CN105354811A - Ground multiline three-dimensional laser radar point cloud data filtering method - Google Patents

Abstract

The present invention belongs to a filtering method and particularly relates to a ground multiline three-dimensional laser radar point cloud data filtering method. The method comprises: 1, establishing a ground curve surface equation, i.e. firstly, determining a ground point, then establishing the ground curve surface equation, and finally, fitting relevant parameters to obtain the ground curve surface equation; and 2, filtering out a misjudgement point. The ground multiline three-dimensional laser radar point cloud data filtering method has the effects that the ground multiline three-dimensional laser radar point cloud data filtering method is simple in implementation, is fast in operation speed, overcomes a limitation of a conventional airborne laser radar and ground single-line laser radar point cloud filtering method, and effectively solves the problem of separation between the ground point and a non-ground point of ground multiline laser radar point cloud data; and a filtering effect of the point cloud data is improved.

Description

A kind of ground filtering method of multi-thread three-dimensional laser radar cloud data

Technical field

The invention belongs to filtering method, be specifically related to the filtering method of a kind of ground with multi-thread three-dimensional laser radar cloud data.

Background technology

Because the multi-thread three-dimensional laser radar in ground can obtain the density three-dimensional cloud data of destination object within a short period of time, obtain the depth information of target, and there is the advantages such as measuring speed is fast, precision is high, direct acquisition ambient condition information, therefore the application in the field such as intelligent vehicle navigation and robot vision is specially adapted to, for intelligent vehicle or robot provide obstacle information.The filtering method of cloud data is one of gordian technique of Three Dimensional Ground laser radar point cloud data treatment research.The filtering of cloud data refers to that an original point cloud data is separated into ground point and non-ground points accurately and rapidly, is Objects extraction, detection of obstacles and follow-up atural object three-dimensional reconstruction or generate the basis of other digital products.

By consulting discovery to existing document, current cloud data filtering method mainly for the Point Cloud Processing of airborne laser radar, and is not suitable for the cloud data of ground multi-line laser radar.This is because, the first, ground multi-line laser radar is multibeam scanning, be mainly used in the field such as obstacle detection, buildings high-speed mapping, general for close shot objective, distance object compared with near, some cloud density is large, data precision is high, the precision required by Data classification is also higher; Second, airborne laser radar is generally single beam scanning, distance objective point is far away, the laser footpoint obtained is sparse distribution at the terrain and its features surface point (buildings then mainly roof point) of extensive area, and ground multi-line laser radar obtains is the geological information of three-dimensional object facade, sampled point is comparatively intensive, cannot continue to use traditional filtering method and process.Therefore filtered classification is carried out to ground multi-line laser radar data and can not directly continue to use airborne laser radar data disposal route.

In addition, part document is had to propose the filtering algorithm based on sweep trace for mobile lidar, but this method for be single line laser radar, and the sweep trace of process is for pressing the equally distributed vertical sweep line of row, and be not suitable for the process of sweep trace by the multi-line laser radar cloud data of row non-uniform Distribution.

Summary of the invention

The present invention is directed to the defect of prior art, a kind of ground filtering method of multi-thread three-dimensional laser radar cloud data is provided.

The present invention is achieved in that the filtering method of a kind of ground with multi-thread three-dimensional laser radar cloud data, comprises the steps:

Step one: set up ground surface equation

First, in the cloud data of input, select minimum point, compared with minimum point by all the other each points, the judgement that difference is less than preset value is ground point, and the judgement being more than or equal to preset value is non-ground points;

Described preset value is the numerical value of outside input;

Then, ground surface equation is set up

Z＝A+BX+CY+DXY+EX ²+FY ²

(X, Y, Z) is wherein laser point cloud coordinate, and A, B, C, D, E, F are quadric surface coefficient to be asked;

Finally, by least square method, for known quantity, by unknown quantity A, B, C, D, E, F matching in above-mentioned formula out, obtain ground surface equation with millet cake fully;

Step 2: filtering erroneous judgement point

For all ground points, with following formulae discovery filtering reference value S

$S=\frac{z_{i+1}-z_i}{\sqrt{{(x_{i+1}-x_i)}^2+{(y_{i+1}-y_i)}^2}}$

Wherein (x, y, z) has been judged to be topocentric laser point cloud coordinate for step one, and i is for representing different points;

Point when judging i=0 is as ground point;

In the result of above-mentioned formulae discovery, if this S value is less than the value preset, judge that this i+1 point is as ground point, otherwise judge that this i+1 point is as erroneous judgement point, and this erroneous judgement point is removed.

