares, the points of each scan line are transformed into the scan line coordinate system. The origin of the scan line coordinate system is the midpoint of the road surface point seeds extracted in S3, the X-axis is along the scan line direction, and the Y-axis is vertically upward perpendicular to the X-axis. As Figure 5 shown, after being transformed into the scan line coordinate system, the coordinates of the road surface seed points extracted in S3 in the scan line coordinate system are represented by r i (x i , y i ). The next point to be determined whether it is a road surface point is p, and the point closest to p is r1. Then, the straight line that best fits the local road surface where p is located can be obtained by minimizing Equation (4):

[0065] l(x) = ax + b (3)

[0066]

[0067] In Equation (4), x i , y i are the coordinates of the current road surface seed point in the scan line coordinate system, w i is the weight of the point r i (x i , y i ). The weight w i is obtained through the exponential weight function:

[0068]

[0069] In Equation (6), d i is the distance from the point r i to the point r1. The weight calculated according to Equation (5) can ensure that the weight of the point r1 closest to p is 1, while the weights of other points are less than 1, and the farther the point is from p, the smaller its weight. h is a parameter that controls the distance range within which the points affect the fitted straight line. Since the value of e -5 is 0.0067, this weight is very small and has a negligible impact on the fitted straight line. Therefore, the relationship between the maximum effective distance D and h is:

[0070]

[0071] Although the embodiments of the present invention have been shown and described, for those of ordinary skill in the art, it can be understood that various changes, modifications, substitutions, and variations can be made to these embodiments without departing from the principles and spirit of the present invention. The scope of the present invention is defined by the appended claims and their equivalents.

Claims

1. A method for extracting road point clouds based on scan lines and sliding least squares fitting, characterized in that Including the following steps: S1. Scan line reconstruction: According to the scan angle of each laser point in the scanner coordinate system, the original laser point cloud is reconstructed into different scan lines; S2. Accurate estimation of the height of the POS system from the road surface: There is a fixed height difference HD between the POS system and the ground, and then the accurate height is obtained through calculation; S3. Robust extraction of road surface seed points: Based on the calculation of the POS data in S2, and then the extraction of seed points is carried out; S4. Complete and robust extraction of road surface points: Based on the seed points obtained in S3, fitting growth is carried out, and then road surface points are obtained. The specific operation method is: Local road surface straight line fitting based on sliding weighted least squares. The points of each scan line are transformed into the scan line coordinate system. The origin of the scan line coordinate system is the midpoint of the road surface point seeds extracted in S3. The X-axis is along the scan line direction, and the Y-axis is vertically upward perpendicular to the X-axis. After transformation into the scan line coordinate system, the coordinates of the road surface seeds extracted in S3 in the scan line coordinate system are represented by r i (x i ,y i ). Let the next point to be determined whether it is a road surface point be p, and the point closest to p be r1. Then the straight line of the local road surface where p is located for the best fit can be obtained by minimizing Equation (4): l(x) = ax + b (3) In formula (4), x i , y i are the coordinates of the current road surface seed point in the scan line coordinate system, and w i is the weight of point r i (x i , y i ). The weight w i is obtained through an exponential weight function: In formula (5), d i is the distance from point r i to point r1. The weights calculated according to formula (5) can ensure that the weight of the point r1 closest to p is 1, while the weights of other points are less than 1, and the farther away from point p, the smaller the weight. h is a parameter that controls the influence of points within a certain distance range on the fitted line.

2. The method for extracting road point cloud based on scan line and sliding least squares fitting according to claim 1, wherein The specific operation method in step S1 is: Based on the on-vehicle laser point cloud or the UAV laser point cloud stored continuously in time sequence, the scan angle of each point in the point cloud can be obtained in the scanner coordinate system. When the scan angles of two consecutive points are converted from 360° to 0°, it means the end of the old scan line and the start of the new scan line. Based on this rule, the above laser point cloud can be divided into scan lines one by one, so as to realize the reconstruction of scan lines.

3. The method for extracting road point cloud based on scan line and sliding least squares fitting according to claim 1, wherein The specific operation method of step S2 is: For a record (t0, x0, y0, z0) in the POS data, first obtain a point set with a time range of [t0 - δt, t0 + δt] from the original data where δt is a relatively small time threshold, and then obtain a point set from the point set in which the distance to (x0, y0) is less than d0 Then the height of the POS system from the road surface can be stably estimated by the following formula: where N is the number of points in the point set , z i is the elevation of the i-th point, and d hd is the estimated height of the POS system from the ground.

4. The method for extracting road point cloud based on scan line and sliding least square fitting according to claim 1, wherein The specific operation method of step S3 is: For each point P in each scan line t (t, x t , y t , z t ), the corresponding POS data at time t can be interpolated from the original POS data. Let O t (t, x, y, z) represent the POS data at time t. Since the POS system is rigidly fixed on the roof platform, although the road surface is not completely flat and the height difference between the POS system and the road surface is not a fixed value, the height difference between the POS system and the road surface is relatively fixed. The height difference between P t and O t and the magnitude relationship between d hd obtained in formula (1) are used to obtain some ground points as seed points; Based on this, the points that satisfy formula (2) can be used as alternative points for road surface seed points: |z t –z–d hd |<ρ hd (2) The points that satisfy formula (2) may include non-road surface points such as vegetation points outside the road in addition to road surface points. In order to robustly obtain road surface seed points, the alternative points for seed points obtained through formula (2) are clustered according to the adjacency of the points in the scan line, and the class with the largest number of points is selected as the final road surface seed points.

Images