CN105115441A - Feature point extraction automatic segmenting method for profile of revolution solid part - Google Patents

Abstract

The invention discloses a feature point extraction automatic segmenting method for the profile of a revolution solid part, and the method comprises the steps: scanning a revolution solid through a sensor, and carrying out uniform sampling at equal intervals; calculating the curvature of each obtained sampling point of the revolution solid in a (2n+1) curvature calculation method through sampling data; obtaining the curvature of each sampling point of the revolution solid after deformation; enabling a point with the curvature being greater than a set threshold value after approximate deformation to serve as a feature point of the revolution solid; extracting the feature point and carrying out segmenting; obtaining different straight lines or curved lines; achieving the extraction and segmenting of feature points of the profile of the revolution solid part in the radial direction of an axis, wherein the revolution solid part has dense saltation and steps and more trenches.

Description

A kind of feature point extraction of revolving parts profile and automatic segmentation method

Technical field

The present invention relates to geometric measurement technical field, be specifically related to a kind of feature point extraction and automatic segmentation method of revolving parts profile.

Background technology

Revolving parts is widely used in the every field such as weapons, Aero-Space, machinery, automobile, and the profile of revolving parts has very large impact to revolving parts assembly performance and using function, therefore needs to detect it and evaluate.Pickup unique point is as the gordian technique of contour segmentation and recognition technology, and lot of domestic and international scholar is studied the segmentation of the outline data by measuring equipment or scanning device acquisition and identification.The pick-up method of existing unique point adopts point with extreme curvature method.Point with extreme curvature method is the unique point being identified curve segmentation by the Curvature varying of curve, ultimate principle is: the approximate curvature first calculating each profile measurement point, then curvature absolute value is greater than the Local Extremum of setting threshold values as waypoint---the angle point of profile.The outline data of the method can produce corresponding noise by curvature during noise, and produces pseudo-extreme point and pseudo-angle point thus.

In the process of revolving parts contour segmentation, number of contours strong point can be subject to the interference of noise, if revolving parts is little along the radial direction sudden change of axis, the impact of noise on feature point extraction is smaller, even negligible, adopt said method can realize extraction to its unique point and segmentation, and reach good effect; But, if revolving parts profile compares comparatively dense along the radial direction sudden change of axis, step, groove are more, the impact of noise on segmentation is larger, now often expect reducing sampling interval, but owing to reducing sampling interval, the number of sampled point is increased, each sampled point needs to calculate curvature, Curvature varying can be caused disorderly and unsystematic, cannot judging characteristic point position, causing cannot extract minutiae, thus cannot carry out segmentation to revolving parts, and therefore the extraction of said method to this type of solid of revolution unique point has some limitations.

Summary of the invention

In view of this, the invention provides a kind of feature point extraction and automatic segmentation method of revolving parts profile, the radial direction along axis can be suddenlyd change than the revolving parts contours extract unique point of comparatively dense, thus realize closely-spaced segmentation.

The feature point extraction of this revolving parts profile and automatic segmentation method, concrete steps are as follows:

Step one, the continuous sweep revolution bodily form are wide, go forward side by side and are divided into the uniform sampling at equal intervals of Δ x in the ranks, obtain sampling number certificate in real time;

Step 2, according to obtaining sampling number certificate in step one, in 2n+1 curvature estimation method, utilizing the curvature of formula (7) to each sampled point of solid of revolution obtained to be out of shape, obtaining the curvature after the distortion of each sampled point of solid of revolution:

$k_i=\frac{{\stackrel{\·\·}{y}}_i}{{\stackrel{\·}{x}}_i^2-{\stackrel{\·}{y}}_i^2}---\left(7\right)$

Wherein, ${\stackrel{\·}{x}}_i=x_{i+n}-x_{i-n},{\stackrel{\·}{y}}_i=y_{i+n}-y_{i-n},{\stackrel{\·\·}{y}}_i=y_{i+n}+y_{i-n}-2y_i;i=\mathrm{1,2},......N,$ N is sampling number, x _ifor the axial coordinate value that sampled point i is corresponding, y _ifor the radial coordinate value that sampled point i is corresponding; k _ifor the curvature after sampled point i distortion; N gets positive integer;

Step 3, analyze the curvature after the distortion that obtained by step 2, judge the angle point in solid of revolution profile data point, extract the unique point of solid of revolution;

Step 4, the unique point obtained by step 3, by solid of revolution Cutting section by section, obtain different straight-line segments or segment of curve.

Preferably, step one also comprises: judge that the sampling number obtained is according to whether there is gross error, and reject sampling number certificate corresponding to gross error; Specific practice is: setting threshold value ζ, threshold value ζ are solid of revolution radial direction border higher limit, reject y _ithe sampling number certificate of>=ζ, retains remaining sampling number certificate.

