CN116188334A - A method and device for automatically repairing lane line point clouds - Google Patents

Abstract

The invention relates to an automatic repair method and device for lane line point cloud, wherein the method comprises the following steps: acquiring vehicle-mounted laser point cloud data, and generating a vehicle-mounted track line according to the vehicle-mounted laser point cloud data; acquiring a plurality of sections on a normal plane of the extending direction of the vehicle-mounted track line; if the peak value position of the first point cloud intensity of the current section changes and the peak value position of the first point cloud intensity of the subsequent section is restored, the lane line between the current section and the subsequent section is missing, the position of the missing lane line is obtained, the supplementing point cloud intensity average value corresponding to the missing lane line is calculated according to the average value of the point cloud intensity of the first peak value interval of the previous section of the current section and the average value of the point cloud intensity of the first peak value interval of the subsequent section, and the supplementing point cloud intensity is supplemented at the position of the missing lane line by the supplementing point cloud intensity average value. According to the defect of the road marking of the expressway, the invention provides a targeted repairing scheme, and the repairing is accurate and efficient.

Description

A lane line point cloud automatic repair method and device

Technical Field

The present invention relates to the technical field of road three-dimensional scene acquisition, and in particular to a method and device for automatically repairing lane line point cloud.

Background Art

With the widespread popularization of informatization, first-class software and hardware technologies such as intelligent construction and 5G communication are gradually entering the commercial market, providing technical support for autonomous driving. Therefore, how to quickly, in real time, and accurately locate the lane level to ensure the safe driving of personnel and vehicles, the concept of high-precision maps came into being. At present, software service providers such as AutoNavi, Baidu, and Tencent each provide their own navigation systems. Lane-level navigation requires accurate and real-time lane line information, and how to accurately obtain its three-dimensional spatial position and morphological characteristics has become the key to the technology.

Vehicle-mounted laser equipment is a new type of surveying and mapping equipment with automatic spatial collection. It is not restricted at night. Compared with sensors such as cameras, the detection distance of laser radar can be 200 to 300m+, with higher resolution. It has more advantages in low visibility environments such as night and rainy days, and can significantly improve the reliability of the autonomous driving system. It has gradually entered the public eye since the first DARPA driverless challenge. With Valeo as the leader, many self-service manufacturers such as RoboSense, Lanwo (DJI), Tudatong, Huawei, and Hesai have continued to cooperate deeply with downstream OEMs, making the laser industry market fiercely involuted, and vehicle-mounted laser equipment has been fully accelerated. Iteration upgrade. Therefore, using high-precision vehicle-mounted laser scanning equipment to scan the road surface can last for a long time, with a collection accuracy better than 10cm, providing reliable data guarantee for the accurate extraction of marking lines.

At present, the extraction of road markings is mainly divided into point cloud feature clustering to extract road markings, and the other is combined with deep learning for extraction. The use of deep learning algorithms requires not only training samples, but also continuous experimentation and iteration to obtain appropriate experience values, which leads to limited automation and cumbersome implementation processes. For the technical route of point cloud feature clustering to extract road markings, point cloud intensity information is used for clustering. This type of method has a high degree of automation, but it relies on the quality of the data itself. For example, road surface strength is affected by road noise, making it difficult to extract high-precision road marking features from vehicle-mounted laser data.

Summary of the invention

The purpose of the present invention is to solve the problem that in the process of clustering point cloud intensity information to extract lane markings, the point cloud data quality is not high, which affects the quality of the extracted lane lines, and propose a lane line point cloud automatic repair method and device.

A lane line point cloud automatic patching method comprises the following steps:

Acquire vehicle-mounted laser point cloud data, and generate a vehicle-mounted trajectory line according to the vehicle-mounted laser point cloud data;

Acquire a plurality of cross sections in a normal plane in the extending direction of the vehicle-borne trajectory line;

If the position of the first point cloud intensity peak of the current section changes, and the position of the first point cloud intensity peak of the subsequent section returns to its original state, then the lane line between the current section and the subsequent section is missing, and the position of the missing lane line is obtained.

The mean value of the supplementary point cloud intensity corresponding to the missing lane line is calculated based on the mean value of the point cloud intensity of the first peak interval of the previous section of the current section and the mean value of the point cloud intensity of the first peak interval of the subsequent section, and the point cloud intensity is supplemented at the position of the missing lane line with the supplementary point cloud intensity mean.

