CN113076949A - Method and system for quickly positioning parts of contact net - Google Patents
Method and system for quickly positioning parts of contact net Download PDFInfo
- Publication number
- CN113076949A CN113076949A CN202110345661.8A CN202110345661A CN113076949A CN 113076949 A CN113076949 A CN 113076949A CN 202110345661 A CN202110345661 A CN 202110345661A CN 113076949 A CN113076949 A CN 113076949A
- Authority
- CN
- China
- Prior art keywords
- state
- positioning
- image
- parts
- steady state
- 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.)
- Granted
Links
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/70—Arrangements for image or video recognition or understanding using pattern recognition or machine learning
- G06V10/74—Image or video pattern matching; Proximity measures in feature spaces
- G06V10/75—Organisation of the matching processes, e.g. simultaneous or sequential comparisons of image or video features; Coarse-fine approaches, e.g. multi-scale approaches; using context analysis; Selection of dictionaries
- G06V10/751—Comparing pixel values or logical combinations thereof, or feature values having positional relevance, e.g. template matching
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/22—Matching criteria, e.g. proximity measures
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/20—Image preprocessing
- G06V10/24—Aligning, centring, orientation detection or correction of the image
- G06V10/245—Aligning, centring, orientation detection or correction of the image by locating a pattern; Special marks for positioning
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Computer Vision & Pattern Recognition (AREA)
- General Physics & Mathematics (AREA)
- Physics & Mathematics (AREA)
- Evolutionary Computation (AREA)
- Multimedia (AREA)
- Artificial Intelligence (AREA)
- Data Mining & Analysis (AREA)
- Medical Informatics (AREA)
- Databases & Information Systems (AREA)
- General Health & Medical Sciences (AREA)
- Health & Medical Sciences (AREA)
- Software Systems (AREA)
- Computing Systems (AREA)
- Life Sciences & Earth Sciences (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Bioinformatics & Computational Biology (AREA)
- Evolutionary Biology (AREA)
- General Engineering & Computer Science (AREA)
- Image Analysis (AREA)
Abstract
The invention belongs to the technical field of contact net detection, and discloses a method and a system for quickly positioning contact net parts, wherein the method comprises an unsteady state positioning step, a system steady state judging step and a steady state positioning step, the system comprises an image acquisition module, an unsteady state positioning module, a system steady state judging module and a steady state judging module, the problem of frame-by-frame calculation is avoided by combining position information between adjacent images, and a template matching algorithm and a linear regression algorithm are introduced to replace the contact net parts quickly positioned by a primary positioning algorithm in a multistage positioning process.
Description
Technical Field
The invention belongs to the technical field of contact network detection, and particularly relates to a method and a system for quickly positioning contact network parts.
Background
The contact net is an important component of rail transit, and is mainly used for providing power for trains. The contact net suspension system is easy to generate the defect phenomena of loosening, falling, cracking and the like due to long-term exposure to the external environment, and potential safety hazards are brought to the continuous and stable operation of the contact net. In a high-speed railway power supply safety detection monitoring system, a contact network suspension state detection monitoring device (CCHM, namely 4C) is installed on a contact network operation vehicle or a special vehicle, periodically carries out high-resolution imaging detection on parts of a contact network suspension system and contact network geometric parameters, particularly parts in a cantilever area, and forms a maintenance suggestion on the basis of automatic identification and analysis of detection data to guide maintenance of the contact network.
The first problem of judging whether parts of the overhead line system have defects or not through an image sampling result of the imaging device is how to determine the accurate positions of the parts of the overhead line system of each type from the acquired image, namely, positioning and identifying the parts of the overhead line system contained in the image. In the processing flow commonly used in the prior art, the positions of parts in an image are positioned frame by using a target positioning method based on traditional image processing or deep learning on an image sequence acquired in the advancing process of a detection vehicle. Meanwhile, as the sizes of parts of the overhead line system are smaller, in order to accurately identify and position the parts of the overhead line system with small sizes from an image, a plurality of target positioning algorithms are often required to be cascaded, and small-size parts are accurately positioned in a multistage positioning mode, for example, in the prior art, the publication number is CN110533725A, which is named as 'a method for positioning various parts of the overhead line system of a high-speed rail based on a structure inference network', and discloses a method for positioning various parts of the overhead line system of the high-speed rail based on the structure inference network, firstly, an image data set of the parts of the overhead line system of the high-speed rail is obtained, and feasibility of detecting by designing the structure inference network is analyzed according to a fixed structure of relative position relations of the parts of the overhead line; designing a structural inference network by combining the classification neural network and the positioning neural network according to the analysis conclusion; and finally, inputting the high-speed rail contact net part image data set into a reasoning network for classification training, and inputting a new picture to obtain the positioning information of the new picture. However, in the solutions in the prior art, a plurality of target positioning algorithms need to be cascaded to ensure accurate positioning of small-sized parts, and each image is individually detected in the multi-algorithm cascading manner, so that image and position similarities contained in adjacent images in a time dimension are ignored, unnecessary computation is increased, and the requirement on computing resources is further increased due to the introduction of a multi-level model.
Disclosure of Invention
Aiming at the defects of the prior art and overcoming the problems and the defects in the prior art, the invention aims to provide a method and a system for quickly positioning parts of a contact network, which combine position information between adjacent images, avoid the problem of frame-by-frame calculation, and simultaneously introduce a template matching algorithm and a linear regression algorithm to replace a primary positioning algorithm in a multistage positioning process.