Preferably, in described step 3, angle point is curvature k _ivalue there is spike and be greater than the position of setting curvature threshold values.

Preferably, the shape exterior feature that axially spaced-apart is large can adopt large sampling interval, and the profile that axial spacing is little can adopt little sampling interval, sampling Δ x little 2 ~ 3 orders of magnitude compared with actual revolving parts spacing.

Beneficial effect:

(1) the present invention improves the method that existing curvature method extracts solid of revolution outlining characters point, the basis adopting equal interval sampling reduces difference that is axial and radial direction, make it possible to the revolving parts profile more more than comparatively dense, step, groove to the radial direction sudden change along axis and carry out feature point extraction, thus realize the segmentation to this part.

(2) can adopt different sampling interval according to the not similar shape exterior feature of revolving parts, the shape exterior feature that axially spaced-apart is large can adopt large sampling interval, and the profile that axial spacing is little can adopt little sampling interval.

Accompanying drawing explanation

Fig. 1 is the process flow diagram of two-dimentional solid of revolution cloud data feature point extraction.

Fig. 2 is the some cloud of an actual measurement Xray films of tested solid of revolution.

Fig. 3 is the profile cloud data curvature that prior art employing three point method obtains.

Fig. 4 is the profile cloud data curvature that the present invention adopts three point method to obtain.

Embodiment

The invention provides a kind of feature point extraction and automatic segmentation method of revolving parts profile, its core concept is: calculate in the method for curvature in prior art, adopt equal interval sampling and reduce axially with radial difference, realize the extraction that the revolving parts profile more more than comparatively dense, step, groove to the radial direction sudden change along axis carries out unique point.

To develop simultaneously embodiment below in conjunction with accompanying drawing, describe the present invention.

The feature point extraction of a kind of revolving parts profile provided by the invention and automatic segmentation method, concrete steps are as follows:

Step one, employing sensor continuous sweep solid of revolution carry out uniform sampling at equal intervals to revolution bodily form exterior feature, obtain the data of each sampled point of solid of revolution in real time, data are sent to industrial computer, if the set of the sampled point of sensor scan acquisition is: A _i=P _i(x _i, y _i), wherein, i=1,2 ... N, x _ifor the axial coordinate value that sampled point is corresponding, y _ifor the radial coordinate value that sampled point is corresponding, N is sampling number.Shape exterior feature refers to the outline line that three-dimensional body projects in some planes, and in the present invention, some planes were a plane of solid of revolution axis.

In prior art, along rotator shaft to the sampling of employing non-uniform spacing, sampling interval is uncertain, and the interval of the outline position sampling that solid of revolution axial spacing can be caused less is comparatively large, thus lost some unique points, makes out of true of sampling.Therefore, the present invention adopts equal interval sampling method, can collect the data of each sampled point of revolving body contour accurately, avoid the loss of some unique point caused by Non uniform sampling, then:

$\left\{\begin{array}{c}{x}_i-x_{i-n}=\mathrm{n\Δx}\\ x_{i+n}-x_i=\mathrm{n\Δx}\\ x_{i+n}-x_{i-n}=2\mathrm{n\Δx}\end{array}\right.---\left(1\right)$

Wherein, n gets positive integer, and two catastrophe point axially spaced-apart larger n values are larger, and interval less n value is little, such as, get 1,2,3, thus can have much suddenly change time, obtain the Curvature varying situation of more closing to reality, taken into account data volume simultaneously.Certainly when two catastrophe point axially spaced-aparts are larger, also can adopt less n value, but the data volume processed can be made to increase to some extent.Δ x is the axially spaced-apart between adjacent two sampled points.Revolving parts axially has a lot of sudden change, and as screwed part, parts classify can be straight-line segment, segment of curve or oblique line section by the point of every two sudden changes, and these sections are called as axis at intervals; The shape exterior feature that axially spaced-apart is large can adopt large sampling interval, and the profile that axial spacing is little can adopt little sampling interval, so namely ensure that the accuracy of sampling turn improves the efficiency of measurement; If the profile that axial spacing is little adopts large sampling interval, then can cause the loss of Partial Feature point.Rule of thumb Δ x general little 2 ~ 3 orders of magnitude compared with actual revolving parts spacing.

Further, judge whether the data of the sampled point obtained exist gross error, and reject the sampling number certificate causing gross error corresponding.Specific practice is: setting threshold value ζ, threshold value ζ are solid of revolution radial direction border higher limit, reject y _ithe sampling number of>=ζ is according to P _i, retain remaining sampling number certificate.

