A shield tunnel segment automatic assembling method and system based on feature extraction

Abstract

The application discloses a kind of shield machine segment automatic assembly system and method based on feature extraction, including image acquisition module, transmission module, host computer, first-order segment position calibration module, second-order segment position calibration module and PLC control module;The measurement of segment edge state is realized using structured light technology and the extraction of feature recess coordinates to complete the pose detection of shield machine segment, combined with the discrimination operation of host computer and the accurate control of PLC control module to realize the automation of single segment assembly.Compared with the prior art, the application has high measurement accuracy, high automation degree and stronger anti-interference ability, and has stronger applicability in shield construction environment.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of segment pose detection and automatic assembly in shield construction, and particularly relates to a segment automatic grabbing and automatic assembly device based on feature object recognition. BACKGROUND

[0002] In shield construction, segment grabbing and assembly is a very important link in construction, and at present, segment grabbing and assembly of most segments in China is completed by manual remote control segment assembly machine, so the speed and accuracy of segment assembly mainly depend on the proficiency of assembly hands. In large shield equipment, sometimes technical workers need to control segment assembly machine at 10m high to observe and adjust the segment pose. This makes the construction safety cannot be guaranteed. At the same time, due to the insufficient light conditions, high noise and humidity temperature and other complex conditions of tunnel construction environment, if there are problems such as segment damage or uneven gap between adjacent segments in the assembly process, the speed and quality of tunnel construction will be greatly affected. Therefore, with the increasing demand for shield construction and the increasing diameter of shield machine, the realization of segment assembly automation of shield machine has great significance.

[0003] The core of realizing assembly automation is segment pose detection and adjustment. At present, there are three commonly used pose detection methods: the first is point array method, which uses multiple ranging sensors to measure the height information of each point on the segment at the same time, and obtains the current pose by combining the mold data of the segment. The second is line array method, which uses line laser or other line array sensors to collect the edge information of the assembled segment and the segment to be assembled, and obtains the current pose state of the segment to be assembled. The third is face array method, also known as image method, which uses cameras and other visual sensors to collect feature information on the surface of the segment, and inverses the current pose state of the segment by calculating the coordinate changes of the feature points, so as to adjust the oil cylinder and realize the control of the segment.

[0004] The advantages and disadvantages of the above three methods are as follows: 1) the point array method uses the principle of laser triangulation, has the advantages of high precision, flexible installation position of sensors, simple hardware, etc., and the disadvantages are complex calibration and pose solving operation, poor anti-interference performance. 2) the line array method is relatively complex in hardware installation, and needs to be matched with light source and camera. The advantages are easy pose solving, good anti-interference performance and self-checking performance. 3) the visual method mainly uses image processing method to obtain the mark information of the segment, and has the smallest operation amount in pose solving. The main implementation principle is to use camera to collect the image on the surface of the segment, use image processing technology to extract the coordinate information of the feature object, and then inverse the pose of the segment, so as to adjust the movement amount of each oil cylinder and realize the splicing of the segment. The disadvantages are that it is easy to be affected by the environment, the recognition accuracy and detection degree of the feature object are not high when the segment is damaged or the surface information is polluted by tung oil. SUMMARY

[0005] In order to overcome the uneven illumination in the industrial environment, the surface information is missing, and the existing technical problems such as the low detection rate of pipe piece features, the present application proposes a kind of shield machine pipe piece automatic assembling method and system based on feature extraction, realize the pipe piece advanced ground calibration and automatic assembly based on structured light image enhancement.