As a preferred solution, the calculation process of the position of the missing lane line is:

The missing lane line is considered to be parallel to the trajectory line from trajectory point _Sp to trajectory point _Sq . The width from the trajectory point to the center of the missing lane line is a fixed length value. According to the fixed length value and the length of the missing lane line, the missing lane line repair range is obtained. The missing lane line is composed of the center position _Fm ( _xm , _ym , _zm ) of the missing lane line.

As a preferred solution, the calculation steps of the supplementary point cloud intensity mean are:

，Step a, calculate the mean value I ₁ of the point cloud intensity of the first peak interval of the previous section of the current section,

,

Where, _kp represents the number of point clouds in the first point cloud peak interval of the section where the trajectory point S _p is located; _Ii is the intensity value corresponding to the point cloud in the first point cloud peak interval of the section where the trajectory point S _p is located.

And calculate the mean value I ₂ of the point cloud intensity of the first peak interval of the subsequent section,

，

,

Where _kq represents the number of point _clouds in the first point cloud peak interval of the cross section where the trajectory point _Sq is located; _Ij is the intensity value corresponding to the point cloud in the first point cloud peak interval of the cross section where the trajectory point Sq is located;

Step b, the calculation formula for the supplementary point cloud intensity mean _Im is:

，

,

Where _Im is the mean value of the supplementary point cloud intensity, _I1 is the mean value of the point cloud intensity of the first peak interval of the previous section of the current section, and _I2 is the mean value of the point cloud intensity of the first peak interval of the subsequent section.

As a preferred solution, a method for extracting lane lines is also included, including the following steps:

S1, obtaining laser point cloud data of the road surface;

S2, filtering the laser point cloud data using a gradient filtering method, deleting non-road point cloud data with large fluctuations, and obtaining point cloud coarse extraction data;

S3, filtering the rough point cloud extraction data using a cloth filtering method to obtain fine point cloud extraction data;

S4, the point cloud extracted data is subjected to dark channel denoising and k-d tree search neighborhood denoising in turn to obtain optimized lane line point cloud data.

As a preferred solution, in step S2, the laser point cloud data is filtered using a gradient filtering method, specifically including: scalaring the Z coordinates in the three-dimensional coordinates of all laser point cloud data, calculating the gradient values of all laser point cloud data, setting the minimum gradient value to δ, and deleting the laser point cloud data with a gradient value greater than δ to obtain rough point cloud extraction data.

As a preferred solution, in step S3, filtering the rough extracted point cloud data using a cloth filtering method specifically includes the following steps:

S301, setting the filter grid size, and performing CSF cloth simulation on the points after gradient filtering;

S302, performing horizontal plane projection on the simulation points and the laser points to find the elevation of the laser point closest to each simulation point;

S303, determining the height of the cloth simulation particles under the influence of gravity;

S304, calculating the internal force between the cloth simulation particles, and calculating the displacement caused by the internal driving factors according to the set cloth rigidity parameters;

S305, repeating step S303 and step S304 to perform calculations until the number of iterations reaches the set maximum number of iterations;

S306, comparing the distance difference between the laser point and the simulation point, when the difference is less than or equal to the threshold, the corresponding laser point is marked as a ground point, and when the difference is greater than the threshold, the laser point is marked as a non-ground point.

As a preferred solution, the dark channel denoising specifically includes the following steps: treating the squeezed damaged road surface as noisy, and processing the point cloud fine extraction data based on the dark channel defogging to correct the partial distortion of the road surface intensity; the dark channel defogging processing is to regard the intensity data I(x) with noise reflection as fog, and use the dark channel defogging model to calculate the corrected fog-free point intensity value, and the dark channel defogging model is:

，

,

Among them, I(x) is the intensity data with noisy reflection, t(x) is the transmittance, A is the atmospheric light intensity, and J(x) is the corrected fog-free intensity value.

As a preferred solution, the modified fog-free point intensity value J(x) formula is as follows:

，

,

Where I(x) is the intensity data with noisy reflection, t(x) is the transmittance, A is the atmospheric light intensity, J(x) is the corrected fog-free intensity value, and _t0 is the minimum critical value of transmittance.

As a preferred solution, the neighborhood denoising of k-d tree search specifically includes the following steps:

Step 1: Calculate the intensity average I _avg within the preset range:

，

,

Among them, I _i is the intensity value of different laser points within a neighborhood in the laser point cloud data, i is the serial number of the laser point, and k is the number of laser points within the search range within a neighborhood;

计算：Step 2: Perform variance calculation within the preset range

calculate:

，

,

Step 3, calculate the standard deviation _Si within the preset range:

，

,

Step 4: If _Si > I _s , the intensity values within the main point range are sorted, and the point with the largest intensity value can be considered as a road noise point and removed. I _s is the standard deviation threshold.

