CN113657193A - Segment damage detection method and system based on computer vision and shield machine - Google Patents
Segment damage detection method and system based on computer vision and shield machine Download PDFInfo
- Publication number
- CN113657193A CN113657193A CN202110850769.2A CN202110850769A CN113657193A CN 113657193 A CN113657193 A CN 113657193A CN 202110850769 A CN202110850769 A CN 202110850769A CN 113657193 A CN113657193 A CN 113657193A
- Authority
- CN
- China
- Prior art keywords
- segment
- damage
- duct piece
- computer vision
- image
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
- G06F18/214—Generating training patterns; Bootstrap methods, e.g. bagging or boosting
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/24—Classification techniques
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/045—Combinations of networks
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- Data Mining & Analysis (AREA)
- Life Sciences & Earth Sciences (AREA)
- Artificial Intelligence (AREA)
- General Physics & Mathematics (AREA)
- General Engineering & Computer Science (AREA)
- Evolutionary Computation (AREA)
- Bioinformatics & Computational Biology (AREA)
- Computational Linguistics (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Health & Medical Sciences (AREA)
- Biomedical Technology (AREA)
- Biophysics (AREA)
- Evolutionary Biology (AREA)
- General Health & Medical Sciences (AREA)
- Molecular Biology (AREA)
- Computing Systems (AREA)
- Mathematical Physics (AREA)
- Software Systems (AREA)
- Investigating Materials By The Use Of Optical Means Adapted For Particular Applications (AREA)
Abstract
The invention relates to a duct piece damage detection method and system based on computer vision and a shield machine, wherein the method comprises the steps of shooting a duct piece transportation and hoisting video and the actual weight of a duct piece in a construction site in real time, inputting the duct piece video into a trained duct piece damage model for damage detection to determine damage classification, constructing the duct piece damage model by a convolutional neural network, comparing the actual weight with the standard weight of the duct piece to obtain a comparison result, determining whether the duct piece needs to be repaired or replaced based on the damage classification and the comparison result, and if so, alarming and reminding. According to the invention, the quality of the construction tunnel can be effectively guaranteed, and the potential safety hazard is reduced.
Description
Technical Field
The invention belongs to the technical field of detection of a prepared segment damage condition when a shield machine works, and particularly relates to a segment damage detection method and system based on computer vision and a shield machine.
Background
In the current construction of underground tunnels, the tunnel construction process which is widely applied is a shield tunnel construction process, and has the advantages of high safety, high efficiency, economy, applicability and the like. The main principle of the shield method is that a shield machine cuts rock mass to complete excavation during underground tunneling, segments are spliced at the tail of a shield and are formed by grouting, and accordingly construction of the whole tunnel is completed in a reciprocating mode. The problems of a large amount of damage, cracks, pits and the like can be caused due to improper operation and quality reasons in the processes of long-distance transportation, loading and unloading, well descending, assembly and the like of the duct piece from a factory to a construction site, and workers in site construction of the problems generally choose to ignore the problems. When the seriously damaged duct pieces are used for assembling the tunnel, hidden dangers can be buried for the quality and the safety of the tunnel, so that some tunnels which are finished and used at present have the events of water leakage, breakage, even collapse and the like.
The chinese patent application with application publication number CN110415241A proposes a method for detecting the surface quality of a concrete member based on computer vision, so as to detect the surface of the concrete member, but this application only detects the damage of the surface, if there is a cavity inside, the problem that the weight does not reach the standard cannot be detected, which may cause a safety accident and cannot effectively ensure the quality of the construction tunnel.
Disclosure of Invention
The invention provides a segment damage detection method and system based on computer vision and a shield machine, which are used for solving the problem that the quality of a constructed tunnel cannot be effectively ensured in the prior art.
In order to solve the technical problem, the invention provides a segment damage detection method based on computer vision, which comprises the steps of collecting a segment video and the actual weight of a segment in real time, inputting the segment video into a trained segment damage model for damage detection to determine damage classification, constructing the segment damage model by a convolutional neural network, comparing the actual weight with the standard weight of the segment to obtain a comparison result, determining whether the segment needs to be repaired or replaced based on the damage classification and the comparison result, and if so, alarming and reminding.
The beneficial effects of the above technical scheme are: before the section of jurisdiction is assembled in the job site preparation, the section of jurisdiction video of utilizing the section of jurisdiction damage model that has trained to gather in real time at the job site carries out the damage detection in order to confirm in real time whether there are damage, crack, pit scheduling problem in section of jurisdiction surface, simultaneously through measuring the actual weight of section of jurisdiction, combines the comparison of damaged classification and actual weight and standard weight, tells the on-the-spot constructor whether this section of jurisdiction needs to be repaired or changed through the mode of warning. From this, can realize the real-time damage detection to the job site section of jurisdiction, and if there is the hole scheduling problem in its inside, can in time discover the problem that weight is not up to standard to ensure construction tunnel's quality effectively, reduce the production of potential safety hazard.