The purpose of the invention is realized by the following scheme:
the invention provides a method for quickly positioning parts of a contact net, which comprises an unsteady state positioning step, a system steady state judging step and a steady state positioning step;
the unsteady state positioning step: setting the initial state of the contact net part positioning system to be unstable; in an unsteady state, firstly, carrying out area rough positioning on the acquired contact network part image by adopting a template matching method, and then finely positioning the position coordinates of the part from the area rough positioning result by adopting a target detection model based on deep learning; after the current frame image is positioned, the system steady state judgment is carried out;
the system steady state determination step: if the variance of the positioning coordinates of the same part in the latest continuous N frames of images by the system does not exceed a threshold value, judging that the current state of the system is in a stable state, and entering a stable positioning step; otherwise, judging that the current state of the system is in an unstable state; the steady-state positioning step: and carrying out linear regression fitting on the first frame of steady-state image by combining part positioning results of the previous continuous N frames of images of the current frame of steady-state image acquired under the system steady-state condition, and positioning the contact net parts in the current frame of steady-state image.
Further, after the position coordinates of the parts are positioned in the steady-state positioning step, acquiring vehicle running state monitoring data, if the vehicle running state data exceeds a state determination threshold, determining that the system returns to an unsteady state, and entering the unsteady-state positioning step; the vehicle operating state detection data includes vehicle operating acceleration and/or vehicle operating vibration data. The acquisition of the vehicle running state monitoring data to judge whether the vehicle enters the unstable state again from the stable state is through the whole detection process, so that the timely switching of the stable state/the unstable state is realized, and the balance between the target positioning accuracy and the positioning speed is realized to the maximum extent.
Furthermore, in the system steady state determination step, if the variance of the positioning coordinates of the same part in the latest continuous N frames of images exceeds a threshold value, a vehicle speed determination step is further included, and if the vehicle speed is detected to be 0, the system ends the positioning of the part; otherwise, the system enters an unsteady state positioning step, namely the whole part positioning process is fully automated without manual intervention.
Preferably, in the unsteady-state positioning step, the template matching-based method compares the contact network part image acquired by the system in an unsteady state with the contact network part image serving as the template by presetting a similarity threshold to obtain similarity, and determines that the contact network part area has a similarity result exceeding the similarity threshold.
Furthermore, in the unsteady positioning step, since some small parts of the overhead line system exist at the edge of the template image, in the image acquisition process, the parts may not be in the matching image due to shaking and the like, so that after the area where the parts of the overhead line system are located is identified from the unsteady image, the range expansion of the identified area where the parts of the overhead line system are located is further performed, and it is ensured that the target is located in the area.
In the unsteady state positioning step, a target detection model based on deep learning is adopted, specifically, the target detection model based on CornerNet algorithm is adopted to identify the outline of the area where the parts of the overhead line system are located and identified by the template matching method, the parts of the overhead line system in the image of the current frame where the parts of the overhead line system need to be positioned are identified, the CornerNet algorithm, classically, reduces the input image size to 1/4 for 1 nxn convolutional layer, then extracting features through a feature extraction network (such as a backbone network, which is formed by connecting a plurality of hourglass modules in series), wherein each node of the feature extraction network firstly reduces the input size through a series of down-sampling operations, and then, the size of the input image is recovered through upsampling, and the method can traverse the image in an effective precision range and identify the specific contour of the contact net part.
Preferably, in the system steady-state determining step, in a short time after the detection device starts to perform image acquisition, due to reasons that the acquisition system may have trigger delay, signal asynchronism, incomplete exposure adaptive adjustment and the like, positions of objects to be detected of the same type in images continuously acquired in a series of images may have floating, which is called as an unsteady state, and due to continuous operation of the detection device, the system gradually enters a stable state, which is shown in the case that the same part continuously acquires the images, and the positions of the parts are kept relatively stable, which is called as a steady state, so that the images in the steady state and the unsteady state are identified in different mannersOtherwise, the identification efficiency is greatly improved, and whether the system is in a steady state or not is judged, specifically, a steady state judgment threshold value is set, if the variance of the coordinates of the parts in the current frame image and the previous continuous N frames of images collected by the system on the contact network parts does not exceed the steady state judgment threshold value, the current state of the system is judged to be in a steady state, otherwise, the current state of the system is judged to be in an unsteady state; more specifically, in the steady state determination step of the system, the coordinates (X, Y) of the center position of a certain catenary component in N frames of images that are continuously acquired by the system most recently are counted, the variance of the X coordinate of the center point and the variance of the Y coordinate of the center point are respectively calculated, when the system gradually enters a stable state, the variance of the X coordinate and/or the Y coordinate of the center point is gradually reduced, when the variance is reduced to a set steady state determination threshold range, the acquisition system is considered to have completely entered the stable state, i.e., a steady state, otherwise, the acquisition system is an unsteady state, the acquired images contain a plurality of catenary components that need to be positioned, and the relative position relationships and sizes of all the catenary components are uniform at the installation positions of the catenary components, so that all the catenary components that are acquired in the steady state or the unsteady state can be determined as long as the catenary component that is selected from the most easily identified among the, the matching process can be quickly finished by the coordinates of the central position of the contact net part, and the state judgment result is obtained at the fastest speed, namely
Wherein the content of the first and second substances,
the average value of the X coordinate of the central point of a certain contact net part in the latest N frames of images,
and the method is the average value of Y coordinates of the central point of the latest N frames, the geometric central point of the contact net part is obtained in the mode of the average value, and then the variance of the coordinates of the part in the latest N frames of images is calculated by the geometric central point of the contact net part.
