CN103170778B - An Automatic Seam Tracking System - Google Patents

Abstract

The present invention discloses an automatic welding seam tracking system. In the system, a camera group (1) collects welding seam images, an image recognition device (3) obtains the welding seam images through an image acquisition card (2), searches and matches the current position of the welding seam center in the welding seam images using a self-template method, generates a control signal according to the matched current position of the welding seam center and outputs it to an automatic control device (4), and the automatic control device (4) adjusts the position of the welding gun arm according to the given control signal to move the welding gun to the welding seam center position. Since the system adopts the self-template method, the anti-interference ability of the recognition process is improved, thereby improving the recognition success rate.

Description

An Automatic Seam Tracking System

technical field

The invention belongs to the technical field of automatic welding seam tracking in welding automation, in particular to an automatic welding seam tracking system using an image recognition method.

Background technique

In the production of large steel structures (such as: oil and gas steel pipes, ship hulls, pressure vessels, etc.), automatic welding technology is widely used. In order to realize automatic welding, it is necessary to keep the relative position of the welding torch and the welding seam stable and accurate through the method of automatic welding seam tracking.

Existing seam automatic tracking methods include: profile tracking method, eddy current tracking method, electromagnetic tracking method, photoelectric tracking method, image tracking method and so on. The image tracking method is to use the camera to obtain the weld seam image in real time, use the theory and method of pattern recognition to find the weld seam in the current image, transmit the current position of the weld seam to the automatic control link, and then issue the command to adjust the position to realize the welding seam Automatic tracking of location.

Image tracking methods include two categories: one is based on structured light methods. The principle is to emit a narrow beam of light perpendicular to the direction of the welding seam, and keep the line of sight of the camera on a plane, which is perpendicular to the tangent plane including the center line of the welding seam, and the line of sight must be in a certain direction with the direction of the welding seam. Angle, so as to obtain an image reflecting the cross-sectional shape of the weld, which generally has two types: V-shape and U-shape. The V-shape or U-shape is then used as a feature for identification. Because the feature is simple and obvious, it is easy to identify, and it can ensure reliable tracking under the condition of guaranteed process conditions. However, this approach is susceptible to interference. When the welding seam is irregular, or when there is a large external light interference, when the obvious V-shape or U-shape cannot be presented, it will cause the recognition failure. The other category is based on unstructured light methods. The principle is to directly acquire the image of the weld, extract various features of the weld to form a feature vector, and then construct a complex recognition method to realize the search for the weld. The difficulty of this method lies in the extraction of feature vectors. The simplest feature extraction method is to take a window containing a weld in a standard weld image as a template, and perform window image matching in the current image. When the matching error Minimal, considered to find the weld. Due to the unevenness and cleanness of the weld surface, coupled with the interference of external light, it is actually difficult to obtain a suitable "standard weld image", and the template image taken is often quite different from the current image. Recognition failed.

Contents of the invention

Aiming at the problems existing in the above-mentioned two types of image tracking methods, the present invention proposes an automatic welding seam tracking system, which is also based on the image tracking method and uses its own template to match its own image to find the current position of the weld seam to improve Anti-interference ability of tracking process.

The technical means adopted in the present invention is: an automatic welding seam tracking system, which is characterized in that it includes: a camera group placed on the welding torch arm, an image acquisition card, an image display and recognition device, and an automatic control device;

The camera group collects the weld image, and the image display and recognition device acquires the weld image through the image acquisition card, uses its own template method to search and match the current position of the weld center in the weld image, and according to the matched current position of the weld center The control signal is generated and output to the automatic control device, and the automatic control device adjusts the position of the welding torch arm according to the given control signal, so as to move the welding torch to the center position of the welding seam.

The system of the present invention uses its own template method to realize the center positioning of the weld; specifically, in the image sequence obtained by the camera group, it is necessary to ensure that about half of the areas of the adjacent two frames of images in the forward direction of the weld overlap, that is, That is to say, the second half of the image in the previous frame will move to the first half of the image in the current frame; therefore, take a window image containing a certain length of weld in the second half of the image in the previous frame and use it as a template, in the current frame The first half of the frame image is matched, and when the best match is found, the image of the template itself is found, and the lateral offset value of the weld is determined; since the image in the template obtained in the previous image must be in the current It appears repeatedly in the first half image of a frame. Considering that the time interval between two consecutive frames is very small, the external environment will not change much, so what is done is to match itself with its own image, which improves the anti-interference of the recognition process ability, thereby improving the recognition success rate.