Further, in order to obtain a segment damage model capable of detecting segment damage types, the invention provides a segment damage detection method based on computer vision, and the segment damage model is trained by adopting a segment data set, wherein the segment data set is obtained by preprocessing a normal segment image and segment images of various damage types, the normal segment image comprises normal images of various angles of the segment, and the segment images of various damage types comprise full images of the damaged segment and local images of the damaged part.
Further, in order to reduce the influence of the external environment on the detection effect of the trained model and better improve the accuracy of the trained segment damage model, the invention provides a segment damage detection method based on computer vision, and the method further comprises the steps of preprocessing, including expansion and expansion of all collected images and parameter normalization processing.
Further, in order to obtain a segment damage model with better detection performance, the invention provides a segment damage detection method based on computer vision, further comprising that the convolutional neural network is a yolov3 algorithm model, the yolov3 algorithm model comprises an input layer, a Darknet layer and a yolo output layer, the input layer adjusts tensor on input images to obtain first intermediate images, the Darknet layer receives the first intermediate images, performs feature extraction on the first intermediate images to output prediction results, and the yolo output layer receives the prediction results and performs position regression and classification.
Further, in order to obtain the damage condition of the surface of the segment, the invention provides a segment damage detection method based on computer vision, which further comprises the damage types including pits, cracks, unfilled corners and/or exposed ribs.
Further, in order to better classify the damage, the invention provides a segment damage detection method based on computer vision, which further comprises the steps of judging whether the actual width of the segment crack is larger than a preset crack threshold value or not so as to determine the crack degree, and determining the damage classification based on the crack degree and a damage detection result obtained after the damage detection is carried out, wherein the actual width of the segment crack is obtained based on the actual size of the segment, the size of the segment in a segment image and the width of the segment crack in the segment image, and the segment image is a corresponding frame image in a segment video extracted by using a distance meter at a specific distance.
Further, in order to better evaluate the overall quality of the duct piece, the invention provides a duct piece damage detection method based on computer vision, which further comprises the step of measuring the actual weight of the duct piece in real time by using a force transfer conversion sensor, and when the ratio of the actual weight to the standard weight is lower than a preset threshold value, judging that the duct piece is unqualified and carrying out alarm reminding.
The invention also provides a duct piece damage detection system based on computer vision, which comprises a video acquisition module, a surface damage detection module and a video processing module, wherein the video acquisition module is used for shooting the video of the duct piece in real time and transmitting the video of the duct piece to the surface damage detection module; the surface damage detection module is used for carrying out damage detection on the segment video by utilizing a trained segment damage model to determine a damage detection result and transmitting the damage detection result to the comprehensive evaluation module, wherein the segment damage model is constructed by a convolutional neural network; the weighing module is used for collecting the actual weight of the duct piece and transmitting the actual weight to the comprehensive evaluation module; the comprehensive evaluation module is used for comparing the actual weight with the standard weight of the duct piece to obtain a comparison result, obtaining damage classification based on the damage detection result, integrating the damage classification and the comparison result to determine whether the duct piece needs to be repaired or replaced, and if so, generating an alarm signal; and the alarm module is used for receiving the alarm signal and carrying out alarm reminding.
Further, in order to better perform damage detection, the invention provides a segment damage detection system based on computer vision, which further comprises the steps of inputting the actual size of a segment before the damage detection is performed, calculating to obtain the actual width of a segment crack by a comprehensive evaluation module based on the actual size of the segment, the size of the segment in a segment image and the width of the segment crack in the segment image, judging whether the actual width of the segment crack is larger than a preset gap threshold value to determine the crack degree, determining damage classification based on the crack degree and a damage detection result, and taking the segment image as a corresponding frame image in a segment video extracted by using a distance measuring tool at a specific distance.
The invention also provides a shield machine, which comprises a duct piece hoisting machine and the duct piece damage detection system based on the computer vision.
Drawings
FIG. 1 is a flow chart of a computer vision based segment breakage detection method of the present invention;
FIG. 2 is an exemplary view of a segment breakage of the present invention;
FIG. 3 is a schematic diagram of the construction of a segment breakage model of the present invention;
FIG. 4 is a functional block diagram of a computer vision based segment breakage detection system of the present invention;
FIG. 5 is a schematic diagram of a computer vision based segment breakage detection and assessment field test provided by the present invention.
Description of reference numerals:
the method comprises the following steps of 1-a first area, 2-a second area, 3-a duct piece, 4-a first industrial camera, 5-a second industrial camera, 6-a distance measuring sensor, 7-a force transfer conversion sensor, 8-a duct piece hoisting machine and 9-a duct piece.
Detailed Description
In order to make the objects, technical solutions and technical effects of the present invention more apparent, the present invention is further described in detail below with reference to the accompanying drawings and specific embodiments.