Furthermore, the variance of the coordinates (X, Y) of the center point of a certain contact net part in the latest N frames of images is
Wherein the content of the first and second substances,
is the variance of the X coordinate of the center point,
is the variance of the Y coordinate of the center point.
A steady state positioning step, wherein when the current state of the system is judged to be in a steady state in the system steady state judging step, linear regression fitting is carried out on a first frame of steady state image acquired after the system is judged to enter the steady state by adopting a method based on linear regression fitting in combination with the identification result of the continuous N frames of images before the first frame of steady state image, the contact net parts in the first frame of steady state image are identified, linear regression fitting is carried out on the images acquired in all steady state states after the system enters the steady state and after the first frame of steady state image, the contact net parts in the images acquired in all steady state states are identified by adopting the method based on linear regression fitting in combination with the identification result of the continuous N frames of images before the images, namely, the X coordinates of the outlines of the recently acquired contact net parts in the images are directly obtained through linear regression fitting, the contours of the recently acquired N frames of images in the steady state are identified, The Y coordinate, the width and the height are used as detection results of the same type of contact net parts, linear regression is a statistical analysis method for determining the interdependent quantitative relation between two or more variables by using regression analysis in mathematical statistics, images acquired in a steady state are not influenced by fluctuation of an operation state, the positions of the contact net parts in the images are uniformly high, the adoption of the method for rapidly identifying the images in the steady state can save a plurality of processes of primary identification and precision identification, a fitting result can also be matched with the images in the whole steady state process, the result can be directly used as the identification results of the contact net parts in the images acquired in other steady states, and the detection effect can be greatly improved.
In the system steady state determination step or the steady state positioning step, when the current state of the system is determined to be stable or the system enters the steady state, the method also comprises the steps of monitoring vehicle running state data and setting a state determination threshold value, wherein the state determination threshold value comprises a vehicle running speed change threshold value and a vehicle running vibration threshold value; and if the vehicle running state data exceeds the state judgment threshold value in the monitoring process, the judgment system returns to the unsteady state again, and the system processes the image acquired by the contact network parts according to the unsteady positioning step.
That is, the system is in a continuous working state along with the running process of the vehicle, the vehicle gradually enters a steady state from a beginning unsteady state, the unsteady state judging step is used for roughly identifying the outline of the parts of the overhead line system on the image acquired under the unsteady state, firstly the rough position of the parts of the overhead line system in the image is drawn, and then CornerNet fine positioning is used for specifically and finely identifying the rough position, so that the image acquired in the rapid change process of the train starting and stopping speed can be accurately identified, the steady state judging step is used for firstly judging the condition of the currently acquired image before identification, if the image is in the steady state, the parts of the overhead line system can be quickly identified by using the images of the last continuous N frames in a linear regression fitting mode, and the identification result is adapted to the identification of all parts of the overhead line system in the same state, the method can be applied quickly, the parts of the contact net can be identified in a more rapid and simple mode in the constant-speed stable running of the train, the compatibility and the applicability of the whole design process are high, the process is saved to the greatest extent, and the design structure is simplified.
Corresponding to the method, the invention also provides a system for quickly positioning the parts of the overhead line system, which comprises an image acquisition module, an unsteady-state positioning module, a system steady-state judgment module and a steady-state judgment module;
the image acquisition module is used for continuously acquiring images of an area where parts of the contact network are located during vehicle traveling, transmitting the images acquired under the unsteady state to the unsteady state positioning module and transmitting the images acquired under the steady state to the steady state positioning module;
the unstable positioning module comprises a template matching unit and a CornerNet target detection model unit, the template matching unit compares the similarity of the image acquired by the image acquisition module with an image of a part of the overhead line system serving as a template, and identifies the region where the part of the overhead line system is located from the image in the unstable state, the CornerNet target detection model unit adopts a target detection model based on a CornerNet algorithm to identify the contour of the region where the part of the overhead line system is located, and the coordinates of the contour of the part of the overhead line system in the image of the current frame where the part of the overhead line system needs to be positioned are identified as a positioning result;
the system steady state judging module is provided with a steady state judging threshold value, the system steady state judging module calculates the variance of the coordinates of the outline of the contact net part positioned in the N frames of images which are continuously and recently collected according to the unsteady state positioning module, if the variance of the coordinates does not exceed the steady state judging threshold value, the current state of the system is judged to be in a steady state, otherwise, the current state of the system is judged to be in an unsteady state, and the system steady state judging result is fed back to the image collecting module;
the steady state judgment module adopts a method based on linear regression fitting, and performs linear regression fitting by combining the identification results of continuous N frames of images before the current frame of image acquired by the image acquisition module in a steady state, so as to identify the contour coordinates of the touch screen parts in all the images acquired by the image acquisition module in the steady state.
Corresponding to the method, the invention also provides computer equipment, which comprises: one or more processors, a memory storing instructions that, when executed by the one or more processors, cause the one or more processors to perform the method described above.
And, a non-transitory machine-readable storage medium storing executable instructions that, when executed, cause the machine to perform the above-described method.