Based on the same concept, a lane line point cloud automatic patching device is also proposed, including at least one processor and a memory communicatively connected to the at least one processor; the memory stores instructions executable by the at least one processor, and the instructions are executed by the at least one processor so that the at least one processor can execute any of the above-mentioned lane line point cloud automatic patching methods.

Compared with the prior art, the present invention has the following beneficial effects:

The method of the present invention calculates the mean value of the supplementary point cloud intensity corresponding to the missing lane line based on the mean value of the point cloud intensity of the first peak interval of the previous section of the current section and the mean value of the point cloud intensity of the first peak interval of the subsequent section, and supplements the point cloud intensity at the position of the missing lane line with the supplementary point cloud intensity mean value, thereby repairing the missing lane line. The present invention provides a targeted repair scheme for the missing lane markings on highways, and the scheme can also be extended to the repair of lane markings on other sections, with accurate and efficient repair.

BRIEF DESCRIPTION OF THE DRAWINGS

FIG1 is a flow chart of a method for automatically extracting lane point clouds in Embodiment 1 of the present invention;

FIG2 is a flow chart of performing CSF cloth filtering in Embodiment 1 of the present invention;

FIG3 is a flowchart of the steps of denoising the method for performing neighborhood search using a k-d tree in Embodiment 1 of the present invention;

FIG4 is a schematic diagram of a cross section in Example 3 of the present invention;

FIG5 is a schematic diagram of the width d from a track point to the center of a missing marking line in Example 3 of the present invention;

FIG6 is a flow chart of a complete lane line point cloud automatic repair method in Embodiment 3 of the present invention.

DETAILED DESCRIPTION

The present invention is further described in detail below in conjunction with test examples and specific implementation methods. However, this should not be understood as the scope of the above subject matter of the present invention being limited to the following embodiments, and all technologies realized based on the content of the present invention belong to the scope of the present invention.

Example 1

A method for automatically extracting lane line point cloud, the flow chart of which is shown in FIG1, comprises the following steps:

S1, obtaining laser point cloud data of the road surface.

The vehicle-mounted laser equipment is used to collect data on a certain high-speed pilot road section, including the left and right main road surfaces. The collected data is laser point cloud data that contains point cloud intensity value information, and the laser point cloud data is composed of several laser points. The collected high-speed road surface laser point cloud data is attributed, and the basic attribute values X, Y, Z, and I are selected, where X, Y, and Z represent the three-dimensional coordinates of the point, and I represents the intensity value (intensity data) of the laser point.

Secondly, the laser point cloud data is preprocessed, including gradient filtering, cloth filtering and denoising.

S2, using the gradient filtering method to filter the laser point cloud data, delete the non-road point cloud data with large fluctuations, and obtain the point cloud coarse extraction data.

：Gradient filtering is used for rough extraction of point clouds, specifically: scalaring the Z coordinate of the ground laser point, performing kd tree 3D spatial indexing on the laser point cloud data, performing neighbor search within the main point range, and obtaining the elevation gradient mean of the main point M.

:

，

,

代表Z方向单位法向量，

代表点之间的空间向量，k代表主点M范围内搜索到的邻点个数，i是主点的个数，随后对

进行筛选，设定筛选值的阈值δ=0.1，对小于δ的点进行保留，否则删除。in,

represents the unit normal vector in the Z direction,

represents the spatial vector between the points, k represents the number of neighboring points searched within the range of the principal point M, i is the number of principal points, and then

For screening, set the screening value threshold δ = 0.1, retain the points less than δ, otherwise delete them.

The non-road point cloud with large fluctuations is deleted through gradient filtering to obtain the rough extraction data of the point cloud.

S3, using the cloth filtering method to filter the rough extraction data of the point cloud to obtain the fine extraction data of the point cloud.

Then, CSF cloth filtering is performed to further screen the data for accurate extraction of road surface point cloud data. The flowchart of CSF cloth filtering is shown in Figure 2. The specific steps are as follows:

Step S301, set the filter grid size. The smaller the grid, the higher the simulation accuracy and the longer the calculation time. Perform CSF cloth simulation on the point _Pi after gradient filtering. The filter grid size can be freely set. The size of the grid setting determines the accuracy of the terrain simulation algorithm. Preferably, the grid size is set to 0.1. Setting 0.1 does not take long to calculate and has good simulation accuracy. Set the highest elevation point. The elevation value of the highest elevation point is greater than the elevation value of the point cloud after gradient filtering.