The embodiment of the duct piece damage detection method based on computer vision comprises the following steps:
the embodiment provides a duct piece damage detection method based on computer vision. The segment damage detection method based on computer vision can be referred to as a segment damage detection method for short. According to the duct piece damage detection method, the quality of the construction tunnel can be effectively guaranteed, and the potential safety hazard is reduced.
Fig. 1 is a flow chart of a segment breakage detection method based on computer vision according to the present invention. Fig. 2 is a view showing an example of breakage of a segment according to the present invention. Fig. 3 is a schematic structural diagram of a segment breakage model according to the present invention.
In this embodiment, the segment damage detection method based on computer vision includes the actual weight of real-time collection segment video and segment, carries out the damage with the segment damage model that segment video input trained and detects in order to confirm that the damage is categorised, and the segment damage model is constructed by the convolutional neural network and is formed, compares actual weight and the standard weight of this segment in order to obtain the comparative result, whether needs to be repaired or replaced based on the damaged classification and the comparative result determination segment, if needs, then reports to the police and reminds.
Specifically, as shown in fig. 1, segment transportation pictures are collected in real time, and segments are detected through a network model and classified as damaged. Wherein, gather section of jurisdiction transportation picture in real time and can acquire through industry camera. The segment video can also comprise a segment hoisting video besides the transportation picture. And inputting the segment video into the network model for breakage detection to determine the breakage detection result of the segment. The damage detection results may include, but are not limited to, broken corners, cracked segments, pits on the inside, exposed ribs, etc. The normal operation of section of jurisdiction hoisting machine is not influenced in the position of industry camera installation, and can ensure that the section of jurisdiction can be shot by the camera in the in-process of transportation and hoist and mount, especially the inboard and inboard corner of section of jurisdiction.
In this embodiment, the network model may be a trained segment damage model belonging to a computer vision model. The step of obtaining a trained segment damage model may include collecting normal segment images and segment images of various damage types, and preprocessing all collected images to create a segment data set. And training the segment damage model by adopting the segment data set to obtain the trained segment damage model. From this, obtain the damaged model of section of jurisdiction that can detect the damaged type of section of jurisdiction. Wherein, the normal segment image may include normal images of the segments at various angles. The segment images for various types of damage include a full shot of the damaged segment and a partial shot of the damage. The local illumination of the damage may include, but is not limited to, various types of images such as corner damage, cracks, pits, exposed ribs, etc. on the inside and outside of the duct piece. The damage types can include corner damage, duct piece crack, concave pit on the inner side and the like. Fig. 2 is a schematic image of segment breakage. As shown in fig. 2, the first area 1 and the second area 2 are 2 corner areas of the tube sheet 3. A first crack is present in the first region 1 and a second crack is present in the second region 2.
In this embodiment, the pre-processing may include an expansion, and parameter normalization process for all the images collected. Therefore, the influence of external environments (such as complex environmental conditions of illumination, darkness, shielding and the like on a construction site) on the model detection and classification effects can be reduced, the anti-interference capability of the model is improved when the model is trained, and the accuracy of the trained segment damage model is better improved. The way of expansion and expansion can be, but is not limited to, reflection transform (rotation), flip transform (flip), scale transform (scale), contrast transform (contrast), noise perturbation (noise). The parameter normalization process may use 0-mean normalization.
In this embodiment, all images collected may be divided into a training set and a test set. The training set and the test set may include normal segment images and segment images of various types of damage, respectively. From this, can improve the degree of accuracy of the damaged model of section of jurisdiction that trains well better. In addition, the preprocessing of all the collected images to establish the segment data set may specifically be preprocessing a training set and a testing set respectively, and establishing the segment data set by using the preprocessed training set and testing set.
In this embodiment, adopt section of jurisdiction data set to train the damaged model of section of jurisdiction, obtain the damaged model of section of jurisdiction that trains well. From this, can train out the damaged model of section of jurisdiction that has better detection classification ability.
In this embodiment, the segment damage model may be constructed by a convolutional neural network. The convolutional neural network may be a model formed based on a deep learning object detection algorithm, in other words, the convolutional neural network may be a yolov3 algorithm model. The yolov3 algorithm model may include an input layer consisting of DBL, Res-n, and Res units, a Darknet layer, and a yolo output layer. The input layer adjusts tensor to the input image to obtain a first intermediate image, the Darknet layer receives the first intermediate image, feature extraction is conducted on the first intermediate image to output a prediction result, and the yolo output layer receives the prediction result and conducts position regression and classification. Therefore, a segment damage model with better detection performance can be obtained.