Has the advantages that:
compared with the prior art, the technical scheme provided by the invention has the following advantages:
1. compared with the traditional method for positioning the contact net parts by using the multistage deep learning model, the method provided by the invention has the advantages that the correlation of the spatial positions of the contact net parts in the detection image is considered, the deep learning model is replaced by introducing the template matching algorithm, and the detection speed is greatly increased.
2. According to the method, through judgment of the stable state, whether the fine positioning of the parts is carried out by using a deep learning model or the positioning is carried out by a coordinate statistical regression mode can be automatically switched. In the single collection process of the collection equipment, the system is in a stable state for most of time, and a regression mode is used for replacing a deep learning detection mode, so that the detection speed is further increased.
Drawings
The foregoing and following detailed description of the invention will be apparent when read in conjunction with the following drawings, in which:
FIG. 1 is a schematic flow chart of a positioning method according to the present invention;
FIG. 2 is a schematic diagram illustrating the logic of the positioning method of the present invention;
FIG. 3 is a schematic diagram of the variance change of the positioning of the component from unstable state to stable state in the positioning method of the present invention.
Detailed Description
The technical solutions for achieving the objects of the present invention are further illustrated by the following specific examples, and it should be noted that the technical solutions claimed in the present invention include, but are not limited to, the following examples.
Example 1
The embodiment discloses a method for quickly positioning parts of a contact network, which comprises an unsteady positioning step, a system steady state judgment step and a steady state positioning step, namely, the main idea of the scheme is as shown in figure 1 and figure 1, a traditional positioning method is replaced by a template matching-based method, and the rough positioning position of the parts is determined; 2. expanding a coarse positioning area obtained through a template to a certain extent, and detecting a fine target (part) by using a target detection model based on deep learning; 3. and (4) counting the coordinate position of the fine target (part) in the complete acquired image. Judging whether the state is stable or not; 4. and if the vehicle enters a steady state, predicting the position of the fine target (part) in the completely acquired image directly through coordinate regression, and returning to the non-steady state judgment again when the vehicle running state changes. Specifically, as shown in fig. 2:
the unsteady positioning step is to set the initial state of a system for positioning the contact network parts to be unsteady; and determining the position coordinates of the contact net parts in the images of the system collected under the unsteady state by using a template matching-based method and a deep learning-based target detection model, and simultaneously performing a system steady state judgment step.
In a short time after the detection device starts to acquire images, due to the fact that the acquisition system may have trigger delay, signals are asynchronous, exposure self-adaptive adjustment is not completed and the like, positions of objects to be detected of the same type in the images can float in a series of continuously acquired images, the states are called as unsteady states, and due to continuous operation of detection equipment, the system gradually enters a stable state, the stable state is shown when the same part acquires the images continuously, and the stable state is shown when the positions are kept relatively stable. A schematic diagram of a curve of variance along with time change is shown in fig. 3, taking X-coordinate variance as an example, each point on the abscissa represents the current acquisition time, and the ordinate represents the variance calculated from the latest N frames, when the variance of the X-coordinate of the central point and the Y-coordinate of the central point gradually decreases and falls within a set threshold range, it is determined to enter a steady state period, and a steady state positioning step is started. According to the example of fig. 3, it can be seen that when the acquisition time comes about 13 times, the variance variation calculated at the subsequent time point is small, and it can be considered that the steady state is entered at this time point.
Therefore, images in a steady state and images in an unsteady state are identified in different modes, the identification efficiency is greatly improved, and whether the system is in the steady state or not is judged, namely, a steady state judgment threshold value is set, if the variance of coordinates of the parts in the current frame image collected by the system for the contact network parts and the previous continuous N frames of images does not exceed the steady state judgment threshold value, the current state of the system is judged to be in the steady state, and if not, the current state of the system is judged to be in the unsteady state.
A steady state positioning step, wherein when the current state of the system is judged to be in a steady state in the system steady state judging step, linear regression fitting is carried out on a first frame of steady state image acquired after the system is judged to enter the steady state by adopting a method based on linear regression fitting in combination with the identification result of the continuous N frames of images before the first frame of steady state image, the contact net parts in the first frame of steady state image are identified, linear regression fitting is carried out on the images acquired in all steady state states after the system enters the steady state and after the first frame of steady state image, the contact net parts in the images acquired in all steady state states are identified by adopting the method based on linear regression fitting in combination with the identification result of the continuous N frames of images before the images, namely, the X coordinates of the outlines of the recently acquired contact net parts in the images are directly obtained through linear regression fitting, the contours of the recently acquired N frames of images in the steady state are identified, The Y coordinate, the width and the height are used as detection results of the same type of contact net parts, linear regression is a statistical analysis method for determining the interdependent quantitative relation between two or more variables by using regression analysis in mathematical statistics, images acquired in a steady state are not influenced by fluctuation of an operation state, the positions of the contact net parts in the images are uniformly high, the adoption of the method for rapidly identifying the images in the steady state can save a plurality of processes of primary identification and precision identification, a fitting result can also be matched with the images in the whole steady state process, the result can be directly used as the identification results of the contact net parts in the images acquired in other steady states, and the detection effect can be greatly improved.
Preferably, in the unsteady-state positioning step, the position coordinates of the catenary component in the image are determined by using a template matching-based method and a deep learning-based target detection model, wherein the position coordinates comprise an X coordinate, a Y coordinate, a width and a height of the outline of the catenary component in the image.
