CN112884821A - Method for making super-long train template image - Google Patents
Method for making super-long train template image Download PDFInfo
- Publication number
- CN112884821A CN112884821A CN202110158020.1A CN202110158020A CN112884821A CN 112884821 A CN112884821 A CN 112884821A CN 202110158020 A CN202110158020 A CN 202110158020A CN 112884821 A CN112884821 A CN 112884821A
- Authority
- CN
- China
- Prior art keywords
- image
- train
- features
- images
- template
- 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
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T7/00—Image analysis
- G06T7/30—Determination of transform parameters for the alignment of images, i.e. image registration
- G06T7/33—Determination of transform parameters for the alignment of images, i.e. image registration using feature-based methods
- G06T7/337—Determination of transform parameters for the alignment of images, i.e. image registration using feature-based methods involving reference images or patches
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T7/00—Image analysis
- G06T7/70—Determining position or orientation of objects or cameras
- G06T7/73—Determining position or orientation of objects or cameras using feature-based methods
- G06T7/74—Determining position or orientation of objects or cameras using feature-based methods involving reference images or patches
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/40—Extraction of image or video features
- G06V10/44—Local feature extraction by analysis of parts of the pattern, e.g. by detecting edges, contours, loops, corners, strokes or intersections; Connectivity analysis, e.g. of connected components
- G06V10/443—Local feature extraction by analysis of parts of the pattern, e.g. by detecting edges, contours, loops, corners, strokes or intersections; Connectivity analysis, e.g. of connected components by matching or filtering
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/20—Special algorithmic details
- G06T2207/20021—Dividing image into blocks, subimages or windows
Landscapes
- Engineering & Computer Science (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- Multimedia (AREA)
- Image Analysis (AREA)
Abstract
The invention discloses a method for manufacturing an overlong train template image, which specifically comprises the following steps: acquiring an image of a train to be manufactured, and importing an existing standard template image as a reference image; adjusting an image, and setting boundary preview; extracting features from the images by using a SURF feature extraction algorithm, pairing the features in the two images by using a K-neighbor-based violence matching algorithm to obtain a mapping relation from the image features of the train to be made to the features of the reference image, and adjusting the image content according to the coordinate information of the mapping relation; and storing the manufactured image result and the characteristic information. The invention has high efficiency, high precision and strong applicability, and can finish the template manufacture no matter how different the reference images are. The method fills the gap of train template image production in the industry.
Description
Technical Field
The invention belongs to the technical field of train image processing, and particularly relates to a method for manufacturing a template image of an ultra-long train.
Background
When using computer vision to perform automated anomaly checking on train components, the train will acquire images at a low speed through the image acquisition points. Due to the non-rigid motion of the train and other reasons, the acquired image has distortion in the transverse direction, and the distortion needs to be eliminated by using an image registration algorithm. The image registration algorithm generally adopts the idea of template matching, so that the result of image registration depends on the quality of a template image. For each new train, a complete, clear, undistorted set of template images is required before the image registration algorithm is used.
Since the first image of each train is also distorted, it is necessary to refer to the images of other trains in the process of making the template image. Similar schemes can use other train number standard images as reference images to directly make the train number template image by using an image registration algorithm. The image registration algorithm has two common ideas. Firstly, according to the idea of template matching, a local reference image is utilized to search a corresponding relation in an image to be registered, and the method has the advantages of high speed and high precision. Secondly, according to the idea of feature mapping, firstly extracting the features of the reference image and the image to be registered, and then pairing the extracted features with each other by using a feature matching algorithm to form a mapping relation.
If the template of the current train is obtained by directly using the image registration algorithm, the difference of the image content is too large because other trains and the current train are different in train number and even different in vehicle type. In practical applications, there are many factors that affect the difference in image content, and the specific analysis and classification are listed below.
1. The camera parameters are different: the two image acquisitions may be at different acquisition points, so that the environments of the acquisition points are different, and the parameters of the line camera used for image acquisition, such as gain, exposure and the like, may also be different.
2. The lighting conditions are different: the two image acquisitions are at different times, and ambient light sources may affect the brightness and the reflection areas in the images.
3. The stains on the car body are different: train cleaning is periodic, the longer the collection time is from the cleaning time, the thicker the dust on the train body can be, and in addition, sewage can also exist on the train body passing through a rainy section. Such stains can severely obscure image content.
4. The train has different structures: usually, the appearance structure, the component position and the component model of two different trains are greatly different. Even if the two trains are of the same model, there are many inherent differences.
Under the condition, the image registration algorithm is used, a large amount of local non-coincidence phenomena occur, and the registration mapping relation is easy to be disordered to cause content dislocation. Even if a sparse matching relationship can be obtained, the accuracy of the image cannot be guaranteed.
Disclosure of Invention
Aiming at the problems, the invention provides a method for manufacturing a template image of an ultra-long train, aiming at solving the problem of manufacturing the template image.
The invention discloses a method for manufacturing an overlong train template image, which comprises the following steps of:
step 1: and (4) collecting and importing images.
S11: the train image is obtained by splicing each frame of image after the train is transversely scanned by the linear array camera, and then the image is adjusted to 180000 pixels in length and 2048 pixels in height.
S12: the image is displayed in pages, the whole train image is divided into 45 sub-images according to the 4000-pixel length, the sub-images are sequentially numbered with page pages in pages from left to right, and only one page of sub-image is displayed each time.
S13: judging whether a train type in the standard template image exists or not, if so, using the train image in the standard template image as a reference image to occupy a red channel R and a blue channel B and display purple in the image, using a train image to be manufactured to occupy a green channel G and display green in the image, and displaying an RGB color image obtained by overlapping the channels in a front end interface; otherwise, only displaying the gray-scale image of the train image to be manufactured in the front-end interface.
Step 2: and adjusting the view.
The boundary preview images are added to the left side and the right side of the sub-image in the front end interface, the left boundary preview image displays the last image segment of the previous sub-image, the right boundary preview image displays the beginning image segment of the next sub-image, and the left boundary preview image of the first page and the right boundary preview image of the last page are blank images and do not display contents.
And step 3: and (6) adjusting the characteristics.
S31: and extracting features from the image by using an SURF feature extraction algorithm, wherein the features comprise point features and line features except transverse lines, so that coordinates, directions and descriptors of the features are obtained, and the features are added, moved and deleted.
S32: carrying out feature pairing on the train image to be manufactured with the reference image, and after extracting features from the reference image and the train image to be manufactured, pairing the features in the two images by adopting a K-nearest neighbor-based violence matching algorithm to obtain a mapping relation from the train image features to be manufactured to the reference image features; in each matching pair, the coordinates of the characteristics of the train image to be manufactured are initial coordinates of the characteristics, and the coordinates of the corresponding characteristics of the reference image are correction coordinates of the characteristics; the image content is adjusted according to its coordinate information.
And 4, step 4: and (5) storing the train template image.
Storing the prepared image result and the characteristic information, wherein the image is stored by using a BMP bitmap format; the characteristic information is the making process of the template image, namely the initial coordinates and the corrected coordinates of all the characteristics, and the storage format of the characteristic information is a TXT file.
Further, in step 2, the boundary preview sets preset ranges, the preset ranges are 1/2, 1/3, 1/4 and 1/8, and the lengths of 1/2, 1/3, 1/4 and 1/8 of the sub-images can be previewed respectively.
Further, step 3 further includes S33: after the features are automatically adjusted, the wrong feature matching in the features is manually deleted, and the features are added and moved in the feature sparse area.
The beneficial technical effects of the invention are as follows:
the invention solves the problem that the registration template can not be accurately manufactured in the industry. In the past, when the template is required to be acquired, the train is scanned at a constant speed by hardware, and when the hardware cannot meet the requirement, the template is manufactured only by correcting the distortion of the image content by using an image registration algorithm. When the template is manufactured by the image registration algorithm, the difference of the reference images is usually large, and the reference images cannot be completely qualified for manufacturing work, so that the image manufacturing is inaccurate, and even the content is disordered. The invention adopts a man-machine interaction strategy and is matched with a template making interaction interface to complete template making, the method has high efficiency, high precision and strong applicability, and the template making can be completed no matter how different the reference images are. The method fills the gap of train template image production in the industry.
Drawings
FIG. 1 is a flow chart of the method for making the template image of the ultra-long train according to the invention.
Detailed Description
The invention is described in further detail below with reference to the figures and specific embodiments.
The method for manufacturing the template image of the ultra-long train disclosed by the invention is shown in figure 1 and comprises the following steps of:
step 1: and (4) collecting and importing images.
S11: the train images are obtained by splicing images of each frame after a linear array camera scans a train transversely, and the images are adjusted to 180000 pixels in length and 2048 pixels in height for making the images more standard subsequently.
S12: because the display area of the front-end interactive interface is limited, all image contents cannot be displayed at one time, the images are displayed in a paging mode, the whole train image is divided into 45 sub-images according to the 4000-pixel length, paging page numbers are sequentially numbered for the sub-images from left to right, and only one paging sub-image is displayed each time.
S13: judging whether a train type in the standard template image exists or not, if so, using the train image in the standard template image as a reference image to occupy a red channel R and a blue channel B and display purple in the image, using a train image to be manufactured to occupy a green channel G and display green in the image, and displaying an RGB color image obtained by overlapping the channels in a front end interface; otherwise, only displaying the gray-scale image of the train image to be manufactured in the front-end interface. In the subsequent template image making process, the adjusted contents are green train images to be made, and the purple reference images are kept unchanged.
Step 2: and adjusting the view.
Because the invention uses the paging strategy, each page of sub-graph has a left boundary and a right boundary, and when the feature is positioned near the boundary, the neighborhood view of the feature is limited. For example, when a feature near the right boundary of the sub-image is adjusted, the content of the neighborhood image outside the right boundary cannot be seen, which is likely to cause erroneous judgment. In order to solve the problems, the boundary preview images are added on the left side and the right side of the sub-image in the front-end interface, the left boundary preview image displays the last image segment of the previous sub-image, the right boundary preview image displays the beginning image segment of the next sub-image, and the left boundary preview image of the first page and the right boundary preview image of the last page are blank images and do not display contents. The boundary preview sets preset ranges of 1/2, 1/3, 1/4 and 1/8, and can preview the lengths of 1/2, 1/3, 1/4 and 1/8 of the sub-images respectively.
And step 3: feature adjustment (the present invention uses an automatic adjustment algorithm + manual adjustment).
The automatic adjustment is as follows: s31: and extracting features from the image by using an SURF feature extraction algorithm, wherein the features comprise point features and line features except transverse lines, so that coordinates, directions and descriptors of the features are obtained, and the features are added, moved and deleted.
S32: carrying out feature pairing on the train image to be manufactured with the reference image, and after extracting features from the reference image and the train image to be manufactured, pairing the features in the two images by adopting a K-nearest neighbor-based violence matching algorithm to obtain a mapping relation from the train image features to be manufactured to the reference image features; in each matching pair, the coordinates of the characteristics of the train image to be manufactured are initial coordinates of the characteristics, and the coordinates of the corresponding characteristics of the reference image are correction coordinates of the characteristics; the image content is adjusted according to its coordinate information.
After the automatic adjustment algorithm is used for processing most of image contents, the advantages and the disadvantages of the two methods can be complemented by matching with the strategy of manual adjustment.
S33: after the features are automatically adjusted, the wrong feature matching in the features is manually deleted, and the features are added and moved in the feature sparse area.
And 4, step 4: and (5) storing the train template image.
Storing the prepared image result and the characteristic information, wherein the image is stored by using a BMP bitmap format; the characteristic information is the making process of the template image, namely the initial coordinates and the corrected coordinates of all the characteristics, and the storage format of the characteristic information is a TXT file. One line is recorded for each individual feature in the file, and the two floating point type coordinates of each line of feature information are separated by a space by default. When the template image is provided with the characteristic information file, the template image can be restored through the mapping relation between the image to be manufactured and the characteristics even if the template image manufacturing result is not available. Therefore, the invention designs a characteristic loading function, can recover the intermediate result stored last time through the image to be made and the characteristic information file stored last time, and continue to complete the template making on the basis.