Step S302, setting cloth simulation points (cloth simulation particles) and laser points according to the grid size in step S301, performing horizontal plane projection on the simulation points and the laser points, and finding the elevation h _i of the laser point closest to each simulation point.

Step S303, the cloth simulation particle is regarded as a movable object. The cloth simulation particle is affected by gravity. The height of the cloth simulation particle is determined under the influence of gravity. When the height of the cloth simulation particle h _{v ≤hi} , the height of the cloth simulation particle is set to h _v and set to an immovable point.

。Step S304, calculate the internal force between the cloth simulation particles, and calculate the displacement caused by the internal driving factors according to the set cloth rigidity parameters.

.

，

,

为垂直方向单位向量，用来P_i指代模拟粒子，并且P_i为P₀的相邻粒子。In the formula, when the particle is in a moving state, the b value is 1, otherwise it is 0.

is a unit vector in the vertical direction, used to represent the simulated particle _Pi , and _Pi is the adjacent particle of _P0 .

Step S305, repeat step 3 and step 4 to calculate until the number of iterations reaches the set maximum number of iterations. Particles need to be affected by both external and internal forces during the simulation process. The external force is gravity, and the internal force is the interaction force between particles. During the particle simulation process, step 3 simulates the external force, and step 4 simulates the internal force until the maximum height change of all particles is small enough or the number of iterations reaches a preset value, then the simulation is stopped. As a preferred solution, the present invention uses the number of iterations as a condition for stopping the simulation. Here, the maximum number of iterations is set to 500 to obtain the final simulated height of the cloth particles. The simulation of the particles is affected by external forces (gravity) and internal forces, and both forces will affect the height of the simulated particles. The more iterations, the more accurate the simulation.

Step S306, compare the distance difference between the laser point and the simulation point. When the difference is less than or equal to the threshold, the corresponding laser point is marked as a ground point. When the difference is greater than the threshold, the laser point is marked as a non-ground point. Here, the threshold is set to 0.5 meters.

Through steps S301 - S306 , the point cloud precise extraction data of all processed road surfaces are obtained.

S4, the point cloud extracted data is subjected to dark channel denoising and k-d tree search neighborhood denoising in turn to obtain the optimized lane line point cloud data.

After a long period of high-load operation, the asphalt material of the highway pavement is squeezed and worn for a long time. The reflection value of the surface asphalt obtained after the vehicle-mounted laser scanning is different from the reflection value of the unworn pavement. This squeezed and damaged pavement is regarded as noisy and denoised. The pavement intensity data is defogged based on the dark channel to correct the partial distortion of the asphalt pavement intensity. The intensity data I(x) with noise reflection is regarded as fog, and the dark channel defogging model is as follows:

，

,

In the formula, J(x) is the fog-free data, and the road point cloud noise is regarded as fog. The corrected fog-free point intensity value J(x) is the road point cloud after noise removal, and t(x) is the transmittance, where A is the atmospheric light intensity. Generally speaking, the larger the transmittance value, the more obvious the image fog. Here, the transmittance can be obtained according to the point cloud intensity distribution frequency p, that is,

，

,

When the transmittance t(x) approaches 0 infinitely, the image is considered to be a fog-free image. However, when t(x)=0, the image color will be oversaturated, so a minimum critical value t ₀ =0.1 is set for t(x) to make the image closer to the effect under natural light conditions. Therefore, the formula for the corrected fog-free point intensity J(x) value is as follows:

，

,

In the formula, A is the atmospheric light intensity, t(x) is the transmittance, _t0 is the minimum critical value of transmittance, and I(x) is the intensity data with noise reflection. The intensity data with noise reflection I(x) is calculated by the above formula to obtain the corrected fog-free point intensity value J(x), thus completing the dark channel point cloud filtering.

After the dark channel point cloud filtering is completed, there are still some irregular white bright spots, whose intensity values are significantly higher than the intensity of the surrounding points. Therefore, the kd tree is used to perform neighborhood search, and the intensity standard deviation of the laser point is calculated at the same time. If the standard deviation is greater than the standard deviation threshold _Is , it is considered that there are strong white noise points within the main point range, and the noise points are removed. The range of the standard deviation threshold _Is is determined according to the point cloud data itself, and it is not a universal fixed value. Different models of vehicle-mounted laser equipment scan the same object, and the intensity values of the points are different. The denoising steps of the method of neighborhood search using the kd tree are shown in Figure 3. Here, individual independent points of the remaining point cloud are denoised after gradient filtering, CSF filtering, and dark channel denoising. The specific steps are as follows:

Step 1: Use the kd tree to perform neighborhood search and calculate the point intensity mean I _avg of all searched neighborhood point sets. The calculation formula is:

，

,

Among them, I _i is the intensity value of different laser points within a neighborhood in the laser point cloud data, i is the serial number of the laser point, and k is the number of laser points within the search range within a neighborhood.