[0006] The present application realizes the following technical solutions:

[0007] A kind of shield machine pipe piece automatic assembling method based on feature extraction, the method includes the following steps:

[0008] Step 1, when the pipe piece assembling machine transports the pipe piece to be spliced to the area to be spliced, open the parallel laser line group, and carry out the first-order pipe piece position calibration:

[0009] Step 2, calculate the end point spacing of parallel laser line, line spacing between parallel laser lines and tangent value of inclination angle, combined with calibration value, calculate the deviation of end point spacing of parallel laser line, line spacing between parallel laser lines and tangent value of inclination angle and calibration value;

[0010] Step 3, judge whether the deviation is within the threshold range, if not, i.e. one of the absolute deviations of end point spacing of parallel laser line, line spacing between parallel laser lines and tangent value of inclination angle and calibration value is greater than the threshold value, then go to step 4; if yes, i.e. the absolute deviation meets the threshold condition, then go to step 5;

[0011] Step 4, adjust the extension, pitch and yaw of the pipe piece assembling machine respectively until the absolute deviation is adjusted to the threshold range, and lock the pitch and yaw of the pipe piece assembling machine;

[0012] Step 5, lock the pitch and yaw of the pipe piece assembling machine, close the parallel laser line group, open all cameras and the attached projection equipment, carry out the second-order pipe piece position calibration, project the color grating with phase shift to the pipe piece groove direction, collect the grating image modulated by the groove height and transmit it to the upper computer;

[0013] Step 6, input the grating fringe into the improved Res-UNet network for processing, the network will extract image features and perform phase analysis, restore the depth information in the image, so as to detect the three-dimensional distribution of the groove in the image, restore the height distribution, calculate the midline coordinates, corner point coordinates and aspect ratio of the groove, combined with calibration value, calculate the deviation;

[0014] Step 7, judge whether the deviation is within the threshold range; if not, i.e. one of the absolute deviations of midline coordinates, corner point coordinates and aspect ratio of the groove is greater than the threshold value, then go to step 8; if yes, then go to step 9;

[0015] Step 8, adjust the extension, roll and rotation of the segment erector respectively until the deviation meets the threshold condition, confirm that the to-be-assembled segment and the assembled segment are parallel, and the detection and correction of the to-be-assembled segment assembly position are completed.

[0016] Step 9, lock the extension, roll and rotation of the cylinder, slide the cylinder, push the segment to complete the automatic assembly of the to-be-assembled segment.

[0017] A shield segment automatic assembly system based on feature extraction, wherein the shield segment automatic assembly system comprises:

[0018] The image acquisition module is configured to acquire parallel laser line images emitted to the assembled segment and the to-be-assembled segment.

[0019] The transmission module is configured to transmit the parallel laser line images to the upper computer.

[0020] The upper computer is configured to perform a first-order segment position adjustment and a second-order segment position adjustment.

[0021] The first-order segment position adjustment module is configured to adjust the position of the segment erector in extension, pitch and yaw using a plurality of parallel laser line groups.

[0022] The second-order segment position adjustment module is configured to adjust the position of the segment erector in extension, roll and rotation using an active vision camera group.

[0023] The PLC control module is configured to control the adjustment and automatic assembly of the segment according to the analysis and calculation of the upper computer.

[0024] Compared with the prior art, the present application has the following technical effects:

[0025] 1) The method of active vision avoids the interference of stray light in the assembly environment on measurement, and also avoids the error introduced by external markers. The combination of structured light and deep learning depth detection is used to adjust the cylinder one by one, which improves the degree of automation and assembly precision.

[0026] 2) High measurement accuracy, simple system and convenient calibration;

[0027] 3) Stronger anti-interference ability, strong applicability in shield construction environment. BRIEF DESCRIPTION OF DRAWINGS

[0028] Figure 1 The flowchart of the shield segment automatic assembly method based on feature extraction of the present application;

[0029] Figure 2 The parallel laser line diagram;

[0030] Figure 3 The improved Res-Unet network structure diagram;

[0031] Figure 4 The pipe piece movement position schematic view;

[0032] Figure 5 The shield pipe piece automatic assembling system module based on feature extraction of the application;

[0033] Figure 6 The shield pipe piece automatic assembling system embodiment based on feature extraction of the application;

[0034] Figure 7 The multi-line laser shooting image schematic view when the pitch, yaw and extension oil cylinders are not adjusted;

[0035] Figure 8 The groove schematic view;

[0036] Figure 9 The groove state schematic view when the rotation oil cylinder, yaw oil cylinder, extension oil cylinder and sliding oil cylinder are not adjusted;

[0037] The figure mark: 1, the first camera, 2, the second camera, 3, the third camera, 4, the first laser parallel line group, 5, the second laser parallel line group, 6, the pipe piece to be spliced, 7, the groove, 8, the spliced pipe piece. DETAILED DESCRIPTION

[0038] The specific embodiment of the application will be further described in detail below with reference to the accompanying drawings.