Claims (3)
1. A method for manufacturing an overlong train template image is characterized by comprising the following steps:
step 1: image acquisition and import:
s11: the train image is obtained by splicing each frame of image after a linear array camera transversely scans a train, and then the image is adjusted to 180000 pixels in length and 2048 pixels in height;
s12: the image is displayed in a paging mode, the whole train image is divided into 45 sub-images according to the 4000-pixel length, paging page numbers are sequentially numbered for the sub-images from left to right, and only one paging sub-image is displayed each time;
s13: judging whether a train type in the standard template image exists or not, if so, using the train image in the standard template image as a reference image to occupy a red channel R and a blue channel B and display purple in the image, using a train image to be manufactured to occupy a green channel G and display green in the image, and displaying an RGB color image obtained by overlapping the channels in a front end interface; otherwise, only displaying a gray scale image of the train image to be manufactured in the front-end interface;
step 2: and (3) view adjustment:
adding boundary preview images to the left side and the right side of the sub-image in the front-end interface, wherein the left boundary preview image displays the last image segment of the previous sub-image, the right boundary preview image displays the beginning image segment of the next sub-image, and the left boundary preview image of the first page and the right boundary preview image of the last page are blank images and do not display contents;
and step 3: characteristic adjustment:
s31: extracting features from the image by using a SURF feature extraction algorithm, wherein the features comprise point features and line features except transverse lines, so as to obtain coordinates, directions and descriptors of the features, and perform adding, moving and deleting operations on the features;
s32: carrying out feature pairing on the train image to be manufactured with the reference image, and after extracting features from the reference image and the train image to be manufactured, pairing the features in the two images by adopting a K-nearest neighbor-based violence matching algorithm to obtain a mapping relation from the train image features to be manufactured to the reference image features; in each matching pair, the coordinates of the characteristics of the train image to be manufactured are initial coordinates of the characteristics, and the coordinates of the corresponding characteristics of the reference image are correction coordinates of the characteristics; adjusting the image content according to the coordinate information;
and 4, step 4: train template image preservation:
storing the prepared image result and the characteristic information, wherein the image is stored by using a BMP bitmap format; the characteristic information is the making process of the template image, namely the initial coordinates and the corrected coordinates of all the characteristics, and the storage format of the characteristic information is a TXT file.
2. The method for making the template image of the ultra-long train as claimed in claim 1, wherein the boundary preview in step 2 sets preset ranges, the preset ranges are 1/2, 1/3, 1/4 and 1/8, and the lengths of 1/2, 1/3, 1/4 and 1/8 of the sub-images can be previewed respectively.
3. The method for making the template image of the ultra-long train according to claim 1, wherein the step 3 further comprises the step of S33: after the features are automatically adjusted, the wrong feature matching in the features is manually deleted, and the features are added and moved in the feature sparse area.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202110158020.1A CN112884821B (en) | 2021-02-05 | 2021-02-05 | Method for making super-long train template image |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202110158020.1A CN112884821B (en) | 2021-02-05 | 2021-02-05 | Method for making super-long train template image |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN112884821A true CN112884821A (en) | 2021-06-01 |
| CN112884821B CN112884821B (en) | 2022-07-05 |
Family
ID=76057331
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202110158020.1A Active CN112884821B (en) | 2021-02-05 | 2021-02-05 | Method for making super-long train template image |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN112884821B (en) |
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN115775269A (en) * | 2023-02-10 | 2023-03-10 | 西南交通大学 | Train image accurate registration method based on line features |
Citations (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20120236334A1 (en) * | 2011-03-18 | 2012-09-20 | Takeshi Suzuki | Image processing apparatus, image processing method, and computer program product |
| US20130063741A1 (en) * | 2011-09-14 | 2013-03-14 | Ricoh Company, Limited | Display processing apparatus, image forming system, and computer program product |
| CN104574339A (en) * | 2015-02-09 | 2015-04-29 | 上海安威士科技股份有限公司 | Multi-scale cylindrical projection panorama image generating method for video monitoring |
| CN107948455A (en) * | 2013-11-28 | 2018-04-20 | 夏普株式会社 | The control method of image processing system and image processing system |
| CN111354026A (en) * | 2020-02-27 | 2020-06-30 | 广州运达智能科技有限公司 | Improved train image accurate registration method and system based on SURF (speeded up robust features) |
| CN112102458A (en) * | 2020-08-31 | 2020-12-18 | 湖南盛鼎科技发展有限责任公司 | Single-lens three-dimensional image reconstruction method based on laser radar point cloud data assistance |
-
2021
- 2021-02-05 CN CN202110158020.1A patent/CN112884821B/en active Active
Patent Citations (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20120236334A1 (en) * | 2011-03-18 | 2012-09-20 | Takeshi Suzuki | Image processing apparatus, image processing method, and computer program product |
| US20130063741A1 (en) * | 2011-09-14 | 2013-03-14 | Ricoh Company, Limited | Display processing apparatus, image forming system, and computer program product |
| CN107948455A (en) * | 2013-11-28 | 2018-04-20 | 夏普株式会社 | The control method of image processing system and image processing system |
| CN104574339A (en) * | 2015-02-09 | 2015-04-29 | 上海安威士科技股份有限公司 | Multi-scale cylindrical projection panorama image generating method for video monitoring |
| CN111354026A (en) * | 2020-02-27 | 2020-06-30 | 广州运达智能科技有限公司 | Improved train image accurate registration method and system based on SURF (speeded up robust features) |
| CN112102458A (en) * | 2020-08-31 | 2020-12-18 | 湖南盛鼎科技发展有限责任公司 | Single-lens three-dimensional image reconstruction method based on laser radar point cloud data assistance |
Non-Patent Citations (2)
| Title |
|---|
| XI DAI: "Navigation of Simultaneous Localization and Mapping by Fusing RGB-D Camera and IMU on UAV", 《 2019 CAA SYMPOSIUM ON FAULT DETECTION, SUPERVISION AND SAFETY FOR TECHNICAL PROCESSES 》, 6 October 2020 (2020-10-06) * |
| 吴琛: "基于图像处理的动车组走行部故障检测方法研究", 《中国优秀博硕士学位论文全文数据库(硕士)工程科技Ⅱ辑》, 15 October 2018 (2018-10-15) * |
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN115775269A (en) * | 2023-02-10 | 2023-03-10 | 西南交通大学 | Train image accurate registration method based on line features |
Also Published As
| Publication number | Publication date |
|---|---|
| CN112884821B (en) | 2022-07-05 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN110084241B (en) | Automatic ammeter reading method based on image recognition | |
| CN111127339B (en) | Method and device for correcting trapezoidal distortion of document image | |
| CN103440629B (en) | Laser labelling is from the digital image processing method of the Video Extensometer of motion tracking | |
| CN111932504B (en) | Edge contour information-based sub-pixel positioning method and device | |
| CN109872309B (en) | Detection system, method, device and computer readable storage medium | |
| CN109409290B (en) | Thermometer verification reading automatic identification system and method | |
| CN105701809B (en) | A kind of method for correcting flat field based on line-scan digital camera scanning | |
| CN103293168B (en) | Fruit surface defect detection method based on visual saliency | |
| CN106559658A (en) | Multi-channel projection fusion band color balance Control Scheme method | |
| CN110363769B (en) | Image segmentation method for cantilever system of high-speed rail contact net supporting device | |
| CN112884821B (en) | Method for making super-long train template image | |
| CN113870267B (en) | Defect detection method, defect detection device, computer equipment and readable storage medium | |
| CN107864365B (en) | Method for eliminating purple border of image | |
| CN116468640B (en) | Video image enhancement method for Internet teaching | |
| CN106815814B (en) | Image pollution processing method applied to paper marking system | |
| CN114283434B (en) | Answer sheet identification method based on machine vision | |
| CN111539421A (en) | Deep learning-based railway locomotive number identification method | |
| CN102592302A (en) | Digital cartoon intelligent dynamic detection system and dynamic detection method | |
| CN110751690A (en) | Visual positioning method for milling machine tool bit | |
| CN112308773B (en) | Unmanned aerial vehicle aerial image nondestructive amplification and splicing fusion method | |
| CN113256563A (en) | Method and system for detecting surface defects of fine product tank based on space attention mechanism | |
| CN106815580A (en) | A kind of license plate locating method | |
| CN108520499B (en) | Image offset correction method based on white-band microscopic imaging | |
| CN110298347A (en) | A kind of recognition methods of the automobile exhaust analyzer screen based on GrayWorld and PCA-CNN | |
| CN113436214B (en) | Brinell hardness indentation circle measuring method and system and computer readable storage medium |
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 |