Step 2, to obtain the curvature that data acquisition 2n+1 method calculates each sampled point of solid of revolution according to step one, the curvature of each sampled point of solid of revolution obtained is calculated, obtains the curvature after the distortion of each sampled point of solid of revolution;

In this step, according to the sampling number certificate of a kind of reservation of step, 2n+1 point curvature method is adopted to calculate the curvature k of the sampling number certificate retained _i, 2n+1 point curvature method refers in calculating curvature process, is calculated the curvature of intermediate point by 2n+1 point:

$k_i=\frac{\left|{y_i}^{\′\′}\right|}{{(1+{y_i}^{\′2})}^{3/2}}---\left(2\right)$

Wherein, y _i' be first order derivative, y _i" be second derivative.

Because the interval calculating curvature is 2n+1 point, then

$\left\{\begin{array}{c}{y}^{\′}=\frac{y_{i+n}-y_{i-n}}{x_{i+n}-x_{i-n}}\\ y^{\′\′}=\frac{y_{i+n}^{\′}-y_{i-n}^{\′}}{x_{i+n}-x_{i-n}}\end{array}\right.---\left(3\right)$

Thus obtain curvature estimation formula and be:

$k_i=\frac{\frac{(y_{i+n}-y_i)(x_i-x_{i-n})-(y_i-y_{i-n})(x_{i+n}-x_i)}{(x_{i+n}-x_i)(x_i-x_{i-n})(x_{i+n}-x_{i-n})}}{{[1+{\left(\frac{y_{i+n}-y_{i-n}}{x_{i+n}-x_{i-n}}\right)}^2]}^{3/2}}---\left(4\right)$

Carry out approximate simplification to above-mentioned curvature formulations to obtain:

$k_i=\frac{(y_{i+n}+y_{i-n}-2y_i)(x_{i+n}-x_{i-n})-(y_{i+n}-y_{i-n})(x_{i+n}+x_{i-n}-2x_i)}{{[{(x_{i+n}-x_{i-n})}^2+{(y_{i+n}-y_{i-n})}^2]}^{3/2}}---\left(5\right)$

Formula (1) is substituted into formula (5), obtains:

$k_i=\frac{4\mathrm{n\Δx}(y_{i+n}+y_{i-n}-2y_i)}{{(4n^2\mathrm{\Δ}{x}^2+{(y_{i+n}-y_{i-n})}^2)}^{3/2}}---\left(6\right)$

As shown in Figure 3, be the curvature of all point positions that through type (6) calculates, visible, because part is more along the radial direction sudden change of axis, the curvature calculated is disorderly and unsystematic, cannot the position of judging characteristic point, causes to carry out segmentation to part.

During the data of the revolution bodily form that collects each sampled point wide, all error can be there is in x direction and y direction, the method of the equal interval sampling of our employing in the x-direction, reduce the impact of x deflection error on curvature, and the value of each sampled point is constantly fluctuating up and down in the y-direction, the impact of error on curvature is larger, in order to eliminate the error of each sampled point of y direction to the impact of curvature, can by increasing the impact of Δ x when calculating curvature, reduce the error of each sampled point of y direction to impact during calculating curvature, therefore the curvature that formula (6) calculates is out of shape, then obtaining the curvature after being out of shape is:

$k_i=\frac{{\stackrel{\·\·}{y}}_i}{{\stackrel{\·}{x}}_i^2-{\stackrel{\·}{y}}_i^2}---\left(7\right)$

Wherein, ${\stackrel{\·}{x}}_i=x_{i+n}-x_{i-n},{\stackrel{\·}{y}}_i=y_{i+n}-y_{i-n},{\stackrel{\·\·}{y}}_i=y_{i+n}+y_{i-n}-2y_i;i=\mathrm{1,2},......N,$ N is sampling number, x _ifor the axial coordinate value that sampled point i is corresponding; y _ifor the radial coordinate value that sampled point i is corresponding;

The curvature of each sampled point that employing formula (7) calculates is as schemed the result of (4).

Visible, 2n+1 point curvature method, can arrange different intervals according to the difformity of part and to count calculating curvature, applied widely.

Step 3, analyze the curvature after the distortion that obtained by step 2, judge the angle point of solid of revolution profile, extract the unique point of solid of revolution.Angle point refers to the position of intersecting point of straight line and straight line, straight line and circular arc, circular arc and circular arc, if curvature k _ivalue there is spike and be greater than the position of setting curvature threshold values, then illustrate that this location point is the angle point of revolution bodily form exterior feature, according to figure (4), find the kurtosis pulse of curvature, be the unique point of solid of revolution, and find the sampling number certificate of Feature point correspondence.