Further, in the unsteady-state positioning step, a template matching-based method specifically identifies the area where the contact net component is located from the image under the unsteady state by comparing the similarity between the image acquired by the system under the unsteady state and the contact net component image serving as the template, and the template matching-based method belongs to coarse matching and mainly marks the area where the contact net component is suspected to be located in the image, so that the range of further fine positioning is conveniently narrowed, and the identification speed is accelerated.
More specifically, in the unsteady-state positioning step, the template matching-based method is to compare the similarity between the image acquired by the system under the unsteady state and the image of the contact network component serving as the template to obtain the similarity by presetting a similarity threshold, and determine that the contact network component area has a similarity result exceeding the similarity threshold, wherein the similarity threshold can be adjusted and set according to the use requirements and the environmental characteristics.
Furthermore, since some small parts of the overhead line system exist at the edge of the template image, in the image acquisition process, the parts may not be in the matched image due to shaking and the like, in the unsteady positioning step, after the area where the parts of the overhead line system are located is identified from the unsteady image, the range of the area where the parts of the overhead line system are located is expanded, and it is ensured that the target is located in the area.
Furthermore, in the unsteady-state positioning step, a target detection model based on deep learning is adopted, specifically, a target detection model based on a CornerNet algorithm is adopted to perform contour recognition on the area where the parts of the overhead line system are located, which is identified by the template matching method, so as to identify the parts of the overhead line system in the image of the current frame where the parts of the overhead line system need to be positioned, the CornerNet algorithm, classically, reduces the input image size to 1/4 for 1 nxn convolutional layer, then extracting features through a feature extraction network (such as a backbone network, which is formed by connecting a plurality of hourglass modules in series), wherein each node of the feature extraction network firstly reduces the input size through a series of down-sampling operations, and then, the size of the input image is recovered through upsampling, and the method can traverse the image in an effective precision range and identify the specific contour of the contact net part.
More specifically, in the steady state determination step of the system, the coordinates (X, Y) of the center position of a certain catenary component in N frames of images that are continuously acquired by the system most recently are counted, the variance of the X coordinate of the center point and the variance of the Y coordinate of the center point are respectively calculated, when the system gradually enters a stable state, the variance of the X coordinate and/or the Y coordinate of the center point is gradually reduced, when the variance is reduced to a set steady state determination threshold range, the acquisition system is considered to have completely entered the stable state, i.e., a steady state, otherwise, the acquisition system is an unsteady state, the acquired images contain a plurality of catenary components that need to be positioned, and the relative position relationships and sizes of all the catenary components are uniform at the installation positions of the catenary components, so that all the catenary components that are acquired in the steady state or the unsteady state can be determined as long as the catenary component that is selected from the most easily identified among the, the matching process can be quickly finished by the coordinates of the central position of the contact net part, and the state judgment result is obtained at the fastest speed, namely
Wherein the content of the first and second substances,
the average value of the X coordinate of the central point of a certain contact net part in the latest N frames of images,
and the method is the average value of Y coordinates of the central point of the latest N frames, the geometric central point of the contact net part is obtained in the mode of the average value, and then the variance of the coordinates of the part in the latest N frames of images is calculated by the geometric central point of the contact net part.
Furthermore, the variance of the coordinates (X, Y) of the center point of a certain contact net part in the latest N frames of images is
Wherein the content of the first and second substances,
is the variance of the X coordinate of the center point,
is the variance of the Y coordinate of the center point.
Preferably, in the step of determining the steady state of the system or the step of positioning the steady state, when the current state of the system is determined to be stable or the system enters the steady state, the method further comprises a process of monitoring vehicle running state data, and a state determination threshold value is set, wherein the state determination threshold value comprises a vehicle running speed change threshold value and a vehicle running vibration threshold value; if the vehicle running state data exceeds the state determination threshold in the monitoring process, the determination system returns to the unstable state again, and the system processes the image acquired by the contact network parts according to the unstable state positioning step, so that the design avoids the situation that the contour coordinate identified in the stable state is always within the threshold, and whether the current vehicle running state changes or not and enters the unstable state again cannot be determined according to the actual driving condition, and factors influencing the vehicle running state are many, such as vehicle acceleration and deceleration.
That is, the system is in a continuous working state along with the running process of the vehicle, the vehicle gradually enters a steady state from a beginning unsteady state, the unsteady state judging step is used for roughly identifying the outline of the parts of the overhead line system on the image acquired under the unsteady state, firstly the rough position of the parts of the overhead line system in the image is drawn, and then CornerNet fine positioning is used for specifically and finely identifying the rough position, so that the image acquired in the rapid change process of the train starting and stopping speed can be accurately identified, the steady state judging step is used for firstly judging the condition of the currently acquired image before identification, if the image is in the steady state, the parts of the overhead line system can be quickly identified by using the images of the last continuous N frames in a linear regression fitting mode, and the identification result is adapted to the identification of all parts of the overhead line system in the same state, the method can be applied quickly, the parts of the contact net can be identified in a more rapid and simple mode in the constant-speed stable running of the train, the compatibility and the applicability of the whole design process are high, the process is saved to the greatest extent, and the design structure is simplified.
Example 2
Corresponding to the methods of embodiments 1 and 2, this embodiment further provides a system for quickly positioning components of a catenary, including an image acquisition module, an unsteady-state positioning module, a system steady-state determination module, and a steady-state determination module.
The image acquisition module is used for continuously acquiring images of areas where parts of the contact network are located during vehicle traveling, transmitting the images acquired under the unsteady state to the unsteady state positioning module and transmitting the images acquired under the steady state to the steady state positioning module.