Specifically, the input layer adjusts the input image to a tensor of 416 × 3, and then inputs the image to the network. DBL is the basic component of yolov3 algorithm model. As shown in fig. 3, the DBL may be composed of a convolution layer (conv), a bn (bath normalization) layer, and a leakage relu active layer. Res-n may generate a relatively deep network structure. Res-n may consist of zero-padding units (zero padding units), DBL, and n Res units (Res _ unit). The res unit (residual unit) may be composed of two DBLs and one add layer. As shown in fig. 3, the Darknet layer may be composed of DBL and res1, res2, res8, res8, res 4. The Darknet layer receives the image output by the input layer, performs feature extraction on the image and outputs a prediction result. The yolo output layer may be composed of DBL and convolutional layers. Finally, feature maps (feature maps) with three scales of 13 × 255, 26 × 255 and 52 × 255 are respectively output, and position regression and classification are carried out on the basis of the feature maps.
In the present embodiment, the selection of the frame image is realized by a distance meter (e.g., a distance measuring sensor), specifically, the distance of the duct piece is measured in real time by the distance meter. If the segment is determined to have a crack by damage detection, when the distance meter detects that the segment is at a specific distance, the corresponding frame image in the segment video is extracted, for example, when the distance meter measures 1 meter from the segment, the system picks out the frame image corresponding to the distance from the video stream of the segment transportation picture.
In this embodiment, the actual size of the segment needs to be input before performing the breakage detection. Calculating the actual width of the segment crack based on the actual size of the segment, the size of the segment in the segment image and the width of the segment crack in the segment image. The segment image is the extracted frame image. Specifically, the method comprises the following steps: the length of all kinds of sections of jurisdiction that use in advance, wide, the high information (be the actual size of section of jurisdiction) storage system of advancing, according to the first distance of section of jurisdiction to distancer under the specific distance circumstances of distancer calculation, because relative position between distancer and the industry camera is fixed, the system can determine the relation between image and the actual size under the specific distance through corresponding coordinate transformation, obtain the width of section of jurisdiction crack in the image through the damaged model of the section of jurisdiction that trains, and then calculate the cracked actual width of section of jurisdiction. From this, can accurately obtain the actual width of section of jurisdiction fracture. The system may determine whether the actual width is greater than a preset gap threshold to determine the extent of the crack. This facilitates subsequent better classification of the damage. The preset gap threshold may be denoted by w. w can be set according to the on-site duct piece size (actual size of the duct piece). w may be any value of 0.2mm to 1 mm. If actual width is less than or equal to and predetermines the gap threshold value, then the crack degree is less, need not repair damaged crack department, and the section of jurisdiction can normal use. If the actual width is larger than the preset gap threshold value, the crack degree is larger, and the damaged crack needs to be repaired.
In the present embodiment, the damage classification is classified into seven categories according to the degree of cracking and the damage detection result. The seven categories are duct pieces with broken corners and capable of being normally used; the segment is provided with cracks but can be normally used; the inner side of the duct piece is provided with a pit which can be normally used; the edges and corners have damaged duct pieces which need to be repaired or replaced; the segment with cracks needs to be repaired or replaced; the inner side of the duct piece is provided with a pit which needs to be repaired or replaced; intact duct piece. Each frame of segment image of the segment video collected in real time is labeled by using the trained segment damage model, and the damage classification to which the segment belongs is recorded in the label. If there appears the corner in the label of the section of jurisdiction image of this section of jurisdiction and has the damaged section of jurisdiction that needs to repair or change, the section of jurisdiction that the section of jurisdiction has the crack to need to repair or change, there is pit in the inboard and needs to repair or change this three kinds of at least one of the section of jurisdiction, then this section of jurisdiction needs to repair or replace, belongs to the section of jurisdiction that can not the direct use. If the three types of the duct piece needing to be repaired or replaced do not appear in the label of the duct piece image of the duct piece, the duct piece does not need to be repaired or replaced.
In this embodiment, compare actual weight and this section of jurisdiction's standard weight in order to obtain the comparison result, whether the section of jurisdiction needs to be repaired or the replacement is determined based on the damaged classification in the label and comparison result, if needs, then reports to the police and reminds. Specifically, judge whether section of jurisdiction needs to be repaired or replaced according to damaged classification, if need, then report to the police and remind. Specifically, as shown in fig. 1, if the damage classification is a classification that needs to be repaired or replaced, the damage degree of the surface of the segment needs to be repaired or replaced, and the segment cannot be directly used, so that an alarm is given. If not required, acquire the real-time weight of section of jurisdiction and compare with standard weight during hoist and mount section of jurisdiction, when actual weight and standard weight's ratio is less than when predetermineeing the threshold value, then section of jurisdiction weight difference is great, has the quality problem, and this section of jurisdiction is unqualified, can not directly use, reports to the police and reminds. The existing quality problems can be the problems of casting quality, cement solidification and unqualified materials or the existence of cavities in the interior. If the number of the segments is not less than the preset threshold value, the segments have no quality problem, can be directly used, are continuously hoisted to the position for assembly, and are continuously transported and hoisted to the next segment. From this, can make the aassessment to the whole quality of section of jurisdiction better. In addition, when the next segment is transported and hoisted continuously, the segment damage detection method of the embodiment is repeatedly executed.