计算：Step 2: Perform point set variance calculation within the neighborhood

calculate:

，

,

Step 3, calculate the standard deviation _Si of the point set within the neighborhood:

，

,

Step 4: If _Si > _Is , the intensity values of the point set within the neighborhood are sorted, and the point with the largest intensity value can be considered as a road noise point and removed.

The intensity value of isolated noise points is obviously greater than that of surrounding points. If we only do sorting, we cannot find isolated noise points. Therefore, we need to calculate the standard deviation to find the standard deviation value between the main point and the adjacent points, filter the standard deviation _Si by the standard deviation threshold _Is , and then sort to exclude isolated noise points.

After the abnormal noise points of the road surface point cloud are removed, the preprocessing steps such as denoising of the point cloud extracted data are completed, and the optimized lane point cloud data is obtained. The above method sequentially performs gradient filtering, cloth filtering, dark channel denoising and k-d tree search neighborhood denoising on the acquired laser point cloud data of the road surface, making the extracted lane line point cloud data more accurate. The extracted lane line can accurately provide the road lane-level position information, with a high degree of automation, and the information provided is accurate, stable and reliable.

Example 3

According to the method of Example 1 or Example 2, filtering and denoising of the point cloud data are implemented to obtain optimized lane point cloud data, so that the optimized lane point cloud data has higher quality, which is helpful for the subsequent high-precision extraction of lane lines.

In addition to the preprocessing of the point cloud data itself, it is also necessary to overcome the problem of lane line occlusion. When the vehicle-mounted laser scans the road surface, the road surface points are easily affected by the vehicle occlusion, resulting in partial missing of the road surface point cloud. Therefore, the POS trajectory points are combined with the road shape to make up for the local linear marking point cloud:

The driving trajectory points are used to track forward from the initial point and obtain cross-sections along the scanning direction. Since there are integral pavements and separated pavements on highways, half of the highway pavement is calculated here.

During the vehicle-mounted laser scanning process, the vehicle-mounted trajectory points are automatically generated. The equation for generating the trajectory line based on the trajectory points is V(s)=[x(s),y(s),z(s)], where x(s),y(s),z(s) are the coordinate information of each point that composes the trajectory line, s is the normalized arc length, and V(s) refers to the trajectory line that the vehicle-mounted laser passes through, which is a set expression of the trajectory points composed of each set of trajectory points.

During the on-board laser scanning process, it was found that the reflection intensity of the marking was significantly higher than that of the asphalt pavement. Therefore, the section position Section(k) of the trajectory point _Sk was defined, and the peak area of the entire intensity value interval of the current section Section(k) was calculated to obtain the marking information.

作为截面法向量

：First, calculate the section Section(k). The section is defined as the position of the normal plane in the extension direction of the trajectory line, according to the trajectory vector of the trajectory point _Sk

As the cross-section normal vector

:

，

,

分别代表轨迹点S_k到S_k+1在x、y、z方向的空间差值。根据法向量

即可构建法平面，截面所在的法平面方程为：In the formula,

Represent the spatial difference between the trajectory point S _k and S _k+1 in the x, y, and z directions respectively.

The normal plane can be constructed, and the equation of the normal plane where the section is located is:

，

,

=5cm为截面宽度。Here is defined

=5cm is the section width.

Affected by the driving on the road, the lane line is blocked and the laser scanning cannot obtain the local spatial point cloud features. In general, it is believed that under the same scanning distance, the reflection intensity of the vehicle-mounted laser of the same structure remains basically unchanged. Therefore, by calculating the intensity distribution position of the cross-sectional point cloud, the situation where the vacant long solid line marking may change is judged. Here, the partial missing of the lane long marking near the central isolation belt is taken as an example.

It should be noted that the method of repairing lane line point cloud in this patent is applicable to the long solid line markings on the left and right sides of the highway, and the subsequent formula is also applicable to the point cloud repair of the long solid line markings between the main lane and the emergency lane. Since most vehicles drive according to the lanes when the vehicle-mounted laser performs scanning operations, it is basically impossible for the short solid line markings between lane 1 and lane 2 to be blocked by vehicles and cannot be scanned. The following formulas and methods are only used for the repair of long solid line markings on both sides of the left and right sides of the road.

Step 1: When there is an obvious point cloud gap in the cross-section point cloud, start the calculation. At this position, the point cloud of the cross-section corresponding to the driving trajectory point changes significantly, that is, the position of the first point cloud intensity peak of the cross-section changes significantly. The point cloud intensity of the first peak interval of the previous trajectory point _Sp cross-section is read and the average value I ₁ is calculated:

，

,

Where _kp represents the number of point clouds in the first point cloud peak interval of the section where the trajectory point _Sp is located.

Step 2: When the first peak position of the section where the driving trajectory point _Sq is located returns to normal, that is, there is no point cloud missing in the current section, record the point cloud intensity of the first peak interval of the current section and calculate the average value I ₂ :

，

,

Where _kq represents the number of point clouds in the first point cloud peak interval of the section where the trajectory point _Sq is located.

Step 3: The interval where the first point cloud intensity peak position changes is the interval from track point _Sp to track point _Sq on the track line. The missing marking line is considered to be parallel to the track line. The section positions corresponding to track points _Sp and _Sq can be used to obtain the section center positions at both ends of the missing marking line under the current section. The center line of the missing marking line can be obtained by connecting the section center positions at both ends of the two missing marking lines. Since the marking line width is a fixed value, the missing marking line point cloud can be repaired according to the center line and marking line width of the missing marking line.

=d，根据缺失标线的中心位置F_m(x_m,y_m,z_m)连接构成的缺失标线的中心线和固定长度d，就可以得到缺失标线修补范围。The distance represented by the width d from the trajectory point to the center of the missing marking is shown in Figure 5. The interval from the trajectory point _Sp to the trajectory point _Sq on the trajectory line can be considered as straight-line driving. The center line of the missing marking formed by connecting the center position _Fm ( _xm , _ym , _zm ) of the missing marking is considered to be parallel to the trajectory line from the trajectory point _Sp to the trajectory point _Sq. The width from the center of the missing marking to the trajectory point is a fixed length value

=d, and the center line of the missing marking line formed by connecting the center position of the missing marking line _Fm ( _xm , _ym , _zm ) and the fixed length d can be obtained to obtain the missing marking line repair range.

The length of d is the distance between the peak position center of the point cloud of the first lane line on the left and the vehicle-mounted laser trajectory point _Sp :

，

,

Where ( _xp , _yp ) represents the coordinates of the track point _Sp . The center position _Fm of the current missing marking and the current vehicle track point _Sq can be considered to be at the same elevation, so _zm = _zp . Therefore, the normal plane equation of the vehicle track point can be transformed into the normal line equation:

，

,

分别代表轨迹点S_k到S_k+1在x、y方向的空间差值，F_m(x_m,y_m,z_m)是缺失标线的中心位置，(x_p,y_p)是车载激光轨迹点S_p在x、y平面上的二维坐标。对上述两式联立即可获得缺失标线的中心位置F_m(x_m,y_m,z_m)，距离方程和法线方程中除了x_p,y_p均为已知数，联立可以求得F_m。In the formula,

Respectively represent the spatial difference between the track point _Sk to _Sk+1 in the x and y directions, _Fm ( _xm , _ym , _zm ) is the center position of the missing marking, ( _xp , _yp ) is the two-dimensional coordinate of the vehicle-mounted laser track point _Sp on the x and y plane. The center position of the missing marking _Fm ( _xm , _ym , _zm ) can be obtained by combining the above two equations. Except for _xp and _yp , all other numbers in the distance equation and the normal equation are known numbers, and _Fm can be obtained by combining them.

The mean values _I1 and _I2 obtained in the above steps 1 and 2 are used to obtain the compensation point cloud intensity I _m within the missing marking repair range, and the missing marking is repaired by point cloud. The calculation formula of the compensation point cloud intensity I _m is:

，

,

Here, the compensation for the missing section of the long solid lane marking is not limited to the compensation for the central isolation zone position marking, but is also applicable to the point cloud compensation for the long solid lane line between the driving lane and the emergency parking lane.

Calculate the lengths Length _p and Length _Q of the peak strength interval P and the non-peak strength interval Q of the cross section (taking a two-way 4-lane highway as an example, this patent method is not limited to the number of lanes on the highway, and is still applicable to other road surfaces). If the lengths Length _p and Length _Q meet the following conditions 1 or 2 (condition 1 is shown in section 1 of the attached figure, and condition 2 is shown in section 2), it is determined that the cross section has a long solid marking line:

，

,

，

,

In the formula, (x _k1 ,y _k1 ) and (x _k2 ,y _k2 ) represent the coordinates of the starting and ending points of the first peak interval of the cross section, and (x _k3 ,y _k3 ) and (x _k4 ,y _k4 ) represent the coordinates of the starting and ending points of the second peak interval of the cross section. Specifically:

1) When there are two point cloud intensity peak intervals P in section Section(k), and the length of the non-peak interval Q meets condition 2, there are two long solid marking lines on the section, and no other types of marking lines, as shown in section 2 in Figure 4;

2) When there are three point cloud intensity peak intervals P in section Section(k), and the length of the non-peak interval Q meets condition 1, there are two long solid marking lines and one short solid marking line on the section, and no other types of marking lines, as shown in section 1 in Figure 4;

3) When there are 4 or more point cloud intensity peak intervals P in section Section(k), and the distance between the length of the first peak interval and the length of the last peak interval satisfies Length _Q2 , it means that there are different types of marking features on the section, with solid markings at both ends and other types of markings such as lane guide arrows at the middle position. The situation is shown in Section 3 of Figure 4;

By counting the peak positions of all sections, retaining the point sets on the peak interval sections in cases 1 and 2, and retaining the point sets on the first and last peak interval sections in case 3, all lane line positions can be obtained. FIG6 is a flow chart of a complete lane line point cloud automatic repair method of the present invention.

The above shows and describes the basic principles and main features of the present invention and the advantages of the present invention. It is obvious to those skilled in the art that the present invention can be implemented in other specific forms without departing from the spirit or basic features of the present invention. Therefore, no matter from which point of view, the embodiments should be regarded as exemplary and non-restrictive. The scope of the present invention is limited by the attached claims rather than the above description, and it is intended that all changes falling within the meaning and scope of the equivalent elements of the claims are included in the present invention. Any figure mark in the claims should not be regarded as limiting the claims involved.

In addition, it should be understood that although the present specification is described according to the implementation mode, the implementation mode does not only include an independent technical solution. This narrative method of the specification is only for the sake of clarity. Those skilled in the art should regard the specification as a whole. The technical solutions in the embodiments can also be appropriately combined to form other implementation modes that can be understood by those skilled in the art.

Claims (10)

1. The automatic lane line point cloud repairing method is characterized by comprising the following steps of:

acquiring vehicle-mounted laser point cloud data, and generating a vehicle-mounted track line according to the vehicle-mounted laser point cloud data;

acquiring a plurality of sections on a normal plane of the extending direction of the vehicle-mounted track line;

if the position of the first point cloud intensity peak value of the current section changes and the position of the first point cloud intensity peak value of the subsequent section is restored, the lane line between the current section and the subsequent section is missing, the position of the missing lane line is obtained,

and calculating a supplementary point cloud intensity mean value corresponding to the missing lane line according to the mean value of the point cloud intensity of the first peak interval of the previous section of the current section and the mean value of the point cloud intensity of the first peak interval of the subsequent section, and supplementing the point cloud intensity at the position of the missing lane line by using the supplementary point cloud intensity mean value.

2. The automatic repair method for lane line point cloud as claimed in claim 1, wherein the calculation process of the position of the missing lane line is:

the missing lane line is regarded as the slave track point S _p To locus point S _q The track line of the line is kept parallel, the width from the track point to the center of the missing line is a fixed length value, the repair range of the missing line is obtained according to the fixed length value and the length of the missing line, and the missing line is repaired by the center position F of the missing line _m (x _m ,y _m ,z _m ) And the connection is formed.

3. The automatic repair method for the lane line point cloud as claimed in claim 1, wherein the calculation step of the supplementary point cloud intensity mean value is as follows:

step a, calculating the average value I of the point cloud intensity of the first peak interval of the previous section of the current section ₁ ，

，

Wherein k is _p Representative track point S _p The number of point clouds in the peak interval of the first point cloud of the section; i _i Is the locus point S _p The corresponding intensity value of the point cloud in the peak value interval of the first point cloud of the section, and the average value I of the intensity of the point cloud in the first peak value interval of the subsequent section is calculated ₂ ，

，

Wherein k is _q Representative track point S _q The number of point clouds in the peak interval of the first point cloud of the section; i _j Is the locus point S _q The corresponding intensity value of the point cloud in the peak value interval of the first point cloud of the section;

step b, supplementing the point cloud intensity mean value I _m The calculation formula is as follows:

，

wherein I is _m Is to supplement the point cloud intensityValue, I ₁ Is the average value of the point cloud intensity of the first peak interval of the previous section of the current section, I ₂ Is the average value of the point cloud intensity of the first peak interval of the subsequent section.

4. The automatic repair method for lane line point cloud according to any one of claims 1 to 3, further comprising a method for lane line extraction, comprising the steps of:

s1, acquiring laser point cloud data of a road surface;

s2, filtering the laser point cloud data by adopting a gradient filtering method, deleting non-road surface point cloud data with large fluctuation change, and obtaining point cloud rough extraction data;

s3, filtering the point cloud coarse extraction data by using a cloth filtering method to obtain point cloud fine extraction data;

and S4, denoising the point cloud extract data sequentially through dark channel denoising and neighborhood denoising of k-d tree searching to obtain optimized lane line point cloud data.

5. The method for automatically repairing the lane line point cloud as claimed in claim 4, wherein in the step S2, filtering the laser point cloud data by using a gradient filtering method specifically comprises: scalar quantity is carried out on Z coordinates in three-dimensional coordinates of all laser point cloud data, gradient values of all laser point cloud data are calculated, the minimum gradient value is set as delta, and laser point cloud data with gradient values larger than delta are deleted to obtain point cloud rough extraction data.

6. The method for automatically repairing the point cloud of the lane line according to claim 4, wherein in the step S3, the filtering of the point cloud rough extraction data by using a cloth filtering method specifically comprises the following steps:

s301, setting the size of a filtering grid, and simulating CSF (human body fluid) distribution of points after gradient filtering;

s302, carrying out horizontal projection on the simulation points and the laser points, and searching the elevation of the nearest laser point of each simulation point;

s303, judging the height of cloth simulation particles under the condition of gravity influence;

s304, calculating the internal force action among cloth simulation particles, and calculating the displacement generated by the influence of internal driving factors according to the set cloth rigidity parameters;

s305, repeating the step S303 and the step S304 to calculate until the iteration number reaches the set maximum iteration number;

s306, comparing the distance difference between the laser points and the analog points, marking the corresponding laser points as ground points when the difference is smaller than or equal to a threshold value, and marking the laser points as non-ground points when the difference is larger than the threshold value.

7. The automatic repair method for lane line point cloud as claimed in claim 4, wherein the denoising through the dark channel comprises the steps of: regarding the extruded damaged pavement as noisy, performing defogging treatment on the point cloud extract data based on a dark channel, and correcting the condition of partial distortion of the pavement intensity; the defogging treatment based on the dark channel is to consider intensity data I (x) with noise reflection as fog, calculate and obtain a corrected fogliless point intensity value by adopting a dark channel defogging model, wherein the dark channel defogging model is as follows:

，

wherein I (x) is intensity data with noise reflection, t (x) is transmittance, a is atmospheric light intensity, and J (x) is corrected fogliless intensity value.

8. The method for automatically repairing a lane line point cloud as claimed in claim 7, wherein the corrected fogliless point intensity value J (x) is formulated as follows:

，

wherein I (x) is intensity data with noise reflection, t (x) is transmissivity, A is atmospheric light intensity, and J (x) is corrected fogliless point intensityDegree value, t ₀ Is the minimum threshold of transmissivity.

9. The method for automatically repairing the lane line point cloud as claimed in claim 4, wherein the neighborhood denoising of the k-d tree search specifically comprises the following steps:

step 1, carrying out intensity average value I in a preset range _avg And (3) calculating:

，

wherein I is _i The method is characterized in that the method comprises the steps that (1) the intensity values of different laser points in a certain neighborhood in laser point cloud data are obtained, i is the serial number of the laser points, and k is the number of the laser points in a searching range in the certain neighborhood;

step 2, performing variance in a preset range

And (3) calculating:

，

step 3, carrying out standard deviation S within a preset range _i And (3) calculating:

，

step 4, if S _i ＞I _s Sorting the intensity values in the main point range, namely taking the point with the maximum intensity value as the road surface noise point, removing, and I _s Is the standard deviation threshold.

10. The lane line point cloud automatic repairing device is characterized by comprising at least one processor and a memory in communication connection with the at least one processor; the memory stores instructions executable by the at least one processor to enable the at least one processor to perform a lane line point cloud auto repair method according to any one of claims 1 to 9.

Images