Effect of the present invention is: realize simple, fast operation, overcome the limitation of conventional on-board laser radar and ground single line laser radar points cloud filtering method, efficiently solve the ground point of ground multi-line laser radar cloud data and being separated of non-ground, improve the filter effect of cloud data.

Accompanying drawing explanation

Accompanying drawing 1 is 64 bundle laser footpoint distribution plans on the vertical plane corresponding to the certain level anglec of rotation;

Accompanying drawing 2 is the ground adjacent laser pin point schematic diagram of single line laser radar;

Accompanying drawing 3 is the ground adjacent laser pin point schematic diagram of multi-line laser radar.

Embodiment

Below in conjunction with accompanying drawing, the present invention is described further:

The present invention is divided into two steps.First relate to based on the more smooth ground point of the filtering method filtering of surface fitting, illustrate for ground 64 line laser radar.64 line laser radars are arranged 64 laser beam in vertical direction, and therefore corresponding each horizontal rotation angle has 64 laser footpoint, and laser footpoint becomes uneven distribution at random, as shown in Figure 1.Defining frame data is the cloud data that laser radar horizontal direction rotation sweep 360 ° obtains.In general, ground point is the minimum point of elevation, and therefore for frame data, find the minimum point of 64 laser footpoint corresponding to each horizontal rotation angle, these minimum points compared, difference of elevation is greater than should removing of the threshold value of setting.Then utilize remaining minimum point to carry out surface fitting, obtain a Quadratic Surface Equation.Then using this curved surface as initial ground, calculate the Fitting height value on all the other laser footpoint and this simulation ground, if Fitting height value with observation height value difference exceeded threshold value just using this point as culture point, if be less than threshold value, it can be used as ground point filtering.Quadric matching solution procedure is:

Suppose that the Quadratic Surface Equation needing matching is:

Z=A+BX+CY+DXY+EX ²+ FY ²(formula 1)

Wherein, A, B, C, D, E, F are quadric surface coefficient to be asked, and (X, Y, Z) is laser point cloud coordinate.Utilize least square method, these minimum points substitution surface equation can be obtained error equation is:

$V=\left[\begin{array}{cccccc}1& X_1& Y_1& X_1{Y}_1& X_1^2& Y_1^2\\ 1& X_2& Y_2& X_2{Y}_2& X_2^2& Y_2^2\\ 1& X_3& Y_3& X_3{Y}_3& X_3^2& Y_3^2\\ .& .& .& .& .& .\\ .& .& .& .& .& .\\ .& .& .& .& .& .\\ 1& X_n& Y_n& X_n{Y}_n& X_n^2& Y_n^2\end{array}\right]\left[\begin{array}{c}A\\ B\\ C\\ D\\ E\\ F\end{array}\right]-\left[\begin{array}{c}{Z}_1\\ Z_2\\ Z_3\\ Z_4\\ .\\ .\\ .\\ Z_n\end{array}\right]$ (formula 2)

Wherein n > 6 is the number of match point.

Order $M=\left[\begin{array}{cccccc}1& X_1& Y_1& X_1{Y}_1& X_1^2& Y_1^2\\ 1& X_2& Y_2& X_2{Y}_2& X_2^2& Y_2^2\\ 1& X_3& Y_3& X_3{Y}_3& X_3^2& Y_3^2\\ .& .& .& .& .& .\\ .& .& .& .& .& .\\ .& .& .& .& .& .\\ 1& X_n& Y_n& X_n{Y}_n& X_n^2& Y_n^2\end{array}\right],x=\left[\begin{array}{c}A\\ B\\ C\\ D\\ E\\ F\end{array}\right],b=\left[\begin{array}{c}{Z}_1\\ Z_2\\ Z_3\\ Z_4\\ .\\ .\\ .\\ Z_n\end{array}\right]$

According to the indirect adjustment principle of least square, at V ^tunder the condition of PV=min, can in the hope of the solution of Unknown curve surface coefficient, P is power battle array, gets unit matrix here:

X=(M ^tm) ^-1m ^tb (formula 3)

Thus quadric surface coefficient can be tried to achieve.Above-mentioned algorithmic method is simple, fast operation, but when the slope at You compare little angle of inclination, ground, may be mistaken for culture point and retain.Therefore need by second step filtering slope land millet cake.

Do not have barrier ideally at ground level, the ground laser footpoint that certain horizontal rotation angle of single line laser radar is corresponding is the scanned straight lines of a level, as shown in Figure 2.Laser footpoint p _i(x _i, y _i, z _i) and p _i+1(x _i+1, y _i+1, z _i+1) obtain for vertical scan direction two adjacent laser footpoint.Then p _i+1point is relative to p _ithe computing formula of the degree of tilt of point can be obtained with formula 4:

$S=\frac{z_{i+1}-z_i}{\sqrt{{(x_{i+1}-x_i)}^2+{(y_{i+1}-y_i)}^2}}$ (formula 4)

And for 64 line laser radars, because 64 laser footpoint are not uniform by column distribution, but planar distribution comparatively at random.Horizontal range variation range between 64 laser footpoint are adjacent is larger.Fig. 3 is two ground laser footpoint p adjacent in vertical direction _i(x _i, y _i, z _i) and p _i+1(x _i+1, y _i+1, z _i+1) schematic diagram, these two points are not on a horizontal line.

Do not have barrier ideally at ground level, product run-down, ground many horizontal scanning lines can form one group of concentric circles centered by initial point.Area due to slope is general all comparatively large, and the sweep trace on slope can be similar to the straight line regarding one group of parallel distribution as in zonule.Here by p _i+1and p _ithe difference of the concentrically ringed radius at place calculates as the horizontal range of 2, and the degree of tilt solved is the degree of tilt of adjacent two sweep traces, and formula is shown below:

$S^{\′}=\frac{z_{i+1}-z_i}{r_{i+1}-r_i}=\frac{z_{i+1}-z_i}{\sqrt{x_{i+1}^2+y_{i+1}^2}-\sqrt{x_i^2+y_i^2}}$ (formula 5)

Wherein r _iand r _i+1be respectively pin point p _iand p _i+1to the horizontal range of initial point.If p _ipoint is ground point, p _i+1point is relative to p _ithe degree of tilt S' of point is less than threshold value, then think p _i+1for ground point; If be greater than threshold value, then think p _i+1for non-ground points.The selection of threshold value comprehensively determines according to the characteristic of actual ground condition and intelligent vehicle.

Core point of the present invention is for ground multi-line laser radar, proposes a kind of based on the cloud data filtering method of surface fitting in conjunction with degree of tilt.Key point of the present invention is that data process frame by frame, first chooses comparatively low spot reject part abnormal elevation, then carries out Quadratic Surface Fitting, reduce filtering error to frame data; Simultaneously different from traditional method solving degree of tilt, this method is using the degree of tilt of adjacent two horizontal scanning lines as comparison other, instead of simple with the degree of tilt of two adjacent laser footpoint for reference; In addition, choosing of threshold value is also very important, should consider according to combined factors such as the characteristics of concrete road conditions condition and vehicle.This method efficiently solves the ground point of ground multi-line laser radar cloud data and being separated of non-ground, improves the filter effect of cloud data.

Claims (1)

1. a ground filtering method for multi-thread three-dimensional laser radar cloud data, is characterized in that, comprise the steps:

Step one: set up ground surface equation

First, in the cloud data of input, select minimum point, compared with minimum point by all the other each points, the judgement that difference is less than preset value is ground point, and the judgement being more than or equal to preset value is non-ground points;

Described preset value is the numerical value of outside input;

Then, ground surface equation is set up

Z＝A+BX+CY+DXY+EX ²+FY ²

(X, Y, Z) is wherein laser point cloud coordinate, and A, B, C, D, E, F are quadric surface coefficient to be asked;

Finally, by least square method, for known quantity, by unknown quantity A, B, C, D, E, F matching in above-mentioned formula out, obtain ground surface equation with millet cake fully;

Step 2: filtering erroneous judgement point

For all ground points, with following formulae discovery filtering reference value S

$S=\frac{z_{i+1}-z_i}{\sqrt{{(x_{i+1}-x_i)}^2+{(y_{i+1}-y_i)}^2}}$

Wherein (x, y, z) has been judged to be topocentric laser point cloud coordinate for step one, and i is for representing different points;

Point when judging i=0 is as ground point;

In the result of above-mentioned formulae discovery, if this S value is less than the value preset, judge that this i+1 point is as ground point, otherwise judge that this i+1 point is as erroneous judgement point, and this erroneous judgement point is removed.。