The unstable positioning module comprises a template matching unit and a CornerNet target detection model unit, the template matching unit compares the similarity of the image acquired by the image acquisition module with an image of a part of the overhead line system serving as a template, and identifies the region where the part of the overhead line system is located from the image under the unstable state, the CornerNet target detection model unit adopts a target detection model based on a CornerNet algorithm to identify the contour of the region where the part of the overhead line system is located, and the coordinates of the contour of the part of the overhead line system in the image of the current frame where the part of the overhead line system needs to be positioned are identified as a positioning result.
The system steady state judgment module is provided with a steady state judgment threshold value, the system steady state judgment module calculates a variance according to coordinates of the outline of the contact net part positioned in the N frames of images which are continuously and recently acquired by the unsteady state positioning module, if the variance of the coordinates does not exceed the steady state judgment threshold value, the current state of the system is judged to be in a steady state, otherwise, the current state of the system is judged to be in an unsteady state, and a system steady state judgment result is fed back to the image acquisition module, specifically, the central position of a certain contact net part in the N frames of images which are continuously and recently acquired by the system is countedSetting coordinates (X, Y), respectively calculating the variance of the X coordinate of the central point and the Y coordinate of the central point, when the system gradually enters a stable state, the variance of the X coordinate and/or the Y coordinate of the central point is gradually reduced, when the variance is reduced to be within a set stable state judgment threshold range, the acquisition system is considered to completely enter the stable state, namely a stable state, otherwise the acquisition system is in an unstable state, the acquired image comprises a plurality of overhead line system parts needing positioning, and the relative position relation and the size of all the overhead line system parts are uniform on the installation positions of the overhead line system parts, so that the overhead line system parts can be judged to be acquired under the stable state or the unstable state by quickly selecting the overhead line system parts which are most easily identified as representatives, and the matching process can be quickly completed by using the central position coordinates of the overhead line system parts, the fastest obtaining a state decision result, i.e.
Wherein the content of the first and second substances,
the average value of the X coordinate of the central point of a certain contact net part in the latest N frames of images,
the method is most direct and quickest, and further, the variance of the coordinates (X, Y) of the central point of a certain catenary component in the latest N-frame images is taken as the variance of the coordinates (X, Y) of the central point of the catenary component in the latest N-frame images
Wherein the content of the first and second substances,
is the variance of the X coordinate of the center point,
is the variance of the Y coordinate of the center point.
The steady state judgment module adopts a method based on linear regression fitting, and performs linear regression fitting by combining the identification results of continuous N frames of images before the current frame of image acquired by the image acquisition module in a steady state, so as to identify the contour coordinates of the touch screen parts in all the images acquired by the image acquisition module in the steady state.
Claims (10)
1. A method for quickly positioning parts of a contact net is characterized by comprising the following steps: the method comprises the steps of unsteady positioning, system steady state judgment and steady state positioning;
the unsteady state positioning step: setting the initial state of the contact net part positioning system to be unstable; in an unsteady state, firstly, carrying out area rough positioning on the acquired contact network part image by adopting a template matching method, and then finely positioning the position coordinates of the part from the area rough positioning result by adopting a target detection model based on deep learning; after the current frame image is positioned, the system steady state judgment is carried out;
the system steady state determination step: if the variance of the positioning coordinates of the same part in the latest continuous N frames of images by the system does not exceed a threshold value, judging that the current state of the system is in a stable state, and entering a stable positioning step; otherwise, judging that the current state of the system is in an unstable state;
the steady-state positioning step: and carrying out linear regression fitting on the first frame of steady-state image by combining the part positioning result of the continuous N frames of images before the current frame of steady-state image under the system steady-state condition, and positioning the contact net parts in the current frame of steady-state image.
2. The method for quickly positioning the parts of the overhead line system of claim 1, which is characterized in that: after the position coordinates of the parts are positioned in the steady-state positioning step, acquiring vehicle running state monitoring data, if the vehicle running state data exceeds a state judgment threshold value, judging that the system returns to an unsteady state, and entering the unsteady-state positioning step; the vehicle operating state detection data includes vehicle operating acceleration and/or vehicle operating vibration data.
3. The method for quickly positioning the parts of the overhead line system of claim 1, which is characterized in that: in the step of judging the steady state of the system, if the variance of the positioning coordinates of the same part in the latest continuous N frames of images exceeds a threshold value, the method also comprises a step of judging the speed of the vehicle, and if the speed of the vehicle is 0, the system finishes the positioning of the part; otherwise, the system enters the unsteady-state positioning step.
4. The method for quickly positioning the parts of the overhead line system as claimed in any one of claims 1 to 3, wherein: in the unsteady state positioning step, a template matching-based method is to compare the similarity of the contact network part image acquired by the system in an unsteady state and the contact network part image serving as a template to obtain the similarity by presetting a similarity threshold, and determine that the contact network part area has a similarity result exceeding the similarity threshold.
5. The method for quickly positioning the parts of the overhead line system of claim 4, which is characterized in that: in the unsteady state positioning step, after the area where the contact network parts are located is identified from the unsteady state diagram, the range expansion of the identified area where the contact network parts are located is further included.
6. The method for quickly positioning the parts of the overhead line system of claim 4, which is characterized in that: in the unsteady state positioning step, a target detection model based on deep learning is adopted, specifically, the target detection model based on the CornerNet algorithm is adopted to perform contour recognition on the area where the contact net parts identified by the template matching method are located, and the contact net parts in the image of the current frame where the contact net parts need to be positioned are identified.