In this embodiment, the alarm reminding mode may be, but is not limited to, one or more of a light alarm and a voice alarm. The preset threshold may be denoted by m. m can be set according to the type of segment used on site. m can be any percentage of 95-98%. And measuring the actual weight of the duct piece in real time by using the force transfer conversion sensor. The normal use of a segment hoisting machine hanging head cannot be influenced by the position where the force transmission conversion sensor is installed, and the weight of a segment in the hoisting process can be obtained.
In this embodiment, the step of obtaining the real-time weight of section of jurisdiction and comparing with standard weight can be gone on simultaneously with the step that utilizes the section of jurisdiction damage model that trains well to carry out damage detection when hoist and mount section of jurisdiction. In addition, whether the duct piece needs to be repaired or replaced or not can be determined based on the damage classification and the comparison result, whether the quality problem exists in the duct piece or not can be determined based on the comparison result, and if the quality problem exists, an alarm is given out. If not, whether the duct piece needs to be repaired or replaced is judged according to the damage classification, and if so, alarm reminding is carried out. If not, the compound can be directly used.
In this embodiment, the segment damage detection method of this embodiment can be applied to the segment hoisting transportation process, and the segment is subjected to on-site real-time damage detection. Under this condition, as long as the section of jurisdiction is still in the field of vision of camera, the real-time weight of acquiring the section of jurisdiction in real time and comparing with standard weight when utilizing the damaged model of section of jurisdiction that trains to carry out the damage detection. However, the application scenario of the segment damage detection method of the embodiment is not limited to this, and in other embodiments, the segment damage detection method may also be used in any link from the factory to before the segment is hoisted.
The segment damage detection method based on the embodiment is characterized in that before segments are assembled on a construction site, a trained segment damage model is used for detecting damage of a segment video acquired in real time so as to determine whether the surface of the segment has problems of damage, cracks, pits and the like or not in real time, the actual weight of the segment is measured simultaneously, and whether the segment needs to be repaired or replaced or not is informed to site construction personnel through an alarm reminding mode by combining damage classification and comparison of the actual weight and standard weight. From this, can realize detecting the real-time damage of job site section of jurisdiction, and if there is the hole scheduling problem in the inside of section of jurisdiction, can in time discover the problem that weight is not up to standard to ensure construction tunnel's quality effectively, reduce the production of potential safety hazard. The segment damage detection method is simple and effective, has strong engineering practicability, is not influenced by the shade of ambient light and the like, reduces the potential safety hazard possibly caused by neglecting segment damage in a construction site, and ensures the construction quality.
Duct piece breakage detection system embodiment based on computer vision:
the embodiment discloses a duct piece breakage detection system based on computer vision. The segment breakage detection system based on computer vision can be referred to as a segment breakage detection system for short. The duct piece damage detection system based on the embodiment can realize real-time damage detection of duct pieces on a construction site, thereby effectively ensuring the quality of a construction tunnel and reducing the generation of potential safety hazards. By the duct piece damage detection system of the embodiment, the duct piece damage detection method based on computer vision introduced in the method embodiment of the invention can be realized.
FIG. 4 is a functional block diagram of a computer vision based segment breakage detection system of the present invention. FIG. 5 is a schematic diagram of a computer vision based segment breakage detection and assessment field test provided by the present invention. In this embodiment, as shown in fig. 4, the duct piece damage detection system may include a video capture module, a surface damage detection module, a weighing module, a comprehensive evaluation module, and an alarm module. The video acquisition module can be used for shooting the video of the segment in real time and transmitting the video of the segment to the surface damage detection module. The surface damage detection module can be used for carrying out damage detection on the segment video by utilizing a trained segment damage model so as to determine a damage detection result, and transmitting the damage detection result to the comprehensive evaluation module, wherein the segment damage model is formed by constructing a convolutional neural network. The weighing module can be used for gathering the actual weight of section of jurisdiction to transmit actual weight to comprehensive judgement module. The comprehensive evaluation module can compare the actual weight with the standard weight of the duct piece to obtain a comparison result, obtain damage classification based on the damage detection result, and determine whether the duct piece needs to be repaired or replaced by integrating the damage classification and the comparison result, if so, an alarm signal is generated. The alarm module can receive the alarm signal to alarm and remind.
In this embodiment, the video capture module may be an industrial camera. The normal operation of section of jurisdiction hoisting machine is neither influenced to the position of industry camera installation, also can ensure that the section of jurisdiction can be shot by the camera in the in-process of transportation and hoist and mount, especially the inboard and inboard corner of section of jurisdiction. The number of industrial cameras in this embodiment may be at least one. If the number of the industrial cameras is 2, the setting can be performed with reference to the positions of the first industrial camera 4 and the second industrial camera 5 in fig. 5. First industry camera 4 and second industry camera 5 set up the both sides at section of jurisdiction hoisting machine 8. In the process of transportation and hoist and mount, section of jurisdiction 9 can be shot in real time to first industry camera 4 and second industry camera 5. The industrial camera collects duct piece transportation pictures in real time.