Step 3, by the unique point extracted by solid of revolution segmentation;

By the unique point that step 2 obtains, by solid of revolution Cutting section by section, just different straight-line segments or segment of curve can be obtained.The characteristic parameter of each section can be obtained according to segmentation for every section targetedly, as pitch, angle of inclination etc., complete feature point extraction and the automatic segmentation of solid of revolution profile.

Instance analysis:

The cloud data of light veil type sensor scan solid of revolution, it contains a large amount of grooves, step.If the sampling number N scanning this solid of revolution is 3569, sampling interval △ x=25um, obtain the radial displacement y that sampled point is corresponding _i, obtain the sampling number of actual measurement solid of revolution profile according to P _i(x _i, y _i), (i=1,2 ... 3569), choose the data mapping of front 2000 sampled points of actual measurement solid of revolution sampled point, as shown in Figure 2, the overall profile sampling number of solid of revolution is according to expanding by periodic law.

According to formula (7) to surveying the sampling number of solid of revolution profile according to P _i(x _i, y _i), (i=1,2 ... ..3569), ask its approximate curvature, obtain all curvature value k _i;

Select k _ithe sampling number of front 2000 points is according to mapping, and as shown in Figure 4, all curvature values of its solid of revolution are expanded by periodic law in Fig. 4.

Analysis chart 4 is known, and the curvature of solid of revolution there will be kurtosis pulse in the corner location of profile, milder at other straight-line segment.By searching Curvature varying situation, finding the position of the kurtosis pulse of curvature, and finding out the position of the revolving body contour of its correspondence, the unique point of getting required by being.After obtaining unique point, by solid of revolution Cutting section by section, just different straight-line segments can be obtained.Different characteristic parameters can be obtained according to different straight-line segment, thus obtain the characteristic parameter of whole solid of revolution.

In sum, these are only preferred embodiment of the present invention, be not intended to limit protection scope of the present invention.Within the spirit and principles in the present invention all, any amendment done, equivalent replacement, improvement etc., all should be included within protection scope of the present invention.

Claims (4)

1. the feature point extraction of revolving parts profile and an automatic segmentation method, it is characterized in that, concrete steps are as follows:

Step one, the continuous sweep revolution bodily form are wide, go forward side by side and are divided into the uniform sampling at equal intervals of Δ x in the ranks, obtain sampling number certificate in real time;

Step 2, according to obtaining sampling number certificate in step one, in 2n+1 curvature estimation method, utilizing the curvature of formula (7) to each sampled point of solid of revolution obtained to be out of shape, obtaining the curvature after the distortion of each sampled point of solid of revolution:

$k_i=\frac{{\stackrel{\·\·}{y}}_i}{{\stackrel{\·}{x}}_i^2-{\stackrel{\·}{y}}_i^2}---\left(7\right)$

Wherein, ${\stackrel{\·}{x}}_i=x_{i+n}-x_{i-n},{\stackrel{\·}{y}}_i=y_{i+n}-y_{i-n},{\stackrel{\·\·}{y}}_i=y_{i+n}+y_{i-n}-2y_i;$ I=1,2 ... N, N are sampling number, x _ifor the axial coordinate value that sampled point i is corresponding, y _ifor the radial coordinate value that sampled point i is corresponding; k _ifor the curvature after sampled point i distortion; N gets positive integer;

Step 3, analyze the curvature after the distortion that obtained by step 2, judge the angle point in solid of revolution profile data point, extract the unique point of solid of revolution;

Step 4, the unique point obtained by step 3, by solid of revolution Cutting section by section, obtain different straight-line segments or segment of curve.

2. the feature point extraction of a kind of revolving parts profile as claimed in claim 1 and automatic segmentation method, it is characterized in that, step one also comprises: judge that the sampling number obtained is according to whether there is gross error, and reject sampling number certificate corresponding to gross error; Specific practice is: setting threshold value ζ, threshold value ζ are solid of revolution radial direction border higher limit, reject y _ithe sampling number certificate of>=ζ, retains remaining sampling number certificate.

3. the feature point extraction of a kind of revolving parts profile as claimed in claim 1 and automatic segmentation method, it is characterized in that, in described step 3, angle point is curvature k _ivalue there is spike and be greater than the position of setting curvature threshold values.

4. the feature point extraction of a kind of revolving parts profile as claimed in claim 1 and automatic segmentation method, it is characterized in that, the shape exterior feature that axially spaced-apart is large can adopt large sampling interval, the profile that axial spacing is little can adopt little sampling interval, sampling Δ x little 2 ~ 3 orders of magnitude compared with actual revolving parts spacing.