[0039] As shown in the figure, it is a shield pipe piece automatic assembling system and method based on feature extraction of the application, and the specific process is as follows: Figure 1

[0040] Step 1: when the pipe piece assembling machine transports the pipe piece to be spliced to the splicing area, the parallel laser line group is started, and the first-order pipe piece position is adjusted;

[0041] Step 2: the end point distance of the parallel laser line, the line distance between the parallel laser lines and the tangent value of the inclination angle are calculated, the deviation of the end point distance of the parallel laser line, the line distance between the parallel laser lines and the tangent value of the inclination angle from the calibration value is calculated, and the specific description of this step is as follows:

[0042] Firstly, the parallel laser line image collected by the upper computer is processed, and the pixel-level parallel laser line coordinates are separated through the methods such as mean filtering, morphological processing, edge detection and edge point filtering; as Figure 2 ​As shown, it is a schematic diagram of parallel laser lines. Affected by the spatial position of the pipe piece, the parallel laser lines in the field of view are divided into two groups, i.e. the upper parallel laser line group up_line of the already assembled pipe piece area and the lower parallel laser line group down_line of the pipe piece to be assembled area, the interval between the parallel laser lines in the same group is obtained, specifically the line interval of the upper parallel laser line group up_line and the lower parallel laser line group down_line (the interval between the top end points in the up_line group and the interval between the top end points in the down_line group), the line interval of the parallel laser lines between groups, specifically the end point interval between the upper parallel laser line group up_line and the lower parallel laser line group down_line (the interval between the tail end point of the up_line group and the top end point of the down_line group), and the tangent value of the inclination angle of the parallel laser lines;

[0043] Step 3, judge whether the deviation is within the threshold range, if not, i.e. one of the absolute deviations of the end point interval of the parallel laser lines, the line interval between the parallel laser lines and the tangent value of the inclination angle from the calibration value is greater than the threshold value, then go to step 4, adjust the position of the three oil cylinders of the pipe piece assembling machine according to the offset, repeat the measurement, and cycle the adjustment until the absolute deviation is adjusted to the threshold range, at this time, lock the two oil cylinders of the pitch and yaw;

[0044] Step 5: If yes, i.e. the absolute deviation meets the threshold condition, at this time, lock the two oil cylinders of the pitch and yaw, close the parallel laser line group, open all cameras and the attached projection equipment, and perform the second-order pipe piece position adjustment, project the color grating with phase shift to the pipe piece groove direction, the camera collects the grating image modulated by the groove height and transmits it to the upper computer;

[0045] Step 6: input the grating fringe into the improved Res-UNet network for processing, the network will extract image features and perform phase analysis, restore the depth information in the image, and thus restore the height distribution, calculate the midline coordinates, corner point coordinates and aspect ratio of the groove, and calculate the deviation combined with the calibration value; this step is specifically described as follows:

[0046] The processing of the grating image is specifically described as follows:

[0047] According to the RGB channel sampling, different color grating fringes are extracted and input into the Res-UNet network with channel attention mechanism and color attention mechanism as input, and the three-dimensional distribution of the groove in the image is detected; for example Figure 3The diagram shows the structure of the improved Res-UNet network. The improved Res-UNet network includes a feature extraction network and a feature fusion network: First, the feature extraction network encodes the image. The first feature extraction network consists of a quadruple downsampling module: two 3×3 convolutional layers with non-linear activation layers and a 2×2 max pooling layer. After the quadruple downsampling module, the image resolution is reduced, leaving only the main feature layers. Then, two attention mechanism networks are used, each consisting of a 2×2 max pooling layer and a SharedMLP layer, and a 2×2 average pooling layer and a SharedMLP layer, respectively. The results from these two attention mechanism networks are summed and then processed by a 3×3 convolutional layer with non-linear activation layers to output the filtered effective features. Finally, the feature fusion network is used for restoration. The second feature extraction network consists of a quadruple upsampling module: a feature concatenation layer, two 3×3 deconvolutional layers, and a 2×2 max pooling layer. Two non-linear activation layers (LeakyReLU) form an upsampling module. To prevent network degradation, the computational data of each downsampling module is added to the sampling result of the upsampling module as a residual (res). To achieve pixel-level segmentation, a 1×1 convolutional layer is added after obtaining the main feature layer to expand the number of channels. After passing through the feature fusion network, another 1×1 convolution layer is used to restore the number of channels to 1. Finally, the normalized feature layer yields the height information of objects within the range of the input image.

[0048] The centerline coordinates of the groove are determined based on the distribution of extreme points in the height information. The difference calculation is performed on the neighborhood of the centerline to extract the height change region, i.e., the difference extreme coordinates, to obtain the distribution of the groove corner points. The circumscribed rectangle is fitted to the groove distribution to calculate the horizontal and vertical coordinates of the centerline of each groove, the length-to-width ratio of the groove, and the size of the groove area. The deviation is calculated by comparing with the calibration value.

[0049] Step 7: Determine if the deviation is within the threshold range; if not, that is, if the absolute deviation of the groove centerline coordinate, corner coordinate, or aspect ratio is greater than the threshold, then proceed to step 8. Control the extension, sway, and rotation cylinders of the segment assembly machine according to the offset, repeat the measurement, and adjust cyclically until the deviation meets the threshold condition, confirm that the segment to be assembled and the assembled segment are strictly parallel, and the detection and correction of the assembly posture is completed.

[0050] Step 9: If so, lock the telescopic, lateral, and rotation cylinders, and use the sliding cylinder to push the segments together, automatically completing the splicing. After completion, add bolts to secure the segments.

[0051] like Figure 4 The diagram shows the pose of the tunnel segment during motion. The segment moves in a three-dimensional xyz coordinate system, forming a θ relationship with the xyz axes.x θ y and θ z

[0052] like Figure 5 The figure shown is a structural diagram of an automatic tunnel segment assembly system based on feature extraction according to the present invention.

[0053] like Figure 6 The figure shown is an embodiment of an automatic tunnel segment assembly system based on feature extraction according to the present invention.

[0054] The system includes an image acquisition module 100, a transmission module 200, a host computer 300, a parallel multi-parallel laser line group 600, a first-order segment position adjustment module 400, an active vision acquisition camera group 700, a second-order segment position adjustment module 500, and a PLC control module. The active vision acquisition camera group 700 includes a first camera 1, a second camera 2, and a third camera 3. The parallel multi-parallel laser line group 600 includes parallel laser line groups composed of a first laser parallel line group 4 and a second laser parallel line group 5. The first camera 1, the second camera 2, and the third camera 3, along with the first laser parallel line group 4 and the second laser parallel line group 5, are spaced apart and installed on the segment assembly machine, ensuring that the segment edges are within the measurement range at the start of assembly.

[0055] The image acquisition module 100 is used to acquire images of parallel laser lines emitted onto the assembled tube segments and the tube segments to be assembled.

[0056] The transmission module 200 is used to transmit the parallel laser line image to the host computer;

[0057] The host computer 300 is used to perform discrimination calculations for first-order segment position adjustment and second-order segment position adjustment;

[0058] The first-order segment position adjustment module 400 is used to adjust the position of the segment assembly machine in the three directions of extension, pitch and roll using the parallel multi-parallel laser line group 600.

[0059] The second-order segment position adjustment module 500 is used to adjust the position of the segment assembly machine in the three directions of extension, sway, and rotation by using the active vision acquisition camera group 700.

[0060] The PLC control module 800 is used to control the adjustment movement and automatic assembly of the tunnel segments based on the analysis and calculation of the host computer.

[0061] In practical applications, the system also includes a communication module.

[0062] In summary, the present application uses structured light technology to realize the measurement of the edge state of the segment and the extraction of the coordinates of the feature groove to complete the pose detection of the segment of the shield machine, and the automation of the single segment assembly is realized by combining the discrimination operation of the upper computer and the accurate control of the PLC. The system comprises an image acquisition module, a transmission module, an upper computer, a parallel multi-parallel laser line group, a first-order segment position adjustment module, and an active vision acquisition camera group and a second-order segment position adjustment module. The control main body is a six-degree-of-freedom shield machine segment assembly machine controlled by a hydraulic cylinder and a segment containing a groove; the contour and center line of the groove on the segment are extracted and recognized by using depth information, and the pose detection of the to-be-assembled segment is realized.

[0063] The method logic of the relevant image processing and motion control extracted by the parallel laser line group in step 2 of the present application is illustrated as follows based on the calculation principle of this step:

[0064] After the computer receives the image transmitted by the camera, the image is first converted to grayscale, then the image is transformed into a binary image using an adaptive threshold, and finally a 3*3 rectangular structured element is used to perform an opening operation on the binary image to remove burrs, isolated small points and small bridges between lasers. A parallel laser line array with clear edges is obtained.

[0065] Then the parallel laser line array is screened and classified. First, the edge of the image after operation is detected to find the edge width and pixel range of each parallel laser line, then each edge is read in a loop, and the bright spots with less than β1 pixel points and the edges with less than β2 linearity are filtered out, (β1 and β2 are threshold values set in the program, and the empirical reference values are β1 = 80 and β2 = 0.65). The median of the edge is taken as the laser center line, 22 parallel laser lines are obtained, and they are classified into the up_line group of the already assembled segment area and the down_line group of the to-be-assembled segment area. After classification, the same group of parallel laser lines is counted, and if the number of the upper and lower parallel laser lines is equal and greater than 10, the parallel laser line group is successfully recognized. Finally, the line spacing, the cotangent of the inclination angle, and the distance between the bottom end point of the up_line group and the top end point of the down_line group are calculated.

[0066] As shown in Figure 7 Fig. 1 is a schematic diagram of a multi-line laser shooting image of a segment assembly machine when the pitch, yaw and extension cylinders are not adjusted.

[0067] 1) Pitch cylinder detection and adjustment mode:

[0068] If the inclination of the upper and lower parallel laser lines is not equal, the pitch cylinder needs to be adjusted. If the cotangent value of the up_line group parallel laser line is greater than the cotangent value of the down_line group parallel laser line, and the difference is greater than the set threshold, it indicates that the segment has a rotation angle towards the direction of the segment to be spliced. At this time, the rotation cylinder needs to be adjusted to rotate towards the direction of the advancing shield. Conversely, if the cotangent of the up_line group parallel laser line is less than the cotangent of the down_line group parallel laser line, and the difference is greater than the set threshold, it indicates that the segment has a rotation angle towards the direction of the advancing shield. At this time, the rotation cylinder needs to be adjusted to rotate towards the direction of the spliced segment.

[0069] 2) Detection and adjustment mode of yawing cylinder:

[0070] If the distance difference of the end points of the parallel laser lines is increasing or decreasing, the yawing cylinder needs to be controlled to correct the error. When the distance difference is decreasing, the yawing cylinder operates clockwise, and when the distance difference is increasing, the yawing cylinder operates counterclockwise, until the distance difference of the spacing between the upper and lower parallel laser lines is less than 3mm.

[0071] 3) Detection and adjustment mode of telescopic cylinder:

[0072] If the line spacing is not uniform, the telescopic cylinder (also known as red and blue cylinder) needs to be adjusted to correct the error. When the line spacing of the upper and lower parallel laser lines is increasing, the left cylinder needs to be adjusted to extend, and when the line spacing of the parallel laser lines is decreasing, the right cylinder needs to be controlled to extend.

[0073] Repeat the above laser detection and cylinder adjustment operation until the deviation is within the error range.

[0074] In the environment of shield construction, there is a lot of stray light that interferes with the detection system. The use of laser projection can greatly improve the recognition speed and accuracy.

[0075] The groove feature point coordinates used in steps 5 and 6 of the present application are used for cylinder adjustment of the segment splicing machine. The calculation principle based on this step is illustrated as follows:

[0076] As shown in Figure 8 , it is a groove diagram. The groove 7 structure is named ABCD in counterclockwise order, and the middle line is assumed to be point E and point F.

[0077] (1) The color raster pattern projected by the projection device on the camera, where the RGB three colors are shifted by three channels compared to the previous channel.

[0078]

[0079]

[0080]

[0081] Where A(x,y) is the ambient light, B(x,y) is the uneven reflectivity of the object surface, f is the grating fringe period, which satisfies the Nyquist sampling theorem while ensuring that the grating pattern covers the entire period. When the grating pattern is projected onto the surface of the object to be measured, and the grating pattern is subjected to a three-dimensional surface with a distribution of h(x,y), an additional phase modulation caused by the height is added:

[0082]

[0083] Where λ is the equivalent wavelength, which is exactly equal to the height causing a 2π phase change. Ideally, the photographic equipment is collected and then separated by color channel in the computer to obtain the phase distribution. At this time, the modulation phase can be solved by the phase shift algorithm:

[0084]

[0085] The difference between the original phase and the phase can obtain the single-valued function related to the height. Thus, the phase is accurately solved.

[0086]

[0087] In reality, the problem of receiving ambient light interference and the color matching problem of the projector camera will reduce the accuracy, and the semantic segmentation neural network can improve the robustness of the detection algorithm while ensuring the accuracy. The steps are as follows: after the camera collects data, it is transmitted into the computer by the communication module. The computer extracts the color components from the captured image according to the channel sampling, and then transmits each color component as input into the improved Res-UNet network to detect the three-dimensional distribution of the grooves in the image.

[0088] After obtaining the spatial distribution of the feature groove, the minimum value of the height is found to find the groove centerline coordinates EF, and the height mutation area near the centerline is found to find the groove corner coordinates ABCD.

[0089] As shown in Figure 9 The groove state diagram of the rotating cylinder, the rolling cylinder, the telescopic cylinder and the sliding cylinder when the adjustment is not completed. The assembly process is to splice the pipe pieces to be assembled along the tunnel excavation direction to the assembled pipe pieces, so that each camera recognizes two grooves. Each camera recognizes a group of grooves, and the groove information to be processed during splicing is two groups of four. The grooves are numbered as 11, 12, 21, 22 in the order of left upper, right upper, left lower and right lower, and the groove state of the rotating cylinder, the rolling cylinder, the telescopic cylinder and the sliding cylinder when the adjustment is not completed.

[0090] 1) By comparing the distance difference of the two groups of grooves, the rotating attitude of the pipe piece is corrected.

[0091] Take coordinate E as an example, E 11 , E 12 , E 21 , E 22 The horizontal and vertical coordinates are represented as (XE 11 , YE 11 ), (XE 12 , YE 12 ), (XE 21 , YE 21 ), (XE 22 , YE 22 ), and α is a threshold value set in the program.

[0092] (XE 11 -XE 12 )–(XE 21 -XE 22 ) takes value ξ x When it is not in the interval [-α1, α1], the pipe splicing machine needs to adjust the displacement of the rotating cylinder to correct the rotating posture, as shown in Figure 9 a), at this time ξ x <-α1. The pipe splicing machine needs to control the pipe to rotate clockwise, and when the difference ξ x is greater than α1, the correction direction of the pipe is changed to counterclockwise rotation.

[0093] 2) By comparing the longitudinal axis coordinate difference of the center lines of the two groups of grooves, the rolling posture of the pipe is corrected.

[0094] YE 11 -YE 12 ≠0 and YE 21 -YE 22 ≠0, the displacement of the rolling cylinder needs to be adjusted, as shown in YE 11 -YE 12 <0, at this time YE 21 -YE 22 is also less than 0, as shown in Figure 9 b), at this time the rolling cylinder will be controlled to rotate counterclockwise until YE 11 -YE 12 takes value in [-α2, α2].

[0095] 3) By comparing the area difference of the two groups of grooves, the stretching posture of the pipe is corrected.

[0096] The area difference of the grooves 11 and 12 When the absolute value of the size difference of the two grooves on the left is greater than the set threshold, that is, ξ S <α5, the left red oil lever is contracted, as shown in Figure 9 c), when the size difference of the two grooves on the right is greater than the threshold, that is, ξ SThe height of the telescopic cylinder is adjusted by adjusting the height of the blue cylinder on the right side.

[0097] 4) The sliding posture of the segment is corrected by comparing the difference of the horizontal axis coordinates of the two groove center lines.

[0098] Repeat the above process until the position information of the groove is consistent with the calibration value or less than the threshold value, and it is considered that all postures except sliding are adjusted, the segment to be assembled is moved forward to the assembled segment by the sliding cylinder control of the assembling machine, and the segment to be assembled is placed on the assembled segment. Figure 9 d) When the camera judges that the groove interval is less than the threshold value XE 11 -XE 12 ∈[-α6,α6], the segment assembly is completed.

[0099] Finally, the jacking cylinder behind the shield cutterhead is extended to fix the segment, the suction cup of the assembling machine is pressurized, the segment is lowered, and the assembling machine returns to the standby area to wait for the next segment to be grabbed and assembled.

Claims

1. A shield tunnel segment automatic assembling method based on feature extraction, characterized in that, The method comprises the following steps: Step 1: When the pipe segment assembling machine transports the to-be-assembled pipe segment to the to-be-assembled area, the parallel laser line group is turned on to perform first-order pipe segment position adjustment: Step 2: Calculate the end point distance of the parallel laser line, the line distance between the parallel laser lines and the tangent value of the inclination angle, and combine the calibration value to calculate the deviation of the end point distance of the parallel laser line, the line distance between the parallel laser lines and the tangent value of the inclination angle from the calibration value; Step 3: Determine whether the deviation is within the threshold range. If not, that is, one of the absolute deviations of the end point distance of the parallel laser line, the line distance between the parallel laser lines and the tangent value of the inclination angle from the calibration value is greater than the threshold value, then go to step 4; if yes, that is, the absolute deviation meets the threshold condition, then go to step 5; Step 4: Adjust the extension, pitch and yaw of the pipe segment assembling machine respectively until the absolute deviation is adjusted to be within the threshold range, and lock the pitch and yaw of the pipe segment assembling machine; Step 5: Lock the pitch and yaw of the pipe segment assembling machine, turn off the parallel laser line group, turn on all cameras and the attached projection equipment, perform second-order pipe segment position adjustment, project color gratings with phase shift to the pipe segment groove direction, collect the grating image modulated by the groove height and transmit it to the upper computer; Step 6: Input the grating stripe into the improved Res-UNet network for processing. The network will extract image features and perform phase analysis to restore the depth information in the image, thereby detecting the three-dimensional distribution of the groove in the image, restoring the height distribution, calculating the centerline coordinates, corner point coordinates and aspect ratio of the groove, combining the calibration value to calculate the deviation; Step 7: Determine whether the deviation is within the threshold range. If not, that is, one of the absolute deviations of the centerline coordinates, corner point coordinates and aspect ratio of the groove is greater than the threshold value, then go to step 8; if yes, then go to step 9; Step 8: Adjust the extension, yaw and rotation of the pipe segment assembling machine respectively until the deviation meets the threshold condition, and confirm that the to-be-assembled pipe segment and the assembled pipe segment are parallel, and the detection and correction of the to-be-assembled pipe segment assembling position are completed; Step 9: Lock the extension, yaw and rotation cylinders, slide the cylinder, push the pipe segment to complete the automatic assembly of the to-be-assembled pipe segment.

2. The shield tunnel segment automatic assembling method based on feature extraction according to claim 1, characterized in that, The parallel laser lines in step 1 include upper and lower groups, which are the upper parallel laser line group up_line of the assembled pipe segment area and the lower parallel laser line group down_line of the to-be-assembled pipe segment area.

3. The shield tunnel segment automatic assembling method based on feature extraction according to claim 1, characterized in that, In step 6, the grating fringe is taken as the input of the improved Res-UNet network, which first encodes the image by using a feature extraction network, the first feature extraction network is composed of four down-sampling modules: two 3*3 convolution layers with nonlinear activation layers and one 2*2 max-pooling layer form a down-sampling module; after the four down-sampling modules, the main feature layer is left; at this time, two attention mechanism networks are used, the two attention mechanism networks are composed of a 2*2 max-pooling layer and a Shared MLP layer, and a 2*2 average-pooling layer and a Shared MLP layer, respectively; after the processing results of the two attention mechanism networks are added and then processed by a 3*3 convolution layer with a nonlinear activation layer, the effective features screened are output; then the feature fusion network is used for recovery processing, the second feature extraction network is composed of four up-sampling modules: one feature concatenation layer, two 3x3 deconvolution layers and one 2*2 max-pooling layer; two nonlinear activation layers form an up-sampling module; after the main feature layer is obtained, a 1*1 convolution layer is added to expand the channel number, and after the feature fusion network, the channel number is also restored to 1 by a 1*1 convolution layer, and finally the normalized feature layer obtains the height information of the objects in the range of the input image.

4. The shield tunnel segment automatic assembling method based on feature extraction according to claim 1, characterized in that, The pitch action of the segment erector is adjusted, and the adjustment process is as follows: If the cotangent value of the up_line group of parallel laser lines is greater than the cotangent value of the down_line group of parallel laser lines, and the difference is greater than the set threshold, it indicates that the segment has a rotation angle towards the direction of the segment to be spliced, and the rotation cylinder is adjusted to rotate towards the direction of the shield advancing; on the contrary, if the cotangent of the up_line group of parallel laser lines is less than the cotangent of the down_line group of parallel laser lines, and the difference is greater than the set threshold, it indicates that the segment has a rotation angle towards the direction of the shield advancing, and the rotation cylinder is adjusted to rotate towards the direction of the spliced segment.

5. The shield tunnel segment automatic assembling method based on feature extraction according to claim 1, characterized in that, The yaw action of the segment erector is adjusted, and the adjustment process is as follows: When the endpoint distance difference of the parallel laser lines decreases, the yaw of the segment erector acts clockwise, and when the endpoint distance difference of the parallel laser lines increases, the yaw of the segment erector acts counterclockwise, until the distance difference of the upper and lower parallel laser lines is less than 3mm.

6. The shield tunnel segment automatic assembling method based on feature extraction according to claim 1, characterized in that, The extension action of the segment erector is adjusted, and the adjustment process is as follows: When the distance between the upper and lower parallel laser lines increases, the segment erector is controlled to extend to the left side, and when the distance between the parallel laser lines decreases, the segment erector is controlled to extend to the right side.

7. A shield tunnel segment automatic assembling system based on feature extraction, which implements the shield tunnel segment automatic assembling method based on feature extraction according to any one of claims 1 to 6, characterized in that, In the automatic segment splicing system of the shield machine: An image acquisition module is configured to acquire images of the parallel laser lines emitted to the spliced segment and the segment to be spliced; A transmission module is configured to transmit the images of the parallel laser lines to a host computer; The host computer is configured to implement a first-order segment position adjustment and a second-order segment position adjustment; A first-order segment position adjustment module is configured to adjust the positions of the segment erector in extension, pitch and yaw by using a plurality of parallel laser line groups; A second-order segment position adjustment module is configured to adjust the positions of the segment erector in extension, roll and rotation by using a camera group of active vision. The PLC control module is used for controlling the adjustment movement and automatic assembly of the segment according to the analysis and calculation of the upper computer.

8. The shield tunnel segment automatic assembling system based on feature extraction according to claim 7, characterized in that, The active vision acquisition camera group comprises a first camera, a second camera and a third camera, and the parallel multi-parallel laser line group comprises a parallel laser line group composed of a first laser parallel line group and a second laser parallel line group.

9. The shield tunnel segment automatic assembling system based on feature extraction of claim 7, wherein, The first camera, the second camera and the third camera and the first laser parallel line group and the second laser parallel line group are arranged at intervals, and are respectively installed on the segment erector and the segment edges are within the range when the segment erector starts to assemble.

Images