Description of drawings

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

Fig. 1 is the schematic diagram of the welding seam automatic tracking system of the present invention;

Figure 2 is a schematic diagram of extracting its own template for search and matching;

Fig. 3 is a data transmission diagram;

Figure 4 is a schematic diagram of the embodiment.

Detailed ways

As shown in FIG. 1 , the welding seam automatic tracking system of the present invention includes: a camera group 1 placed on a welding torch arm, an image acquisition card 2 , an image recognition device 3 , and an automatic control device 4 . The camera group 1 collects the weld seam image, and the image recognition device 3 obtains the weld seam image through the image acquisition card 2, uses its own template method to search and match the current position of the weld seam center in the weld seam image, and according to the current position of the weld seam center matched The principle of position and servo system control generates a control signal and outputs it to the automatic control device 4. The automatic control device 4 adjusts the position of the welding torch arm according to the given control signal to move the welding torch to the center of the weld.

The camera group can contain one or two cameras. Let the size of the field of view of a frame of weld image be F _v =a×b, a is the length perpendicular to the direction of weld advance, its coordinate direction is set to x, and b is the length along the weld seam. The length in the forward direction of the seam, and its coordinate direction is set as y. The principle of selecting one or two cameras is: set the moving speed of the weld seam S _w , and the duration of one frame of weld seam image T _V , when the distance between S _w , b and T _V satisfies the relation When , within the duration T _V of one frame of weld image, the weld only moves less than b/2 in the field of view, so only one camera can be used to provide enough image overlap for the self-template method. When S _w , b and T _V satisfy the relation When , it is necessary to use two cameras to collect two frames of adjacent weld images respectively, and in order to apply the self-template method, at least half of the adjacent two frames of weld images need to overlap. Therefore, the centers of the fields of view of the two cameras are at The distance on the direction of the weld is L _VF , then: L _VF ≤T _V ×S _w .

Among them, there are two standards for TV _V : under the PAL system, TV _V =40ms; under the NTS system, TV _V =30ms. The specific values of a, b, x and y are determined by the resolution and the TV system; the resolution is defined as a pixel in the weld image representing the size of the actual weldment, and its specific value is determined by the production process's requirements for control accuracy; set px is the resolution of the pixel in the x direction, and the maximum allowable control error is ±Er _c , then px≤Er _c /2, and the value range of x is x=1,...,N; let py be the resolution of the pixel in the y direction rate, the value of y is y=1,...,M. The commonly used image acquisition card is N=768 pixels, M=572 pixels; so a=N px, b=M py; the aspect ratio of the commonly used cameras is 4.8/3.6, 4.8/3.6≈768/572≈1.3, Then py=px/1.3; the way to determine the actual position on the spot is flexible, and the method recommended by the present invention is: calculate the actual values of a and b according to the above principles, place a standard length ruler under the camera, adjust Position the camera and focus the lens to observe the image on the display screen so that a fills the width of the entire video display as much as possible.

The process of searching and matching the current position of the weld seam center in the weld seam image by the image recognition device 3 using its own template method is described in detail below:

Step 1: Set three image sequence variables, denoted as Q _i , C _i and i=1,2,..., U, U=L/b, L is the total length of the weld passing through the field of view during the continuous welding process. Among them, when the camera group 1 is one camera, Q _i (i=1,...,U) is the image sequence obtained by sequentially arranging the continuous frame weld images collected by the camera group; when the camera group 1 is two When using a camera, the odd frames in the image sequence are weld images collected by one camera, and the even frames in the image sequence are weld images collected by another camera. Q _i (x, y) (i=1, ..., U) represent the sequence of pixel gray values in Q _i located at the (x, y) coordinate point in the weld image; C _i (i=1,..., U) is the cursor image sequence, C _i (x, y )(i=1,...,U) indicates the cursor gray value sequence of (x, y) coordinate points in the weld image, the shape of the cursor is determined by the designer, and a cross is commonly used; represents the sequence of images with the cursor attached, express The pixel gray value sequence located on the (x, y) coordinate points of the weld image in . For all x and y in a frame of image, execute The operation can superimpose the cursor sequence into the weld image frame, ⊕ means XOR operation. Hereinafter, β _i and β _i+1 are collectively referred to as the current frame and the next frame respectively, and β can represent Q, and C; It should be noted that from the perspective of the algorithm, the number of layers of the above three sequences can be set by the designer according to the working cycle of the welding system, that is, the maximum value U of i can have a large change. In actual implementation, if it is implemented with a computer, the camera is connected to the image card, and the computer reads data from the frame memory of the image card by setting multi-threading, and the above three sequences form a queue; if the self-made processor composed of DSP and FPGA , in order to save memory space, the above image sequence can only have two layers, that is, U=2, and use the traditional two-body cross structure frame accessor to realize the compression of memory space. When a camera is used, two memory The two-body crossover structure, when there are two cameras, use four frame access memories to form two pairs of two-body crossover structures (introduced in the literature of various memory structures);

Step 2: The operator performs initial positioning on the center of the weld: the operator moves the cursor to the center of the weld through a manual input device such as a keyboard and a mouse on the video display, and the image recognition device 3 reads the position indicated by the cursor, Use this position as the initial location of the weld center.

There are two rules for the operator to do the initial positioning: first, one operator locates, which is further divided into: 1. Operators with long-term operating experience can move the cursor to what they see by observing the image on the video display. 2. While moving the cursor, the image recognition device 3 sends the moving signal of the cursor to the automatic control device 4, and the automatic control device 4 controls the welding torch to align with the center of the weld seam; or put the video display of the image recognition device 3 on Near the welding torch, one operator adjusts the position of the welding torch to align it with the center of the weld, and at the same time observes and moves the cursor position on the video display to complete the initial cursor positioning; Positioning: The first operator directly measures the relative deviation between the welding torch and the center of the welding seam near the welding torch, and at the same time instructs the second operator in front of the display to move the cursor closer to the center of the welding seam. At the same time, instruct the second operator to stop moving the cursor at the center of the seam, the initial positioning of the cursor is completed, and the current cursor value is stored. Of course, the automatic tracking system starts after the initial cursor positioning. The initial positioning frame is used as the weld seam image Q _j of the first current frame in the subsequent algorithm.

Step 3: Take an area in the second half of the weld image Q _j of the current frame as its own template P _j . The center of P _j is set on the center of the weld, and the coordinates of this center are recorded in the intermediate variable C _o ; the principle of determining the size of P _j is: let d _px be the length of the template perpendicular to the direction of the weld, satisfying: d _px ≥d _w +2(d' _w -d _w ), d _w is the average width of the weld, d' _w is the maximum width of the weld; d _py is the length of the template along the direction of the weld, and its value is the weld speed stability, the present invention recommends

Step 4: use P _j in the first half of the next frame of weld image Q _j+1 , and use the search and matching algorithm to complete the positioning of the current position of the weld center in the Q _j+1 frame of weld image. The search refers to moving the own template P _j , and the matching refers to comparing the own template P _j with the sub-image at the searched position in the Q _j+1 frame weld image and recording the comparison result. The movement of the self-template P _j must be carried out in a certain path sequence, and different designers can adopt different path sequences; most of them adopt the method of first moving along the x direction and then moving one line in the y direction, and so on, forming a "zigzag" path sequence The order that x and y are exchanged can also be adopted; perhaps a special path sequence is designed by the designer, and the present invention takes the "zigzag" path sequence as an example; the concrete steps that step 4 includes again:

Step 1: Determine a search matching area in the middle area of the first half of the next frame of weld image Q _j+1 , set the starting point coordinates of the search matching area as (x _s0 , y _s0 ), and the end point coordinates (x _s0 +L _sx , y _s0 +L _sy ). The size of the search matching area is determined according to the following principles: the search matching area is defined as a rectangle, and its size should ensure that it has been searched during the search and match process. Continue to do several search matches, and then find the correct matching position through backtracking; let the length in the x direction be L _sx , in order to ensure that the search is done, it is required that L _sx ≥ d' _w +k _x1 Δd _wm +k _x2 , Δd _wm >0 is the maximum allowable weld seam deviation during the whole tracking process, k _x1 and k _x2 are set as adjustable parameters in the man-machine interface on the video display or a special adjustment key is set, and k _x1 is adjusted by the operator and the value of k _x2 ; k _x1 is set to compensate for the stability of the system. When the system is stable, k _x1 can be small, and when the system is not stable, k _x1 should take a large value; k _x2 is set to ensure over-searching. In theory, as long as k _x2 = py, but considering noise interference, k _x2 > py is required. In the actual system, k _x2 can be adjusted to a reasonable value through trial operation; y The length of the direction is L _sy , and the value of L _sy is related to the stability of the welding seam moving speed. It is suggested that L _sy ≥d _py +2ΔS _w T _V +3py, and ΔS _w ≥0 is the maximum deviation of the welding seam moving speed. This value It is a given in the design and operation of automatic welding systems.

Step 2: Set two intermediate array variables M ₁ (z) and M ₂ (z), where z=1,...,Z, respectively record the sum of matching operations, and the current matching on Q _j+1 frame weld image to the location coordinates.

Step 3: Place the upper left corner of the own template P _j on (x _s0 , y _s0 ), start from z=1, first move the own template P _j in the x direction and then the y direction with a step of one pixel, and do z=z+1 until Z to do the following matching operations in a loop:

M ₁ (z)=P _j (x,y)-Q _j+1 (x,y)

This operation generates a data sequence M ₁ (z), z=1,...,Z, the value of Z must ensure that the template P _j covers the entire search matching area during the moving process, that is, the right side of the template P _j must be ensured. The lower corner can finally reach the coordinates (x _s0 +L _sx , y _s0 +L _sy ), where L _sx and L _sy are the lengths in the x and y directions, respectively.

Among them, there are two kinds of step distances for the template P _j to move: one is the minimum step distance, that is, one pixel is used as a step distance unit; the other is a large step distance, and several pixels are used as a step distance unit. In the case that the operation time can meet the requirements, the minimum step distance is generally used. When the operation time is too long, a combination of a large step distance and the minimum step distance is required, which is called a combination algorithm. The combination algorithm is to use the large step distance for rough positioning first, and then use the minimum step distance for precise positioning. The length of the operation time is determined by the production process and the capability of the image processing device; when the operation of the image processing device can complete a search and matching operation within the interval between two control commands required by the production process with the minimum step distance, the minimum step is used. distance, otherwise the combined algorithm is used.

Step 4: Obtain the minimum value of the sequence M ₁ (z), z=1,..., Z, this minimum value corresponds to an exact match, and record the matching coordinate point corresponding to this minimum value, that is, M ₂ (z)=( x, y) gives the exact matching position on the weld image Q _j+1 , then the center of the own template P _j at this position is the center of the weld.

Step 5: Perform weld seam deviation calculation ΔC=C _o -M ₂ (z), change the position of the cursor in C _j+1 (x,y) according to the coordinate value recorded by ΔC, and at the same time send a control command to the control device to move the welding torch, Then do the calculation ${\hat{Q}}_{j+1}(x,\mathrm{the\; y})=Q_{j+1}(x,\mathrm{the\; y})\⊕C_{j+1}(x,\mathrm{the\; y}),$ will later ${\hat{Q}}_{j+1}(x,\mathrm{the\; y})$ send to the monitor.

Step 5: Check whether there is a stop command, if so, stop the above calculations and operations, otherwise go back to step 3, redo fetching templates, searching for matches, rewriting cursor position, sending Operations such as displaying and issuing control commands. There are two situations for issuing the stop command: (1) The system sends out a fault alarm, (2) The work is stopped manually.

Figure 2 shows a schematic diagram of extracting its own template for search and matching. The weld advances along the Y-axis direction, and deviations occur along the X-axis direction. Three coordinate planes are drawn in the figure, which are recorded as x(t _j-1 ), y(t _j-1 ), x(t _j ), y(t _j ), x(t _j+1 ), y(t _j+1 ), in space, these three planes are actually on one plane, but they are represented as three planes on the time axis (marked as t), and their time coordinates are marked as (t _j-1 ),(t _j ),(t _j+1 ). Three frames of weld images Q _j-1 , Q _j , Q _j+1 are drawn respectively on the three coordinate planes; the edge of the weld in the figure is divided into multiple segments in the y direction, and each segment is represented by a different line type. It is to indicate the corresponding relationship between the position changes during the moving process of the weld (further pointed out by the solid double arrow), from this corresponding relationship, it can be seen that about half of the images of the two adjacent frames of the three weld images overlap; P _j and P _j-1 denote templates in Q _j and Q _j-1 frames, respectively, The character symbol indicates the matching center, where P _j-1 is obtained in the second half of Q _j-1 , and the search and match is performed in the first half of Q _j , P _j is obtained in the second half of Q _j , and it is performed in the first half of Q _j+1 A search match, the variation relationship of this search match is indicated by the dotted double arrow. In the figure, x _j-1 , x _j , and x _j+1 represent the coordinates of the weld center respectively.

The image recognition device 3 of the present invention can be ordered or customized. At present, there are generally two types of structures, one is a general structure based on industrial computers, and the other is a special structure based on components such as DSP, FPGA, and single-chip microcomputers. . The specific structure of the dedicated structure is various, but the image acquisition and transmission flow are basically the same as the aforementioned general structure. Fig. 1 shows a schematic diagram of a general structure, in which only components related to the present invention are drawn.

In Figure 1, the ordered or customized image acquisition card with dual camera interfaces is inserted into the corresponding slot of the industrial computer. The image recognition device 3 further includes: a memory, a video display and a processor; the memory further includes a memory buffer 1 , a memory buffer 2 and a memory buffer 3 . Memory buffer 1 stores the weld seam image frame Q _2n+1 collected by one camera, memory buffer 2 stores the weld seam image frame Q _2n collected by another camera, memory buffer 3 stores the cursor image, and the graphics are updated by the algorithm Decision; the processor uses the self-template method to search and match the current position of the weld center in the weld image, generates a control signal according to the matched current position of the weld center, and outputs it to the automatic control device 4 . As pointed out in the foregoing description, the number of layers of memory buffers 1, 2, and 3 can be reduced to two layers. The above-mentioned relevant parameters can be placed in the rest of the storage space.

FIG. 3 shows the transmission paths of image data and control signals generated by recognition results.

Figure 4 is a diagram illustrating an embodiment. The image on the left is the weld seam image of the previous frame, and the image on the right is the image of the next adjacent frame. In the figure on the right, it can be seen that the weld seam image framed by the template in the lower half of the previous frame image has moved to the upper half in the next adjacent frame image. To indicate this, we artificially paste a white triangle. It can be seen from the figure that the center of the weld is shifted to the left by about 4mm (Note: This is just a demonstration picture, and there is no control feedback to adjust the deviation. In practice, it is impossible to allow the welds of two adjacent frames to deviate so much .)

Claims (4)

1. a seam tracking system, is characterized in that comprising: be placed in the shooting unit (1) on welding gun arm, image pick-up card (2), image display and recognition device (3), automaton (4);

Shooting unit (1) gathers weld image, image display and recognition device (3) obtain this weld image by image pick-up card (2) and are stored in image display and recognition device (3), in image display and recognition device (3), utilize self form method search in this weld image and mate Weld pipe mill current location, generate control signal according to the Weld pipe mill current location matched and export to automaton (4), automaton (4) is according to the position adjusting welding gun arm to control signal, so that welding gun is moved to Weld pipe mill position, image display and recognition device (3) utilize self form method to search in weld image and the step of mating Weld pipe mill current location comprises:

Step one: three image sequence variablees are set, are designated as Q _i, C _iwith wherein i=1,2 ..., U, U=L/b, L be in continuous welding process weld seam through the total length of visual field; Wherein, when unit (1) of making a video recording is a video camera, Q _i(i=1,2 ..., U) sequentially arrange the image sequence obtained successively for the successive frame weld image of this shooting unit collection; When unit (1) of making a video recording is two video cameras, the odd-numbered frame in this image sequence is the weld image of a camera acquisition, and the even frame of this image sequence is the weld image of another camera acquisition, uses Q _i(x, y) (i=1,2 ..., U) represent Q _iin be arranged in pixel grey scale value sequence in weld image coordinate points, (x, y), x=1,2 ..., N, y=1,2 ..., M, N=a/px, M=b/py, px and py are the image resolution ratios in x and y direction, also the size of i.e. pixel; C _ifor cursor glyph sequence, C _i(x, y) (i=1,2 ..., U) represent C _ithe cursor gray value sequence of (x, y) coordinate points in middle weld image; (i=1,2 ..., U) represent the image sequence adding cursor, represent in be positioned at pixel grey scale value sequence in weld image (x, y) coordinate points; For x and y all in a two field picture, perform computing is by cursor its superimposition in weld image frame, and ⊕ represents XOR;

Step 2: carry out initialization by operator's butt welded seam center and locate;

Step 3: at the weld image Q of present frame _jlatter half get a rectangular area as self form P _j, wherein Q _jbelong to image sequence Q _iin piece image, formula is expressed as Q _j∈ Q _i, j ∈ i, P _j(x, y) x, y represents template P _jin coordinate value, P _jcenter fix on Weld pipe mill, this centre coordinate is recorded in intermediate variable C _oin; P _jthe determination principle of size be: establish d _pxfor walking template length upwards perpendicular to weld seam, meet: d _px>=d _w+ 2 (d' _w-d _w), d _wfor the mean breadth of weld seam, d' _wfor the Breadth Maximum of weld seam, d _pyfor walking template length upwards along weld seam,

Step 4: use P _jat next frame weld image frame Q _j+1in first half, by use search and matching algorithm, complete butt welded seam picture frame Q _j+1the location of middle Weld pipe mill current location;

Step 5: checked whether and ceased and desisted order; If any, then stop above-mentioned computing and operation, otherwise get back to step 3, delivery plate of reforming, search mate, rewrite cursor position, send showing and send control command operation, sending two kinds of situations of having ceased and desisted order: (1) system sends fault alarm, and (2) people is for quitting work.

2. system according to claim 1, is characterized in that a frame weld image visual field is of a size of F _v=a × b, a is the length perpendicular to weld seam direction of advance, and its coordinate direction is decided to be x, and b is the length along weld seam direction of advance, and its coordinate direction is decided to be y, a frame weld image duration T _v, weld seam is at y direction translational speed S _w, when meeting relation time, shooting unit is a video camera, and now adjacent two two field pictures just have the overlap of more than 1/2; And work as time, also will ensure that adjacent two two field pictures have the overlap of more than 1/2, then require that shooting unit is 2 video cameras, these 2 video cameras gather the weld image of two consecutive frames respectively; The distance along weld seam trend at these two camera field of view centers is L _vF, require L _vF≤ T _v× S _w; Wherein, T under PAL-system _v=40ms, under NTS system, T _v=30ms.

3. system according to claim 1, is characterized in that step 4 also comprises:

Step 1: at Q _j+1the zone line of frame weld image the first half determines a search Matching band, if search

Rope Matching band is rectangle, and the coordinate of the upper left corner in picture frame is (x _s0, y _s0), lower right corner coordinate

(x _s0+ L _sx, y _s0+ L _sy), L _sxand L _sybe respectively the length in x and y direction;

Step 2: aray variable M in the middle of arranging two ₁(z) and M ₂(z), wherein, z=1 ..., Z, record matching operation result, and Q respectively _j+1the current position coordinates matched on frame weld image;

Step 3: by self form P _jthe upper left corner be placed on Q _j+1(the x of frame weld image _s0, y _s0) on, from z=1, first x direction again y direction with a pixel for step pitch moves self form P _j, often moving moves a step is z=z+1 until z=Z only goes cyclically to do following matching operation:

M ₁(z)＝P _j(x,y)-Q _j+1(x,y)

Operation result is recorded in M ₁z, in (), the value of Z will ensure self form P _jthe lower right corner can finally arrive coordinate (x _s0+ L _sx, y _s0+ L _sy), this computing produces a data sequence M ₁(z), z=1 ..., Z;

Step 4: try to achieve sequence M ₁(z), z=1 ..., the minimum of a value of Z, simultaneously the coupling coordinate points of record this minimum of a value corresponding, and use M ₂z () record, even M ₂(z)=(x, y), thus provide Q _j+1the position of accurate match on frame weld image, then self form P on this position _jcenter be exactly the center of weld seam;

Step 5: be weld seam deviation computing Δ C=C _o-M ₂z (), changes C according to the deviate that Δ C records _j+1in (x, y), the position of cursor sends the control command of mobile welding gun simultaneously to control device, then does computing afterwards will display is sent to show.

4. system according to claim 3, the size and the position that it is characterized in that searching in step 1 Matching band are determined by following principle: search Matching band meets L _sx>=d'+k _x1Δ d _wm+ k _x2and L _sy>=d _py+ 2 Δ S _wt _v+ 3py, Δ d _wm>0 is the maximum weld seam deviation allowed in whole tracing process, k _x1and k _x2for arrange adjustable parameter, its position in the central authorities of the first half of a two field picture, Δ S _w>=0 is the maximum deviation of weld movement speed.