7. The method for quickly positioning the parts of the overhead line system of claim 1, which is characterized in that: in the steady state judgment step of the system, the latest continuous statistics of the system are carried outThe method comprises the steps that coordinates (x, y) of the center position of a certain contact net part in N acquired frame images are respectively calculated, the variance of the x coordinate of the center point and the y coordinate of the center point is calculated, when the variance is reduced to be within a set steady state judgment threshold range, the acquisition system is considered to be in a steady state completely, namely a steady state, or else, the acquisition system is in an unsteady state; namely, it is
Wherein the content of the first and second substances,
the average value of the X coordinate of the central point of a certain contact net part in the latest N frames of images,
and calculating the geometric center point of the contact net part in an average manner for the average value of the Y coordinate of the center point of the latest N frames, and then calculating the variance of the coordinates of the part in the latest N frames of images by using the geometric center point of the contact net part.
8. The utility model provides a quick positioning system of contact net spare part which characterized in that: the system comprises an image acquisition module, an unsteady-state positioning module, a system steady-state judgment module and a steady-state judgment module;
the image acquisition module is used for continuously acquiring images of an area where parts of the contact network are located during vehicle traveling, transmitting the images acquired under the unsteady state to the unsteady state positioning module and transmitting the images acquired under the steady state to the steady state positioning module;
the unstable positioning module comprises a template matching unit and a CornerNet target detection model unit, the template matching unit compares the similarity of the image acquired by the image acquisition module with an image of a part of the overhead line system serving as a template, and identifies the region where the part of the overhead line system is located from the image in the unstable state, the CornerNet target detection model unit adopts a target detection model based on a CornerNet algorithm to identify the contour of the region where the part of the overhead line system is located, and the coordinates of the contour of the part of the overhead line system in the image of the current frame where the part of the overhead line system needs to be positioned are identified as a positioning result;
the system steady state judging module is provided with a steady state judging threshold value, the system steady state judging module calculates the variance of the coordinates of the outline of the contact net part positioned in the N frames of images which are continuously and recently collected according to the unsteady state positioning module, if the variance of the coordinates does not exceed the steady state judging threshold value, the current state of the system is judged to be in a steady state, otherwise, the current state of the system is judged to be in an unsteady state, and the system steady state judging result is fed back to the image collecting module;
the steady state judgment module adopts a method based on linear regression fitting, and performs linear regression fitting by combining the identification results of continuous N frames of images before the current frame of image acquired by the image acquisition module in a steady state, so as to identify the contour coordinates of the touch screen parts in all the images acquired by the image acquisition module in the steady state.
9. A computer device, characterized by: comprising one or more processors, a memory storing instructions that, when executed by the one or more processors, cause the one or more processors to perform the method of any of claims 1-7 above.
10. A non-transitory machine-readable storage medium, characterized in that: stored with executable instructions that, when executed, cause the machine to perform the method of any of the preceding claims 1-7.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202110345661.8A CN113076949B (en) | 2021-03-31 | 2021-03-31 | Method and system for quickly positioning parts of contact net |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202110345661.8A CN113076949B (en) | 2021-03-31 | 2021-03-31 | Method and system for quickly positioning parts of contact net |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN113076949A true CN113076949A (en) | 2021-07-06 |
| CN113076949B CN113076949B (en) | 2023-04-18 |
Family
ID=76613989
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202110345661.8A Active CN113076949B (en) | 2021-03-31 | 2021-03-31 | Method and system for quickly positioning parts of contact net |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN113076949B (en) |
Citations (11)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| EP2015101A2 (en) * | 2007-07-12 | 2009-01-14 | Vodafone Group PLC | System and process for enabling location queries for locating a user of a mobile terminal for mobile networks which adds dimensions of time, are, probability and distance beween users |
| CN102521651A (en) * | 2011-12-22 | 2012-06-27 | 北京交通大学 | Bow net contact force prediction method based on NARX neural networks |
| WO2016204385A1 (en) * | 2015-06-15 | 2016-12-22 | 한국철도기술연구원 | Method and apparatus for detecting vibration information of electric railway vehicle |
| CN107123161A (en) * | 2017-06-14 | 2017-09-01 | 西南交通大学 | A kind of the whole network three-dimensional rebuilding method of contact net zero based on NARF and FPFH |
| CN107154033A (en) * | 2016-03-03 | 2017-09-12 | 成都交大光芒科技股份有限公司 | A kind of high ferro contact net rotation ears vertical openings pin missing detection method and system |
| CN107705256A (en) * | 2017-09-13 | 2018-02-16 | 西南交通大学 | A kind of forward direction Vehicular video image enchancing method semantic based on contact net |
| CN108509950A (en) * | 2018-03-27 | 2018-09-07 | 吴泽彬 | Railway contact line pillar number plate based on probability characteristics Weighted Fusion detects method of identification |
| CN108844542A (en) * | 2018-04-23 | 2018-11-20 | 合肥工业大学 | A kind of error analysis and compensation method based on unmanned detection vehicle automatic station-keeping system |
| CN110276785A (en) * | 2019-06-24 | 2019-09-24 | 电子科技大学 | One kind is anti-to block infrared object tracking method |
| CN110490166A (en) * | 2019-08-26 | 2019-11-22 | 石家庄铁道大学 | High iron catenary pipe cap automatic testing method based on depth learning technology |
| CN111383273A (en) * | 2020-03-07 | 2020-07-07 | 西南交通大学 | High-speed rail contact net part positioning method based on improved structure reasoning network |
-
2021
- 2021-03-31 CN CN202110345661.8A patent/CN113076949B/en active Active
Patent Citations (12)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| EP2015101A2 (en) * | 2007-07-12 | 2009-01-14 | Vodafone Group PLC | System and process for enabling location queries for locating a user of a mobile terminal for mobile networks which adds dimensions of time, are, probability and distance beween users |
| CN102521651A (en) * | 2011-12-22 | 2012-06-27 | 北京交通大学 | Bow net contact force prediction method based on NARX neural networks |
| WO2016204385A1 (en) * | 2015-06-15 | 2016-12-22 | 한국철도기술연구원 | Method and apparatus for detecting vibration information of electric railway vehicle |
| CN107743577A (en) * | 2015-06-15 | 2018-02-27 | 韩国铁道技术研究院 | For the method and apparatus for the vibration information for detecting electronic rail truck |
| CN107154033A (en) * | 2016-03-03 | 2017-09-12 | 成都交大光芒科技股份有限公司 | A kind of high ferro contact net rotation ears vertical openings pin missing detection method and system |
| CN107123161A (en) * | 2017-06-14 | 2017-09-01 | 西南交通大学 | A kind of the whole network three-dimensional rebuilding method of contact net zero based on NARF and FPFH |
| CN107705256A (en) * | 2017-09-13 | 2018-02-16 | 西南交通大学 | A kind of forward direction Vehicular video image enchancing method semantic based on contact net |
| CN108509950A (en) * | 2018-03-27 | 2018-09-07 | 吴泽彬 | Railway contact line pillar number plate based on probability characteristics Weighted Fusion detects method of identification |
| CN108844542A (en) * | 2018-04-23 | 2018-11-20 | 合肥工业大学 | A kind of error analysis and compensation method based on unmanned detection vehicle automatic station-keeping system |
| CN110276785A (en) * | 2019-06-24 | 2019-09-24 | 电子科技大学 | One kind is anti-to block infrared object tracking method |
| CN110490166A (en) * | 2019-08-26 | 2019-11-22 | 石家庄铁道大学 | High iron catenary pipe cap automatic testing method based on depth learning technology |
| CN111383273A (en) * | 2020-03-07 | 2020-07-07 | 西南交通大学 | High-speed rail contact net part positioning method based on improved structure reasoning network |
Non-Patent Citations (5)
| Title |
|---|
| QIAODI WEN等: ""Deep Learning Approaches on Defect Detection in High Resolution Aerial Images of Insulators"", 《SENSORS》 * |
| 刘建仁等: ""高速铁路弓网动态图像检测技术研究综述"", 《装备制造技术》 * |
| 占栋: ""线结构光视觉测量关键技术及在轨道交通巡检中应用"", 《中国博士学位论文全文数据库工程科技Ⅱ辑》 * |
| 赵冰: ""基于深度学习的铁路图像智能分析关键技术研究与应用"", 《中国博士学位论文全文数据库工程科技Ⅰ辑》 * |
| 郭雅婕: ""基于轮廓线波动规律的高铁接触网绝缘子提取方法研究"", 《电工技术》 * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN113076949B (en) | 2023-04-18 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| TWI497422B (en) | A system and method for recognizing license plate image | |
| CN111640089B (en) | Defect detection method and device based on feature map center point | |
| KR100247969B1 (en) | Apparatus and method for massive pattern matching | |
| CN113344967B (en) | Dynamic target identification tracking method under complex background | |
| JP2019133306A (en) | Image processing system and image processing method | |
| CN112419289B (en) | Intelligent detection method for urban subway rail fastener defects | |
| CN111179233A (en) | Self-adaptive deviation rectifying method based on laser cutting of two-dimensional parts | |
| CN111417960A (en) | Information processing apparatus, information processing method, and program | |
| CN107247967B (en) | Vehicle window annual inspection mark detection method based on R-CNN | |
| CN101320477B (en) | Human body tracing method and equipment thereof | |
| CN117746077B (en) | Chip defect detection method, device, equipment and storage medium | |
| CN113076949B (en) | Method and system for quickly positioning parts of contact net | |
| CN113989179A (en) | Train wheel set tread defect detection method and system based on target detection algorithm | |
| CN111695582A (en) | Flutter texture detection method and device | |
| CN111563886B (en) | Unsupervised feature learning-based tunnel steel rail surface disease detection method and device | |
| CN116714437B (en) | Hydrogen fuel cell automobile safety monitoring system and monitoring method based on big data | |
| CN116468312A (en) | Logistics vehicle access management system and method based on unattended operation | |
| Aydin et al. | A deep learning-based hybrid approach to detect fastener defects in real-time | |
| Liu et al. | An end-to-end steel strip surface defects detection framework: Considering complex background interference | |
| CN112766306B (en) | Deceleration strip area identification method based on SVM algorithm | |
| CN112419304B (en) | Multi-stage target detection method and device for one-dimensional data | |
| CN111860289B (en) | Time sequence action detection method and device and computer equipment | |
| CN114862786A (en) | Retinex image enhancement and Ostu threshold segmentation based isolated zone detection method and system | |
| CN113205118A (en) | Map matching method, and method and device for determining map matching model | |
| CN110807456A (en) | Method and device for positioning bank card number |
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 | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |