CN112215796A - Railway wagon vehicle image cutting method suitable for railway freight inspection - Google Patents
Railway wagon vehicle image cutting method suitable for railway freight inspection Download PDFInfo
- Publication number
- CN112215796A CN112215796A CN202010951540.3A CN202010951540A CN112215796A CN 112215796 A CN112215796 A CN 112215796A CN 202010951540 A CN202010951540 A CN 202010951540A CN 112215796 A CN112215796 A CN 112215796A
- Authority
- CN
- China
- Prior art keywords
- image
- vehicle
- window
- cutting
- convolution
- 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
- G06T7/0002—Inspection of images, e.g. flaw detection
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T7/00—Image analysis
- G06T7/10—Segmentation; Edge detection
- G06T7/11—Region-based segmentation
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T7/00—Image analysis
- G06T7/10—Segmentation; Edge detection
- G06T7/187—Segmentation; Edge detection involving region growing; involving region merging; involving connected component labelling
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T7/00—Image analysis
- G06T7/60—Analysis of geometric attributes
- G06T7/62—Analysis of geometric attributes of area, perimeter, diameter or volume
-
- 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
- G06T2207/30—Subject of image; Context of image processing
- G06T2207/30248—Vehicle exterior or interior
- G06T2207/30252—Vehicle exterior; Vicinity of vehicle
Abstract
A rail wagon vehicle image cutting method suitable for rail freight inspection specifically comprises the steps of obtaining a high-definition vehicle passing image, preprocessing the image, vertically cutting the image along a horizontal center line, splicing two adjacent images in pairs, performing three-pixel convolution and five-pixel convolution, performing seven-pixel convolution on a three-pixel convolution kernel result image, performing window sliding by using a window width of a five-time difference value of a vehicle joint, inputting a full-connection layer for multi-window parallel judgment, performing overlapping degree judgment, mapping to an original image size, completing vehicle joint labeling, and vertically cutting the vehicle joint along the horizontal center line. The method solves the outstanding problem of image distortion caused by inaccurate cutting of the truck image, and meets the operation requirements of human eye image judgment and machine image judgment on the accuracy of the truck image.
Description
Technical Field
The invention relates to the technical field of freight inspection, in particular to a railway wagon vehicle image cutting method suitable for railway freight inspection.
Background
The railway freight inspection is an important link for inspecting the on-road loading state of the freight car and is an important component for the safe operation of the railway freight; the loading state image of the railway wagon is remotely acquired, manual remote image judgment is realized to replace manual post setting pre-inspection, and the freight inspection working strength can be effectively reduced; the manual remote judging image is used for judging the side and top three-way views of the whole vehicle so as to check and confirm whether the loading appearance state of the truck is good or not, and a complete and accurate vehicle appearance image is required.
The current cutting of goods loading state image mainly passes through methods such as infrared correlation, magnet steel count, owing to receive factors such as the motorcycle type is complicated, the speed of a motor vehicle changes, outdoor environment is changeable influence, has following not enough: due to lack of time synchronization with the video, the cut image is compressed, stretched or miscut; under the condition of vehicle speed change, the magnetic steel counting method still carries out calculation according to a constant speed, so that vehicle type discrimination errors are caused, and vehicle image cutting errors are caused; therefore, the accuracy of the cutting image cannot be ensured by the existing cutting method for the loading state image of the rail wagon.
Disclosure of Invention
The technical problem to be solved by the invention is to provide a railway wagon vehicle image cutting method suitable for railway freight inspection, which can effectively solve the problems provided in the background technology.
In order to solve the problems, the technical scheme adopted by the invention is as follows: a rail wagon vehicle image cutting method suitable for rail freight inspection comprises the following steps:
s1: shooting by a three-way linear array camera to obtain original uncut high-definition vehicle passing image data;
s2: preprocessing images according to camera imaging parameter setting, and sequencing the images according to a three-dimensional shooting sequence to form n original pictures;
s3: the image is vertically cut into the minimum operation unit of the identification image along a horizontal center line;
s4: splicing every two adjacent images to realize that the number of the pictures participating in cutting is 2 n-1;
s5: performing three-pixel convolution, calculating the gray value of a three-pixel matrix to obtain a first accumulated sum, and quantizing a plurality of pixel values for input of a full connection layer;
s6: performing five-pixel convolution, and calculating a gray value of a five-pixel matrix to obtain a second accumulated sum;
s7: performing seven-pixel convolution on the three-pixel convolution kernel result image, and calculating a seven-pixel matrix gray value to obtain a third accumulated sum;
s8: carrying out window sliding by using the window width of a difference value five times larger than the vehicle joint, and carrying out convolution image interception at the carriage joint on the image after convolution to obtain a plurality of window sub-images;
s9: inputting a full connection layer to perform multi-window parallel judgment, identifying whether the sub-window is positioned at the connection part or a part of the connection part, if the output probability value of the full connection layer is more than or equal to 0.85, reserving the window area, otherwise, discarding the window area;
s10: judging the overlapping degree of the window areas reserved in the step S9, combining the areas with the overlapping area larger than 30%, and taking the area smaller than 30% as an independent vehicle joint area; obtaining coordinates of a vehicle joint;
s11: mapping to the size of the original image, and mapping the coordinates of the joint obtained on the convolution image to the original image;
s12: after the marking of the vehicle joints is finished, removing coordinates of the vehicle joints with large horizontal coordinate deviation and small vertical distance;
s13: and (5) vertically cutting the vehicle joint along a horizontal center line to finish the image cutting of the vehicle joint.
As a further preferable scheme of the present invention, the two-to-two stitching in S4 is to divide each acquired original image into two images with equal width according to pixels, and each divided image is respectively stitched with the front image and the rear image to form a new large image.
According to the further preferable scheme of the invention, three pixels, five pixels and seven pixels are used as convolution kernel calculation parameters to perform feature abstraction according to the size of the original image and the size of the feature area at the vehicle joint, so that various image expression features with scaled sizes are obtained, and a small-scale neural network is used for effective judgment.
As a further preferable embodiment of the present invention, the S9 obtains rectangles formed by the window sliding image, and if the intersection area of each two rectangles exceeds 30% of the area of the smaller rectangle, the two rectangles are merged.
As a further preferable scheme of the invention, in the horizontal direction, the objects at the vehicle joints are uniformly distributed in a nonlinear manner along the train advancing direction, and the objects are removed; and in the vertical direction, removing the target with the center of the rectangle deviating from the horizontal central line by more than 10%.
Compared with the prior art, the invention provides the railway wagon vehicle image cutting method suitable for railway freight inspection, which has the following beneficial effects:
the method changes the traditional vehicle image cutting mode depending on hardware, directly utilizes the image processing technology to realize the accurate cutting of the rail wagon, can provide complete and accurate single-section vehicle image results, can carry out self-adaptation on the sampling frequency and the image stretching or compression phenomenon caused by vehicle speed change, and solves the distortion problems of the traditional vehicle image compression, miscut and the like.
Drawings
FIG. 1 is a general flow chart of the present invention;
FIG. 2 is a diagram of a truck top image a acquired on site and segmented in the present invention;
FIG. 3 is an image b adjacent to an image a of the top of a truck collected on site and segmented in the present invention;
FIG. 4 is a schematic diagram of a2+ b1 in two-by-two splicing in the present invention;
FIG. 5 is a diagram of a sliding window ensemble before vehicle junctions are identified in the present invention;
FIG. 6 is a diagram of a sliding window set meeting a threshold in accordance with the present invention;
FIG. 7 is a comparison graph of the merged intersecting sliding windows after vehicle joint identification in accordance with the present invention;
FIG. 8 is a comparison of a vehicle joint of the present invention after screening;
FIG. 9 is a vehicle image cut-out of the present invention;
Detailed Description
The technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the drawings in the embodiments of the present invention, and it is obvious that the described embodiments are only a part of the embodiments of the present invention, and not all of the embodiments.
Referring to the specification and the attached figure 1, the invention provides a railway wagon vehicle image cutting method suitable for railway freight inspection, which comprises the following steps:
s1: shooting by a three-way linear array camera to obtain original uncut high-definition vehicle passing image data;
s2: preprocessing images according to camera imaging parameter setting, and sequencing the images according to a three-dimensional shooting sequence to form n original pictures;
s3: the image is vertically cut into the minimum operation unit of the identification image along a horizontal center line;
s4: splicing every two adjacent images to realize that the number of the pictures participating in cutting is 2n-1, and ensuring that the vehicle joint is completely positioned in one picture;
s5: performing three-pixel convolution, calculating the gray value of a three-pixel matrix to obtain a first accumulated sum, and quantizing a plurality of pixel values for input of a full connection layer;
s6: performing five-pixel convolution, and calculating a gray value of a five-pixel matrix to obtain a second accumulated sum;
s7: performing seven-pixel convolution on the three-pixel convolution kernel result image, calculating a seven-pixel matrix gray value, obtaining a third accumulation sum, and making up for the condition that the five-pixel accumulation sum characteristic representation is insufficient;
s8: carrying out window sliding by using the window width of a difference value five times larger than the vehicle joint, and carrying out convolution image interception at the carriage joint on the image after convolution to obtain a plurality of window sub-images;
s9: inputting a full connection layer to perform multi-window parallel judgment, identifying whether the sub-window is positioned at the connection part or a part of the connection part, if the output probability value of the full connection layer is more than or equal to 0.85, reserving the window area, otherwise, discarding the window area;
s10: judging the overlapping degree of the window areas reserved in the step S9, combining the areas with the overlapping area larger than 30%, and taking the area smaller than 30% as an independent vehicle joint area; obtaining coordinates of a vehicle joint;
s11: mapping to the size of the original image, and mapping the coordinates of the joint obtained on the convolution image to the original image;
s12: after the marking of the vehicle joints is finished, removing coordinates of the vehicle joints with large horizontal coordinate deviation and small vertical distance;
s13: and (5) vertically cutting the vehicle joint along a horizontal center line to finish the image cutting of the vehicle joint.
As a further preferable scheme of the present invention, the two-to-two stitching in S4 is to divide each acquired original image into two images with equal width according to pixels, and each divided image is respectively stitched with the front image and the rear image to form a new large image, so as to ensure that the vehicle joint completely appears in a certain stitched image.
As a further preferable scheme of the invention, according to the size of an original image and the size of a characteristic region at a vehicle joint, three pixels, five pixels and seven pixels are used as convolution kernel calculation parameters to perform characteristic abstraction, so that image expression characteristics with various size scales are obtained, a small-scale neural network is used for effective judgment, the characteristic interference caused by local over-brightness and local over-darkness is avoided, and the algorithm accuracy is improved.
As a further preferable embodiment of the present invention, the S9 obtains rectangles formed by the window sliding image, and if the intersection area of each two rectangles exceeds 30% of the area of the smaller rectangle, the two rectangles are merged.
As a further preferable scheme of the invention, in the horizontal direction, the objects at the vehicle joints are uniformly distributed in a nonlinear manner along the train advancing direction, and the objects are removed; and in the vertical direction, removing the target with the center of the rectangle deviating from the horizontal central line by more than 10%.
As a specific embodiment of the present invention:
acquiring three-way image information of the loading of the railway wagon by using a linear array camera to obtain adjacent top views of the two wagons in the figures 2 and 3, and cutting the obtained images from the middle to obtain a1+ a2 and b1+ b 2; splicing the images a1, a2, b1 and b2 in a pairwise manner according to the adjacent sequence to obtain images a1+ a2, a2+ b1 and b1+ b 2; and carrying out multi-size convolution on the spliced image to obtain a pixel numerical value based on the sizes of various convolution kernels. Pixel values of a plurality of convolution kernel sizes are used as input parameters and are distributed into a sliding window in the next step;
as shown in fig. 5, taking a2+ b1 as an example, a rectangular window is formed according to the marked size of the car compartment joint, the window is overlapped and slides in parallel in the spliced car images, and a sliding window complete set is obtained after sliding;
as shown in fig. 6, on the basis of the sliding window complete set, the full connection layer of the neuron network is used to calculate the sub-window probability value in parallel, and the sub-windows exceeding the threshold value 0.85 are reserved to obtain the window set exceeding the algorithm threshold value;
as shown in fig. 7, taking two intersecting rectangles in the middle of fig. 6 as an example, in the sub-window set meeting the threshold, rectangles whose intersecting areas exceed 30% are merged;
as shown in fig. 8, the sub-windows with large deviation of the center of gravity position in the horizontal and vertical directions are eliminated;
as shown in fig. 9, the line coordinates in the last remaining vehicle joint area are mapped into the original image, and vehicle image cutting is completed.
The above examples are only intended to illustrate the technical solution of the present invention, but not to limit it; although the present invention has been described in detail with reference to the foregoing embodiments, it will be understood by those of ordinary skill in the art that: the technical solutions described in the foregoing embodiments may still be modified, or some technical features may be equivalently replaced; and such modifications or substitutions do not depart from the spirit and scope of the corresponding technical solutions of the embodiments of the present invention.
Claims (5)
1. A rail wagon vehicle image cutting method suitable for rail freight inspection comprises the following steps:
s1: shooting by a three-way linear array camera to obtain original uncut high-definition vehicle passing image data;
s2: preprocessing images according to camera imaging parameter setting, and sequencing the images according to a three-dimensional shooting sequence to form n original pictures;
s3: the image is vertically cut into the minimum operation unit of the identification image along a horizontal center line;
s4: splicing every two adjacent images to realize that the number of the pictures participating in cutting is 2 n-1;
s5: performing three-pixel convolution, calculating the gray value of a three-pixel matrix to obtain a first accumulated sum, and quantizing a plurality of pixel values for input of a full connection layer;
s6: performing five-pixel convolution, and calculating a gray value of a five-pixel matrix to obtain a second accumulated sum;
s7: performing seven-pixel convolution on the three-pixel convolution kernel result image, and calculating a seven-pixel matrix gray value to obtain a third accumulated sum;
s8: carrying out window sliding by using the window width of a difference value five times larger than the vehicle joint, and carrying out convolution image interception at the carriage joint on the image after convolution to obtain a plurality of window sub-images;
s9: inputting a full connection layer to perform multi-window parallel judgment, identifying whether the sub-window is positioned at the connection part or a part of the connection part, if the output probability value of the full connection layer is more than or equal to 0.85, reserving the window area, otherwise, discarding the window area;
s10: judging the overlapping degree of the window areas reserved in the step S9, combining the areas with the overlapping area larger than 30%, and taking the area smaller than 30% as an independent vehicle joint area; obtaining coordinates of a vehicle joint;
s11: mapping to the size of the original image, and mapping the coordinates of the joint obtained on the convolution image to the original image;
s12: after the marking of the vehicle joints is finished, removing coordinates of the vehicle joints with large horizontal coordinate deviation and small vertical distance;
s13: and (5) vertically cutting the vehicle joint along a horizontal center line to finish the image cutting of the vehicle joint.
2. The method for image segmentation of rail wagon vehicles for railway freight inspection as claimed in claim 1, wherein the step of pairwise stitching in S4 is to segment each of the acquired original images into two images with equal width according to pixels, and each of the segmented images is respectively stitched with the front image and the rear image to form a new large image.
3. The method of claim 1, wherein three pixels, five pixels and seven pixels are used as convolution kernel calculation parameters to perform feature abstraction according to the original image size and the feature area size of the vehicle joint, so as to obtain image expression features with various size scales, and a small-scale neural network is used for effective discrimination.
4. The image cutting method for railway wagon vehicles suitable for railway freight inspection as claimed in claim 1, wherein the S9 is a rectangle formed by window sliding images, and if the intersection area of each two rectangles exceeds 30% of the area of the smaller rectangle, the two rectangles are merged.
5. The method for cutting the railway wagon vehicle image suitable for the railway freight inspection as claimed in claim 4, wherein the method comprises the steps of carrying out rejection operation on objects which are uniformly distributed on the vehicle joints in a nonlinear mode along the running direction of the train in the horizontal direction; and in the vertical direction, removing the target with the center of the rectangle deviating from the horizontal central line by more than 10%.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202010951540.3A CN112215796B (en) | 2020-09-11 | 2020-09-11 | Railway wagon vehicle image cutting method suitable for railway freight inspection |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202010951540.3A CN112215796B (en) | 2020-09-11 | 2020-09-11 | Railway wagon vehicle image cutting method suitable for railway freight inspection |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN112215796A true CN112215796A (en) | 2021-01-12 |
| CN112215796B CN112215796B (en) | 2022-12-23 |
Family
ID=74049912
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202010951540.3A Active CN112215796B (en) | 2020-09-11 | 2020-09-11 | Railway wagon vehicle image cutting method suitable for railway freight inspection |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN112215796B (en) |
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN114022537A (en) * | 2021-10-29 | 2022-02-08 | 浙江东鼎电子股份有限公司 | Vehicle loading rate and unbalance loading rate analysis method for dynamic weighing area |
Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20030190058A1 (en) * | 2002-04-04 | 2003-10-09 | Lg Industrial Systems Co., Ltd. | Apparatus and method for measuring queue length of vehicles |
| CN107240102A (en) * | 2017-04-20 | 2017-10-10 | 合肥工业大学 | Malignant tumour area of computer aided method of early diagnosis based on deep learning algorithm |
| US10397524B1 (en) * | 2016-05-18 | 2019-08-27 | UL See Inc. | Three-dimensional around view monitoring system of vehicle and method thereof |
| WO2020140371A1 (en) * | 2019-01-04 | 2020-07-09 | 平安科技（深圳）有限公司 | Deep learning-based vehicle damage identification method and related device |
-
2020
- 2020-09-11 CN CN202010951540.3A patent/CN112215796B/en active Active
Patent Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20030190058A1 (en) * | 2002-04-04 | 2003-10-09 | Lg Industrial Systems Co., Ltd. | Apparatus and method for measuring queue length of vehicles |
| US10397524B1 (en) * | 2016-05-18 | 2019-08-27 | UL See Inc. | Three-dimensional around view monitoring system of vehicle and method thereof |
| CN107240102A (en) * | 2017-04-20 | 2017-10-10 | 合肥工业大学 | Malignant tumour area of computer aided method of early diagnosis based on deep learning algorithm |
| WO2020140371A1 (en) * | 2019-01-04 | 2020-07-09 | 平安科技（深圳）有限公司 | Deep learning-based vehicle damage identification method and related device |
Cited By (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN114022537A (en) * | 2021-10-29 | 2022-02-08 | 浙江东鼎电子股份有限公司 | Vehicle loading rate and unbalance loading rate analysis method for dynamic weighing area |
| CN114022537B (en) * | 2021-10-29 | 2023-05-05 | 浙江东鼎电子股份有限公司 | Method for analyzing loading rate and unbalanced loading rate of vehicle in dynamic weighing area |
Also Published As
| Publication number | Publication date |
|---|---|
| CN112215796B (en) | 2022-12-23 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| DE102018006765B4 (en) | PROCEDURE AND SYSTEM (S) FOR THE MANAGEMENT OF CARGO VEHICLES | |
| EP3454298B1 (en) | Camera device and method for recording a flow of objects | |
| CN109829400B (en) | Rapid vehicle detection method | |
| US20030099330A1 (en) | Method for automatically detecting casting defects in a test piece | |
| ES2727829T3 (en) | Pantograph monitoring system and method | |
| CN104486618B (en) | The noise detecting method and device of video image | |
| US11495022B2 (en) | Method for recognizing an object of a mobile unit | |
| CN114049557A (en) | Garbage sorting robot visual identification method based on deep learning | |
| CN113177565A (en) | Binocular vision position measuring system and method based on deep learning | |
| CN113284154B (en) | Steel coil end face image segmentation method and device and electronic equipment | |
| CN109101932A (en) | The deep learning algorithm of multitask and proximity information fusion based on target detection | |
| CN112215796B (en) | Railway wagon vehicle image cutting method suitable for railway freight inspection | |
| CN114372919B (en) | Method and system for splicing panoramic all-around images of double-trailer train | |
| CN113553916A (en) | Orbit dangerous area obstacle detection method based on convolutional neural network | |
| DE102018006764B4 (en) | PROCEDURE AND SYSTEM (S) FOR THE MANAGEMENT OF CARGO VEHICLES | |
| CN110969135B (en) | Vehicle logo recognition method in natural scene | |
| CN115359306B (en) | Intelligent identification method and system for high-definition images of railway freight inspection | |
| KR102416714B1 (en) | System and method for city-scale tree mapping using 3-channel images and multiple deep learning | |
| DE10142457A1 (en) | Photogrammetric method for determination of the center of optical signal marks on moving objects, e.g. for use in crash-test evaluation, whereby adjoining pixel groups are combined and evaluated to yield fast data compression | |
| CN105574874B (en) | A kind of pseudo- variation targets minimizing technology of sequence image variation detection | |
| CN114519799A (en) | Real-time detection method and system for multi-feature seat state | |
| CN117274843B (en) | Unmanned aerial vehicle front end defect identification method and system based on lightweight edge calculation | |
| CN116580277B (en) | Deep learning-based bottom electronic identification tag missing image identification method | |
| Zhang et al. | ERCM: bionic event-based registration method based on contrast minimum for intelligent unmanned systems | |
| CN116612390B (en) | Information management system for constructional engineering |
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 |