In this embodiment, the surface damage detection module may include a trained segment damage model. The surface damage detection module can input the obtained segment video into the trained segment damage model to carry out damage detection, and determines the damage detection result of the segment. The damage detection results may include, but are not limited to, broken corners, cracked segments, pits on the inside, etc. The related contents may specifically refer to corresponding descriptions in the method embodiments, and are not described herein again. In addition, the surface damage detection module can obtain the width of the segment crack in the segment image.
In this embodiment, the weighing module may be a force transfer transducer. The force transfer conversion sensor may be arranged with reference to the position of the force transfer conversion sensor 7 in fig. 5. The force transfer conversion sensor 7 is arranged at the stress position of a hanging head of the segment hoisting machine 8. The normal use of a segment hoisting machine hanging head cannot be influenced by the position where the force transmission conversion sensor is installed, and the weight of a segment in the hoisting process can be obtained. When section of jurisdiction hoisting machine 8 hoists section of jurisdiction 9, use the actual weight that weighing module can obtain section of jurisdiction 9, transmit actual weight to the standard weight who synthesizes the judgment module and this section of jurisdiction and contrast to this judges that whether quality of pouring, solidification effect and material of section of jurisdiction are qualified, and whether there is the cavity inside. In addition, the weighing module can also be performed simultaneously with the surface damage detection module.
In this embodiment, the segment breakage detection system may further include a distance measurement tool. The comprehensive evaluation module needs to select the image frame corresponding to the specific distance measured by the distance measuring tool. Reference may be made to corresponding descriptions in the method embodiments, which are not described herein again. The ranging tool may be a ranging sensor. The ranging sensor may be, but is not limited to, a laser ranging or ultrasonic ranging sensor. The ranging sensor may be arranged with reference to the position of the ranging sensor 6 in fig. 5. The distance measuring sensor 6 is arranged in the middle of a hanging bracket of the segment hoisting machine 8. The position of installing the distance measuring sensor can accurately measure the position distance between the hanger and the duct piece under the condition of not influencing the normal use of the duct piece hoisting machine.
In this embodiment, before carrying out the damage detection, still need to input the actual size of section of jurisdiction in the damaged detecting system of section of jurisdiction (can be simply referred to as the system), synthesize the cracked width of section of jurisdiction in judging the module and calculating the cracked actual width of acquisition section of jurisdiction based on the actual size of section of jurisdiction, the size of section of jurisdiction in the section of jurisdiction image and the section of jurisdiction image. The segment image is a corresponding frame image in the segment video extracted at a specific distance using a ranging tool. Specifically, the comprehensive evaluation module in the system can determine the relationship between the image and the actual size at a specific distance through corresponding coordinate conversion, and receive the width of the segment crack output by the surface damage detection module in the segment image, so as to calculate the actual width of the segment crack. The comprehensive evaluation module judges whether the actual width is larger than a preset gap threshold value or not so as to determine the degree of the crack, and determines the damage classification based on the degree of the crack and the damage detection result. This enables better classification of damage. Reference may be made to corresponding descriptions in the method embodiments, which are not described herein again.
In this embodiment, the comprehensive evaluation module can obtain damage classification based on the damage detection result of the output of the surface damage detection module, and the comprehensive damage classification and the weighing module output result judge whether there is a great risk in the continued use of the duct piece. If the risk is high, triggering alarm reminding, and repairing or replacing the duct piece; if judge that this section of jurisdiction does not have damage or when damaged degree does not influence the normal use of this section of jurisdiction, then continue to hoist to the position in order to prepare for the assembly to continue to transport hoist and mount next section of jurisdiction. The method for specifically obtaining the damage classification method and the method for judging the damage classification method have been described in detail in the above method embodiments, and for those skilled in the art, the judgment mode of the comprehensive judgment module can be known according to the segment damage detection method, and details are not described here. In addition, the next duct piece needing to be hoisted and conveyed by the duct piece trolley can be repeatedly used for detecting and classifying the duct piece in real time when the duct piece appears in the visual field of the camera.
Segment damage detecting system based on this embodiment, the damage on segment surface, crack, pit scheduling problem are detected and are categorised to automatic real-time use based on yolov3 algorithm of degree of depth learning under the condition that does not need artifical supplementary and shield structure machine normal work to use the actual weight of weighing module measurement segment, combine to detect and classify and actual weight and standard weight's weight ratio, make the warning through alarm module when the current segment damaged condition is more serious needs to be repaired or change and remind in order to remind the site construction personnel to repair or change the segment. The quality of the tunneling tunnel is effectively guaranteed, and the potential safety hazard of the tunnel is greatly reduced. The problem of current shield constructs quick-witted job site and lacks real-time damaged detection before using to prefabricated section of jurisdiction is solved, and simple, effective, easy to assemble operation, the engineering practicality is stronger.
The embodiment of the shield machine comprises:
the embodiment also provides a shield machine which can comprise a duct piece hoisting machine and a duct piece damage detection system based on computer vision in the system embodiment of the invention. From this, can be in need not artifical supplementary automatic real-time section of jurisdiction that will use to the building site scene and detect classification and aassessment for the quality of tunnelling tunnel has obtained effectual assurance, has reduced the production of tunnel potential safety hazard greatly.
Claims (10)
1. A segment breakage detection method based on computer vision is characterized by comprising the following steps:
the method comprises the steps of collecting segment videos and the actual weight of a segment in real time, inputting the segment videos into a trained segment damage model to carry out damage detection to determine damage classification, wherein the segment damage model is formed by a convolutional neural network, comparing the actual weight with the standard weight of the segment to obtain a comparison result, determining whether the segment needs to be repaired or replaced or not based on the damage classification and the comparison result, and if so, alarming and reminding.
2. A segment damage detection method based on computer vision as claimed in claim 1, characterized in that the segment damage model is trained using a segment data set, the segment data set is obtained by preprocessing a normal segment image and segment images of various damage types, the normal segment image includes normal images of each angle of the segment, and the segment images of various damage types include full-shot of the damaged segment and local-shot of the damaged portion.
3. The computer vision-based segment damage detection method of claim 2, wherein the preprocessing comprises an expansion, expansion and parameter normalization process for all collected images.
4. A segment damage detecting method based on computer vision according to claim 1, characterized in that the convolutional neural network is yolov3 algorithm model, the yolov3 algorithm model includes an input layer composed of DBL, Res-n and Res units, a Darknet layer and a yolo output layer, the input layer adjusts tensor on the input image to obtain a first intermediate image, the Darknet layer receives the first intermediate image, performs feature extraction on the first intermediate image to output a prediction result, and the yolo output layer receives the prediction result, and performs position regression and classification.
5. The computer vision-based segment breakage detection method of claim 2, wherein the breakage type includes a pit, a crack, a missing corner, and/or a bare rib type.
6. The method of detecting segment damage based on computer vision of claim 5, further comprising determining whether the actual width of segment cracks is greater than a preset crack threshold to determine the degree of cracks, and determining damage classification based on the degree of cracks and the damage detection result obtained after the damage detection, wherein the actual width of segment cracks is obtained based on the actual size of the segment, the size of the segment in the segment image, and the width of segment cracks in the segment image, and the segment image is a corresponding frame image in a segment video extracted at a specific distance by using a distance meter.
7. The computer vision-based segment breakage detection method according to claim 1, wherein a force transfer conversion sensor is used for measuring the actual weight of the segment in real time, and when the ratio of the actual weight to the standard weight is lower than a preset threshold value, the segment is judged to be unqualified, and an alarm is given.
8. A segment breakage detection system based on computer vision, comprising:
the video acquisition module is used for shooting a segment video in real time and transmitting the segment video to the surface damage detection module;
the surface damage detection module is used for carrying out damage detection on the segment video by utilizing a trained segment damage model to determine a damage detection result and transmitting the damage detection result to the comprehensive evaluation module, wherein the segment damage model is constructed by a convolutional neural network;
the weighing module is used for collecting the actual weight of the duct piece and transmitting the actual weight to the comprehensive evaluation module;
the comprehensive evaluation module is used for comparing the actual weight with the standard weight of the duct piece to obtain a comparison result, obtaining damage classification based on the damage detection result, integrating the damage classification and the comparison result to determine whether the duct piece needs to be repaired or replaced, and if so, generating an alarm signal;
and the alarm module is used for receiving the alarm signal and carrying out alarm reminding.
9. The duct piece damage detection system based on computer vision of claim 8, characterized in that before damage detection, the actual size of the duct piece is required to be input, the comprehensive evaluation module calculates to obtain the actual width of the duct piece crack based on the actual size of the duct piece, the size of the duct piece in the duct piece image and the width of the duct piece crack in the duct piece image, judges whether the actual width of the duct piece crack is larger than a preset crack threshold value to determine the crack degree, determines damage classification based on the crack degree and the damage detection result, and the duct piece image is a corresponding frame image in a duct piece video extracted at a specific distance by using a distance measuring tool.
10. A shield machine comprises a segment hoisting machine and is characterized by further comprising a segment breakage detection system based on computer vision in any one of claims 8-9.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202110850769.2A CN113657193A (en) | 2021-07-27 | 2021-07-27 | Segment damage detection method and system based on computer vision and shield machine |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202110850769.2A CN113657193A (en) | 2021-07-27 | 2021-07-27 | Segment damage detection method and system based on computer vision and shield machine |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN113657193A true CN113657193A (en) | 2021-11-16 |
Family
ID=78478787
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202110850769.2A Pending CN113657193A (en) | 2021-07-27 | 2021-07-27 | Segment damage detection method and system based on computer vision and shield machine |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN113657193A (en) |
Citations (7)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US6615648B1 (en) * | 1997-12-22 | 2003-09-09 | The Roads And Traffic Authority On New South Wales | Road pavement deterioration inspection system |
| CN103605348A (en) * | 2013-11-25 | 2014-02-26 | 深圳市九洲电器有限公司 | Electronic product quality control method and system |
| CN109095059A (en) * | 2018-07-06 | 2018-12-28 | 青岛智能产业技术研究院 | unmanned loading system and its loading method |
| CN110806352A (en) * | 2019-11-28 | 2020-02-18 | 广西大学 | Shield segment damage detection method and detection system thereof |
| CN111667011A (en) * | 2020-06-08 | 2020-09-15 | 平安科技（深圳）有限公司 | Damage detection model training method, damage detection model training device, damage detection method, damage detection device, damage detection equipment and damage detection medium |
| CN112233091A (en) * | 2020-10-15 | 2021-01-15 | 青岛理工大学 | Wind turbine blade image damage detection and positioning method |
| CN113034480A (en) * | 2021-04-01 | 2021-06-25 | 西安道法数器信息科技有限公司 | Blast furnace damage analysis method based on artificial intelligence and image processing |
-
2021
- 2021-07-27 CN CN202110850769.2A patent/CN113657193A/en active Pending
Patent Citations (7)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US6615648B1 (en) * | 1997-12-22 | 2003-09-09 | The Roads And Traffic Authority On New South Wales | Road pavement deterioration inspection system |
| CN103605348A (en) * | 2013-11-25 | 2014-02-26 | 深圳市九洲电器有限公司 | Electronic product quality control method and system |
| CN109095059A (en) * | 2018-07-06 | 2018-12-28 | 青岛智能产业技术研究院 | unmanned loading system and its loading method |
| CN110806352A (en) * | 2019-11-28 | 2020-02-18 | 广西大学 | Shield segment damage detection method and detection system thereof |
| CN111667011A (en) * | 2020-06-08 | 2020-09-15 | 平安科技（深圳）有限公司 | Damage detection model training method, damage detection model training device, damage detection method, damage detection device, damage detection equipment and damage detection medium |
| CN112233091A (en) * | 2020-10-15 | 2021-01-15 | 青岛理工大学 | Wind turbine blade image damage detection and positioning method |
| CN113034480A (en) * | 2021-04-01 | 2021-06-25 | 西安道法数器信息科技有限公司 | Blast furnace damage analysis method based on artificial intelligence and image processing |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN106049210B (en) | A kind of track condition Intelligent Measurement platform | |
| CN112945139B (en) | Shield engineering auxiliary system combining three-dimensional scanning with BIM technology | |
| CN111322985B (en) | Tunnel clearance analysis method, device and system based on laser point cloud | |
| CN106224007A (en) | Safety for tunnel engineering monitoring and warning management system | |
| KR102112046B1 (en) | Method for maintenance and safety diagnosis of facilities | |
| US10641898B1 (en) | Structural displacement measurement using unmanned aerial vehicles equipped with lasers | |
| CN114485570A (en) | Intelligent monitoring, measuring and early warning system and method for construction safety of tunnel under construction | |
| CN108765974A (en) | A kind of traffic conditions monitoring device, monitoring method and system | |
| CN110245734A (en) | Tunnel structure disease recognition System and method for based on rail mounted crusing robot | |
| Huang et al. | Rapid surface damage detection equipment for subway tunnels based on machine vision system | |
| KR101606002B1 (en) | Method for Safety Inspection and Maintenance of Infrastructure Based Information Communication Technology | |
| Belloni et al. | Tack project: Tunnel and bridge automatic crack monitoring using deep learning and photogrammetry | |
| Liu et al. | An approach for auto bridge inspection based on climbing robot | |
| CN114312905A (en) | Switch point rail form image real-time supervision device | |
| CN113657193A (en) | Segment damage detection method and system based on computer vision and shield machine | |
| CN113569428A (en) | Tunnel abnormity detection method, device and system | |
| CN115755069A (en) | Track foreign matter detection device, detection system and detection method | |
| Huang et al. | Machine inspection equipment for tunnels: a review | |
| CN113074694B (en) | Automatic monitoring device for tunnel section deformation | |
| Leonidas et al. | The development of an automatic inspection system used for the maintenance of rail tunnels | |
| CN115035087A (en) | Novel railway line image detection method and system | |
| CN113378726A (en) | Application method of unmanned aerial vehicle in highway construction management | |
| Sjölander et al. | TACK–an autonomous inspection system for tunnels | |
| Loupos et al. | Integrated robotic system for tunnel structural assessment—the ROBO-SPECT EC project | |
| CN113450357B (en) | Segment image online analysis subsystem and subway shield detection system |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination |