CN104732553B - A kind of Feature Points Extraction based on many laser assisted targets - Google Patents

Abstract

A feature point extraction method based on multiple laser-assisted targets of the present invention belongs to the field of image processing and computer vision detection, and particularly relates to a method for processing image data in laser-assisted multi-view data splicing, specifically refers to a method based on multiple laser-assisted Target design and feature point extraction method. This method uses a laser array composed of non-contact multiple lasers to project laser light strips to construct a stitching target. In the target, each intersection point of the laser grid is used as a stitching feature point, and the target coordinate position is extracted using a high-precision target feature point extraction algorithm. The present invention adopts laser-assisted targets, which have better adaptability compared with sticky targets, and effectively improves the accuracy and efficiency of feature point extraction by first determining the region of interest and then accurately extracting feature points.

Description

A feature point extraction method based on multiple laser-assisted targets

technical field

The invention belongs to the field of image processing and computer vision detection, and in particular relates to a method for processing image data in laser-assisted multi-view data splicing, specifically a method for target design and feature point extraction based on multiple laser-assisted methods.

Background technique

In multi-view data stitching, the design of the target and the precise extraction of feature points have an important impact on the stitching accuracy. At present, the commonly used splicing targets are roughly divided into three types: sticking type, placement type, and projection type. The sticking type and placement type are easy to block the shape information of the measured object, and the on-site adjustment is inconvenient, and the adaptability to harsh environments is poor. However, the commonly used projective splicing targets often have the problems of difficulty in feature point extraction and low precision. The feature points of the target are used as the feature points of the final stitching. The traditional feature points generally have a specific shape distribution. Taking the ordinary circular spot as an example, because there is no specific grayscale distribution, the calibration is only calculated by fitting the feature edge The location of feature points inevitably has errors. The cost of improving the positioning accuracy of feature points simply by improving the resolution of the hardware is quite huge, so it is necessary to consider the use of high-precision feature point extraction algorithms to determine the coordinate positions of feature points.

In July 2006, Sun Junhua and others published the article "Visual measurement data splicing method based on planar baseline target" in the journal "Journal of Mechanical Engineering". As the feature point of splicing, the target has high precision, but in the case of partial occlusion of the measured object, due to the uncertainty of the occlusion situation, the corner points of the grid often have mismatching situations, and the target cannot be measured with It needs to be adjusted flexibly, and it has poor adaptability to harsh industrial and mining conditions such as high temperature. In April 2009, Liu Xiaoli and others published the article "Global Matching Method of Depth Data Using Marker Points" in the journal "Acta Optics Sinica", which proposed a method of pasting white circular marker points in the public field of view as stitching targets. Remove image noise through Gaussian filtering, use edge detection operator to achieve rough edge positioning, then further precisely position ellipse, and finally perform least square fitting on edge feature points to obtain sub-pixel positioning of the center of the circle, this method has relatively cumbersome pasting And clearing the mark work, the point cloud of the sticking mark part is missing, which is easy to damage the surface of the measured object.

Contents of the invention

The technical problem to be solved by the present invention is to overcome the defects of the prior art, and to invent a multi-laser-assisted target, which uses a line laser array to obtain a laser grid pattern, and uses the intersection of horizontal and vertical light strips as the splicing feature point. The non-contact laser projection method constructs the mosaic target, and uses the high-precision target feature point extraction algorithm to extract the target coordinate position to improve the accuracy and stability of the target feature point extraction.

The technical solution adopted by the present invention is a feature point extraction method based on multiple laser-assisted targets, which is characterized in that the method uses a laser array composed of non-contact multiple lasers to project laser light strips to construct a spliced target. In the target, the laser network Each intersection point of the grid is used as a splicing feature point, and the target coordinate position is extracted by using a high-precision target feature point extraction algorithm; first, the initial position of the feature point is obtained through an erosion operation, and then the initial position is used to determine the region of interest. In the area, the iterative reweighted least squares method is used to extract the coordinates of multiple points on the horizontal and vertical light stripes at fixed intervals by using the invented laser light stripe model, and two fitting curves of the horizontal and vertical light stripes are obtained, and the intersection points are used as The precise position of the feature point; the specific steps of the extraction method are as follows:

Step 1: Based on laser-assisted target design

The present invention adopts a plurality of line lasers to form a laser array, projects laser light bars onto the surface of the object to be measured, constructs a laser grid as a laser auxiliary target, and then shoots target images; in order to ensure the robustness of laser light bar extraction, By adjusting the camera exposure parameters to obtain a partially overexposed laser light strip image, the laser light strip spacing can be flexibly adjusted by changing the position of the laser; each grid intersection formed by the laser-assisted target is used as a stitching feature point;

Step 2: Laser Light Bar Model

For the laser light stripe projected in step 1, establish a partially overexposed laser light stripe model:

Among them, u is the pixel coordinate of the normal direction of the laser light strip, n is the number of items, a _i , b _i , and c _i are the coefficients of each item; the laser light strip model can be written as:

According to the gray level distribution in the cross-sectional direction of the laser light strip, the local extremum in the saturation area of the light strip is defined as the center point of the light strip;

Step 3: Determination of the initial position of feature points and the region of interest

The target image obtained in step 1 is preprocessed by median filter. The preprocessed image not only removes image impulse noise, but also retains useful detail information to a certain extent; performs binarization and corrosion operations on the image, Reasonably set the erosion operation threshold, obtain the isolated connected areas of each grid intersection, and use the gray center of the connected area as the initial position of the feature point Among them, i, j respectively refer to the number of rows and columns of feature points in the laser grid, i=1,2...5, j=1,2...5, u _ij and v _ij respectively refer to the corresponding initial positions in The pixel coordinates of the u-axis and v-axis in the image.

The region of interest is a circular region with the initial position of the feature point as the center and r _ROI as the pixel radius; quickly obtaining the rough position of the feature point in the laser grid image and identifying the region of interest can effectively reduce the search area and improve Feature point extraction efficiency;

Step 4: Determine the precise coordinates of the feature points

In the region of interest identified in step 3, the center point of the laser light stripe is calculated according to the locally saturated laser light stripe model proposed in step 2; for the laser light stripe in the vertical direction, v represents the search position with a fixed pixel interval d, which is the matching Light stripe information, 10 center coordinates are extracted from each light stripe; the same v-axis v _k is used in the region of interest, and a series of laser light stripe data on the section can be obtained by scanning, and these original data are processed by a saturated laser light stripe model match, x _k is the u-axis coordinate of the central point; the central point of the light bar is written as (x _k , v _k ), then a point set V={V _k (x _k , v _k )} including all central points is obtained, where k=1,2...10; also get the central point set U={U _k (u _k ,y _k )} of the horizontal light bar, where (u _k ,y _k ) is the kth fixed u axis u _k Center point, k=1,2...10;

Use the vertical and horizontal center point sets of the laser light strips to fit the curve respectively, and use the intersection point of the two laser light strip fitting curves as the precise coordinates of the feature points; p(i,j) is defined as the intersection of two fitting laser lines, where i=1,2...5, j=1,2...5; similarly, the precise position of each feature point on the projected target can be obtained.

The beneficial effect of the present invention is that the use of laser-assisted targets has better adaptability compared with sticky targets, and the accuracy and efficiency of feature point extraction are effectively improved by first determining the region of interest and then accurately extracting feature points.

Description of drawings

Figure 1 shows the projection pattern based on multiple laser-assisted targets.

Fig. 2 is the gray scale distribution curve diagram of four typical partial overexposure laser light stripe models, wherein, the axis of abscissa u (pixel), the axis of ordinate is the gray value corresponding to the light stripe, (a), (b) And (c) is the gray distribution curve of the light bar with only one peak, (d) is the gray distribution curve of the light bar with three extreme values, and a is the center position of the extraction point.

Fig. 3 is a map of feature point extraction in the region of interest, where p ⁰ (i, j) represents the initial position of the feature point, and ROI is the region of interest determined by the initial position.

Fig. 4 shows the precise feature points p(i, j) obtained by curve fitting in the region of interest.

Figure 5 is a graph of the extraction results of feature points based on laser-assisted targets, and the left image is an enlarged image of the extraction results of the feature points p(1,1) in the first row and first column.

detailed description

The specific implementation manner of the present invention will be described in detail below in conjunction with the technical scheme and the accompanying drawings. The present invention is based on multiple laser-assisted target and feature point extraction methods. In the embodiment, 10 line lasers are used to project a 5×5 laser grid on the surface of the measured object, and the intersection of horizontal and vertical light strips is used as the splicing feature. Points, after taking the image, preprocessing is performed to obtain the initial position of the feature point, and then the region of interest is determined, and the precise coordinates of the feature point are further obtained.

Step 1: Based on laser-assisted target design

The present invention uses ten line lasers to form a laser array, projects laser light strips onto the surface of the measured object, constructs a 5×5 laser grid as a laser auxiliary target, and forms 25 grid intersections as splicing feature points. The wavelength of the laser light bar is 650nm, and the image of the laser-assisted target is taken, as shown in Figure 1.

Step 2: Laser Light Bar Model

For the laser light stripe projected in step 1, a partially overexposed laser light stripe model is established. In accompanying drawing 2, show the gray scale distribution curve figure of four kinds of typical partial overexposure laser light bar models, wherein, the abscissa is the pixel position of the light bar, and the ordinate is the corresponding gray value of the light bar, (a) , (b) and (c) are gray distribution curves with only one peak, and (d) are gray distribution curves with three extreme values.

Using formula (1):

Establish the laser light bar model, where u is the pixel coordinate of the normal direction of the laser light bar, n is the number of items, and a _i , b _i , _ci are the coefficients of each item. Due to the limitation of the width of the light bar in the image, n is set to 2, and there are 6 parameters in the fitting function: a ₁ , b ₁ , c ₁ , a ₂ , b ₂ , c ₂ ; the derivative of the model is as the above formula (2) shown. Using MATLAB to obtain the gray distribution map of the laser light stripe, according to the gray distribution of the laser light stripe, the local extremum of the saturation area of the light stripe is defined as the center point of the light stripe. If there is only one peak, the corresponding position of n is the center point. If there are 3 extrema in the saturation region, the extremum in the middle of the two extrema with the characteristic of the second derivative is the center point.

Step 3: Determination of the initial position of feature points and the region of interest

The target image obtained in step 1 is preprocessed by median filter. The preprocessed image not only removes image impulse noise, but also retains useful detail information to a certain extent. Perform binarization and erosion operations on the image, reasonably set the threshold of the erosion operation, obtain 25 isolated connected areas, and use MATLAB to obtain the gray center of the connected area as the initial position of the feature point where i=1,2...5, j=1,2...5.

The region of interest is a circular region with the initial position of the feature point as the center and r _ROI as the pixel radius. Attached Figure 3 is a map of the region of interest determined by the initial position. According to the experimental site image, set the r _ROI to 80 pixels. Quickly obtaining the rough position of feature points in the laser grid image and identifying the region of interest can effectively reduce the search area and improve the efficiency of feature point extraction.

Step 4: Determine the precise coordinates of the feature points

In the region of interest identified in step 3, the center point of the laser light stripe can be calculated by the locally saturated laser light stripe model proposed in step 2. vertical laser beams, represents the search position with a fixed pixel interval d, where d is set to 4, and 10 center coordinates are extracted for each light bar. Using the same v-axis v _k in the region of interest, a series of laser light stripe data on the section can be obtained by scanning. These original data can be matched with a saturated laser light stripe model, and x _k is the u-axis coordinate of the center point. The central point of the light bar is written as (x _k , v _k ), then a point set V={V _k (x _k , v _k )} including all central points can be obtained. Similarly, the center point set U={U _k (u _k ,y _k )} of the horizontal light bar can be obtained, where (u _k ,y _k ) is the center point of the kth fixed u-axis u _k , k =1,2...10.

After obtaining the center point of the light bar, in order to eliminate outliers, the technique based on iterative reweighted least squares method is adopted. Therefore, the precise position p(i,j) of the feature point is defined as the intersection point of two fitted laser lines, where i=1,2...5, j=1,2...5. Accompanying drawing 5 is the graph based on laser-assisted target feature point extraction results, p(1,1) is the feature point extracted from the position of the first row and first column in the laser grid. In addition, if the surface on which the laser is projected is not a plane but a free-form surface, the two laser beam curves can be fitted by polynomial equations, and the intersection point can also be calculated.

The present invention designs a laser-assisted target, uses ten line lasers to form a laser array, projects a 5×5 laser grid, and uses 25 grid intersections as splicing feature points. Firstly, coarse positioning is used to obtain the region of interest and then The method of accurately finding feature points realizes the efficient and high-precision extraction of feature points.

Claims (1)

1. a kind of Feature Points Extraction based on many laser assisted targets, it is characterized in that, this method uses many of noncontact In the laser array projection laser striation construction splicing target that laser is constituted, target, each intersection point of laser network is made For the characteristic point of splicing, target co-ordinates position is extracted using high-precision target feature point extraction algorithm；Pass through etching operation first The initial position of characteristic point is obtained, then area-of-interest is determined using initial position, finally in the area-of-interest identified It is interior, using iteration weight weighted least-squares method technology, transverse and longitudinal striation is extracted with fixed intervals using the laser striation model of invention The coordinate of upper multiple points, is obtained two matched curves of transverse and longitudinal striation, the exact position of characteristic point is used as using its intersection point；Extraction side Method is comprised the following steps that：

Step 1：Based on laser assisted drone design

The present invention constitutes laser array using many laser line generators, and laser striation is projected to the surface of measured object, constructed One laser network shoots target image afterwards as laser assisted target；To ensure the robustness of laser striation extraction, pass through Regulation camera exposure parameters obtain the laser optical strip image of local overexposure, and laser light stripe pitch is by changing the position of laser It is adjusted flexibly；Laser assisted target formation each grid intersection point as splicing characteristic point；

Step 2：Laser striation model

For the laser striation projected in step 1, a kind of laser striation model is set up：

$f\left(u\right)=\underset{i=1}{\overset{n}{\Σ}}{a}_{i}sin(b_{i}u+c_i)---\left(1\right)$

Wherein, u is the pixel coordinate of laser striation normal direction, and n is item number, a_i,b_i,c_iIt is the coefficient of each single item；Laser striation model To being denoted as after u derivations：

$\left\{\begin{array}{c}{f}^{\′}\left(u\right)=\frac{df}{du}=\underset{i=1}{\overset{n}{\Σ}}{a}_i{b}_i\cos (b_{i}u+c_i)\\ f^{\′\′}\left(u\right)=\frac{d^{2}f}{{\mathrm{du}}^2}=\underset{i=1}{\overset{n}{\Σ}}-a_i{b_i}^2\cos (b_{i}u+c_i)\end{array}\right.---\left(2\right)$

According to the intensity profile of laser striation cross-wise direction, the local extremum of striation zone of saturation is defined as to the center of striation Point；

Step 3：The determination of characteristic point initial position and area-of-interest

The target image that step 1 is obtained is pre-processed using medium filtering to image, and pretreated image is not only removed Image impulse noise, and the detailed information remained with；Binary conversion treatment and etching operation are carried out to image, rationally sets rotten Operational threshold is lost, the isolated connected region of each grid intersection point is obtained, it is first using the gray scale center of connected region as characteristic point Beginning positionWherein, i, j refer to line number and columns of the characteristic point in laser network respectively, i=1, 2 ... 5, j=1,2 ... 5, u_ijAnd v_ijRefer to the pixel coordinate of each corresponding initial position u axles and v axles in the picture respectively；

Area-of-interest is the r using characteristic point initial position as the center of circle_ROIFor the border circular areas of pixel radius；In laser network image The rough position of middle quick obtaining characteristic point, and identify area-of-interest and can effectively reduce region of search, improve characteristic point Extraction efficiency；

Step 4：Determine characteristic point accurate coordinates

In the area-of-interest that step 3 is identified, the laser striation model meter that laser striation central point is proposed according to step 2 Calculate；The laser striation of vertical direction, v represents the searching position with fixed pixel interval d, for matching striation information, every light Bar extracts 10 centre coordinates；Identical v axles v is used in area-of-interest_k, a series of laser on sections is obtained by scanning Striation data, these initial data are matched with laser striation model, then obtain including the point set V={ V of all central points_k(x_k, v_k), wherein, (x_k,v_k) it is k-th of fixation v axles v_kCentral point, k=1,2 ... 10；It is similarly obtained the center of horizontal striation Point set U={ U_k(u_k,y_k), wherein, (u_k,y_k) it is k-th of fixation u axles u_kCentral point, k=1,2 ... 10；

Vertical, the laser striation central point collection matched curve of horizontal direction is utilized respectively, with the friendship of two laser striation matched curves Put the accurate coordinates as characteristic point；Using based on iteration weight weighted least-squares method technology, by characteristic point exact position p (i, J) intersection point of two fitting laser straight lines, wherein i=1,2 ... 5, j=1,2 ... 5 are defined as；Equally, projection target puts on each spy The exact position levied a little can be obtained.