CN109934135A - A kind of rail foreign matter detecting method decomposed based on low-rank matrix - Google Patents
A kind of rail foreign matter detecting method decomposed based on low-rank matrix Download PDFInfo
- Publication number
- CN109934135A CN109934135A CN201910151851.9A CN201910151851A CN109934135A CN 109934135 A CN109934135 A CN 109934135A CN 201910151851 A CN201910151851 A CN 201910151851A CN 109934135 A CN109934135 A CN 109934135A
- Authority
- CN
- China
- Prior art keywords
- rail
- matrix
- low
- foreign matter
- rank matrix
- 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
Landscapes
- Image Analysis (AREA)
- Image Processing (AREA)
Abstract
The present invention discloses a kind of rail foreign matter detecting method decomposed based on low-rank matrix, belongs to technical field of computer vision.Implementation process includes: 1) to carry out straight-line detection to unmanned plane railway image and screen to straight line group, finds rail edge straight line, determines the region for needing detection processing;2) to rail extracted region pixel vectors and clustering processing is carried out, two subsets of sleeper and stone can be divided into;3) low-rank matrix decomposition is carried out to the matrix that two subsets are constituted, and obtains prospect matrix as difference with the low-rank matrix with background information acquired;4) prospect matrix is filtered and Threshold segmentation can determine position of the foreign matter in rail.The present invention detects rail foreign matter by a kind of algorithm decomposed based on low-rank matrix, quickly confirms foreign matter using the picture that space base monitoring is passed back in real time, can be used for the fields such as railway security monitoring.
Description
Technical field
The invention belongs to technical field of computer vision, are related to digital image processing techniques and target detection technique, specifically
Refer to a kind of rail foreign matter detecting method decomposed based on low-rank matrix.
Background technique
Rail foreign bodies detection refers to guarantee train driving safety, needs to carry out rail region the measure of foreign matter investigation.
It harmful must not be arranged according to the pertinent regulations that China's train drives safely in rail track protective zone and its adjacent domain
The foreign matter of vehicle normally travel is in the railway clearance of national regulation, in order to avoid because after driver can just see foreign matter in closer distance
Carry out brake processing, but brake not in time, braking distance is too long bump against with foreign matter caused by property loss and safety accident.
When the rail that inspector enters no train driving overhauls, since job specification is likely to entrained peace
Other articles such as full cap, work spanner are retained between rail or train passenger throws the foreign matters such as rubbish outside window, cause to arrange
Vehicle security risk.
In recent years, with the continuous development of national unmanned plane industry, the airborne camera lens of UAV flight is usually utilized to execute
The tasks such as special area anomaly monitoring, auxiliary disaster inspection.In field of track traffic, Along Railway is longer, and manpower detects iron
Time-consuming for rail remaining foreign matter, difficulty is big, therefore, by unmanned plane image real-time perfoming rail foreign bodies detection, to train energy
No safe and smooth traveling has very important booster action.
Summary of the invention
The present invention provides a kind of rail foreign matter detecting method decomposed based on low-rank matrix, carries out not needing any sample
In the case that study acquisition model is detected, using sleeper between rail and stone characteristic of cycle repetition using cluster and low-rank square
Battle array decomposes the method for solving low-rank matrix, carries out foreign bodies detection.Due to the respective rgb value tool of rail region sleeper and stone region
The method for having repetition period property, therefore nature being used to cluster is by two class region disconnectings for two subsets.And abnormal object quilt
It is sorted in two sons to concentrate, and dissimilar with place subset other compositions.Can be by picture breakdown by low-rank matrix decomposition
As soon as the low-rank matrix comprising most of similarity redundancy information can make the prospect comprising abnormal object information using poor method is made
Target is stripped from script subset and comes, it is possible to carry out low-rank matrix decomposition under respective type using the method, ask
Obtain foreign substance information.
Specifically, a kind of rail foreign matter detecting method decomposed based on low-rank matrix provided by the invention, including walk as follows
It is rapid:
Step 1: receiving the Aerial Images of UAV flight's camera shooting and inputting in computer, image is located in advance
Then reason carries out straight-line detection to the overall situation by line detection algorithm, obtains straight line group;According to rail length and rail and pillow
The correlation of included angle of straight line between wood screens required straight line group, calculates rail edge straight line, divides rail area
Domain.
Step 2: successively taking rail region to be handled, for the left side edge of each rail edge, along sleeper direction
To rail extracted region pixel vectors, pixel vectors group is clustered using the repeatability in potential stone and sleeper region, point
From at sleeper subset and stone subset and constituting corresponding classification matrix.
Step 3: carrying out low-rank matrix decomposition under respective type to classification matrix, low-rank matrix is found out.To classification matrix
Foreground target foreign substance information is acquired as difference with low-rank matrix.
Step 4: carrying out base image algorithm process, including filtering operation and Threshold segmentation to foreground target foreign substance information.
Result after Threshold segmentation is positioned and is returned in original image, foreign matter position in original image is determined with this, completes rail
Foreign bodies detection.
The beneficial effects of the present invention are:
Deep learning is all based on about the method for rail foreign bodies detection at present and carries out foreign bodies detection, but is based on depth
The method of habit, which needs to learn great amount of samples and train, to be obtained model with this and detects, once in face of unknown foreign matter, different
Situations such as larger, the practical foreign matter of object jump and sample foreign matter are not met, just will appear missing inspection and false retrieval, shows based on depth
Practise the unstability of detection.Therefore, the present invention proposes a kind of rail foreign bodies detection algorithm decomposed based on low-rank matrix, is being not required to
Under the premise of wanting any sample learning, the detection to all foreign matters of rail, this method tool are realized using base image Processing Algorithm
There is preferable robustness.
Detailed description of the invention
A kind of Fig. 1: rail foreign matter detecting method flow chart decomposed based on low-rank matrix;
A kind of Fig. 2: rail original image of taking photo by plane for the rail foreign matter detecting method embodiment decomposed based on low-rank matrix;
A kind of Fig. 3: straight-line detection rail region stroke for the rail foreign matter detecting method embodiment decomposed based on low-rank matrix
Divide result figure;
A kind of Fig. 4: rail foreign bodies detection result figure for the rail foreign matter detecting method embodiment decomposed based on low-rank matrix.
Specific embodiment
Detailed process of the present invention is discussed in detail with reference to the accompanying drawings and detailed description.
The present invention provides a kind of rail foreign matter detecting method decomposed based on low-rank matrix, specific implementation flow such as Fig. 1 institute
Show, specifically comprises the following steps:
Step 1 receives the Aerial Images of UAV flight's camera shooting and inputs in computer.To the image received
It is pre-processed first, it is then complete to image with line detection algorithm to reduce the influence of noise including filtering operation etc.
Office carries out straight-line detection, obtains straight line group, and the straight line group includes the letter such as linear equation, line segment length, starting point, end point
Breath.Straight-line detection processing is related to spatial resolution, and spatial resolution is bigger, and the acquired straight line the more accurate more, but calculation amount
Increase.
According to rail trend in the picture and the relevance of rail straight line and sleeper included angle of straight line to obtained straight
Line group is screened, and calculates rail edge straight line, and find out rail region as sense according to the constraint condition of straight wire spacing
Interest region is handled.Assuming that obtained rail region has N number of, then area-of-interest is also N number of.
Step 2 is successively handled N number of rail region, remembers that each rail region maximum width is respectively lN, take iron
Each pixel of straight line on the left of rail edge, along sleeper direction to rail extracted region length be lNPixel vectors and composition
Vector Groups { v, v2..., vk, k is the number of pixel vectors.To Vector Groups { v, v2..., vkCluster as 2 subsets, this experiment
It is clustered using Kmeans clustering method, is expressed as { m1, m2..., mpAnd { n1, n2..., nq, p and q are respectively indicated
The number for the pixel vectors that two sons are concentrated, and have p+q=k.Pixel vectors in each subset are respectively constituted into correspondence
Matrix DzAnd DsRow vector, in which:
Dz=[m1, m2..., mp] (1)
Ds=[n1, n2..., nq] (2)
Step 3 is to obtained matrix DzAnd DsLow-rank matrix decomposition is carried out, it can be using Robust Principal Component Analysis (letter
Claim: RPCA) method obtain the interference of no high-frequency information, the consistent low-rank matrix A_z and A_s of background.This algorithms selection is based on adding
Fast proximal end gradient algorithm (referred to as: APG) is solved, and detailed process is as follows:
A it) is relaxed to constraint condition is convex in objective function, obtains Augmented Lagrangian Functions;
B it) is approached using secondary model;
C) derivation and abbreviation, fixed another unknown matrix, solve low-rank matrix A.
Iterative threshold algorithm can also be used by decomposing to low-rank matrix, Conjugate Search Algorithm, augmented vector approach alternating
Direction method etc..Due to the division in step 1 and two to the extraction of area-of-interest and the affiliated subset of pixel so that into
Row low-rank matrix greatly reduces operand when decomposing, shorten operation time.
It is poor to make to matrix D and low-rank matrix A:
E=D-A (3)
Wherein, D=[Dz, Ds], A=[A_z, A_s]
It obtains removing background information and the matrix E containing abnormal foreground target information in area-of-interest.
Step 4 pre-processes matrix E, including mini-value filtering operation and Threshold segmentation, detects foreign matter and determines
Position, detailed process is as follows:
Element in matrix E is traversed, center pixel value assignment is completed most with minimum pixel value in 3 × 3 grids
Small value filtering realizes the denoising effect to high-frequency noise point.
Then Threshold segmentation is carried out to matrix E, threshold is the obtained threshold value of adaptive method Threshold segmentation, the i-th row
The pixel value of i-th column is Eij
Eij=255, ifEij> threshold
Eij=0, ifEij≤threshold
Wherein pixel value is that 255 elements can characterize the foreign matter detected.Pass through the row for being 255 to pixel value in matrix E
Number is searched, and position of the foreign substance information in D_z and D_s can be calculated and return in original image, reaches foreign bodies detection and fixed
The effect of position.
The present invention can be used for taking photo by plane detection and positioning of the rail image to foreign matter, and manpower is driven safely and reduced for train
Consume important in inhibiting.
Embodiment
For the present embodiment is chosen based on representative rail foreign matter image is obtained captured by space base platform, such as Fig. 2
It is shown.Straight-line detection is carried out to source images, acquired results are as shown in figure 3, and carry out screening extraction sense to the straight line being calculated
The rail region of interest.
To region of interesting extraction pixel vectors and clustering processing is carried out, is divided into two subsets of sleeper and stone and to two
The matrix that a subset is constituted carries out low-rank matrix and decomposes to obtain low-rank matrix D, obtains prospect as difference to original matrix and low-rank matrix
Matrix E is simultaneously filtered prospect matrix and Threshold segmentation determines foreign matter position and marks in source images, experimental result such as Fig. 4
It is shown.
The foregoing is merely presently preferred embodiments of the present invention, is not intended to limit the invention, it is all in spirit of the invention and
Within principle, any modification, equivalent replacement, improvement and so on be should all be included in the protection scope of the present invention.
Claims (6)
1. a kind of rail foreign matter detecting method decomposed based on low-rank matrix, it is characterised in that: include the following steps,
Step 1: receiving the Aerial Images of UAV flight's camera shooting and inputting in computer, image is pre-processed, so
Straight-line detection is carried out to the overall situation by line detection algorithm afterwards, obtains straight line group;According to rail length and rail and sleeper it
Between the correlation of included angle of straight line required straight line group is screened, calculate rail edge straight line, divide rail region;
Step 2: rail region is successively taken to be handled, for the left side edge of each rail edge, along sleeper direction to iron
Rail extracted region pixel vectors, cluster pixel vectors group, are separated into two subsets and constitute corresponding classification matrix;
Step 3: carrying out low-rank matrix decomposition under respective type to classification matrix, low-rank matrix is found out;To classification matrix and low
Order matrix acquires matrix E, i.e. foreground target foreign substance information as difference;
Step 4: pre-processing, being positioned to the result after Threshold segmentation and being returned in original image to matrix E, determined with this different
Rail foreign bodies detection is completed in object position in original image.
2. a kind of rail foreign matter detecting method decomposed based on low-rank matrix according to claim 1, it is characterised in that: step
Rapid two specifically,
Successively N number of rail region is handled, remembers that each rail region maximum width is respectively lN, take straight on the left of rail edge
Each pixel of line, along sleeper direction to rail extracted region length be lNPixel vectors and composition Vector Groups v,
v2..., vk, k is the number of pixel vectors;To Vector Groups { v, v2..., vkCluster be 2 subsets, be expressed as { m1,
m2..., mpAnd { n1, n2..., nq, p and q respectively indicate the number for the pixel vectors that two sons are concentrated, and have p+q=k;It will
Pixel vectors in each subset respectively constitute homography DzAnd DsRow vector, in which:
Dz=[m1, m2..., mp] (1)
Ds=[n1, n2..., nq] (2)。
3. a kind of rail foreign matter detecting method decomposed based on low-rank matrix according to claim 2, it is characterised in that: institute
The cluster stated uses Kmeans clustering method.
4. a kind of rail foreign matter detecting method decomposed based on low-rank matrix according to claim 1, it is characterised in that: step
Low-rank matrix is decomposed using based on acceleration proximal end gradient algorithm, iterative threshold algorithm, Conjugate Search Algorithm or increasing described in rapid three
Wide method of Lagrange multipliers alternating direction implicit.
5. a kind of rail foreign matter detecting method decomposed based on low-rank matrix according to claim 1, it is characterised in that: step
Pretreatment described in rapid four includes filtering operation and Threshold segmentation.
6. a kind of rail foreign matter detecting method decomposed based on low-rank matrix according to claim 1, it is characterised in that: step
Rapid four detailed process is as follows:
Element in matrix E is traversed, minimum value is completed to center pixel value assignment with minimum pixel value in 3 × 3 grids
The denoising effect to high-frequency noise point is realized in filtering;
Then Threshold segmentation is carried out to matrix E, threshold is the obtained threshold value of adaptive method Threshold segmentation, the i-th row i-th
The pixel value of column is Eij
Eij=255, if Eij> threshold
Eij=0, if Eij≤threshold
Wherein pixel value is that 255 elements characterize the foreign matter detected;By being searched for 255 line number pixel value in matrix E,
Position of the foreign substance information in D_z and D_s is calculated and returns in original image, achievees the effect that foreign bodies detection and positioning.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201910151851.9A CN109934135B (en) | 2019-02-28 | 2019-02-28 | Rail foreign matter detection method based on low-rank matrix decomposition |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201910151851.9A CN109934135B (en) | 2019-02-28 | 2019-02-28 | Rail foreign matter detection method based on low-rank matrix decomposition |
Publications (2)
Publication Number | Publication Date |
---|---|
CN109934135A true CN109934135A (en) | 2019-06-25 |
CN109934135B CN109934135B (en) | 2020-04-28 |
Family
ID=66986192
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201910151851.9A Active CN109934135B (en) | 2019-02-28 | 2019-02-28 | Rail foreign matter detection method based on low-rank matrix decomposition |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN109934135B (en) |
Cited By (4)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN111582084A (en) * | 2020-04-24 | 2020-08-25 | 北京航空航天大学 | Weak supervision learning-based rail foreign matter detection method and system under empty base view angle |
CN112329604A (en) * | 2020-11-03 | 2021-02-05 | 浙江大学 | Multi-modal emotion analysis method based on multi-dimensional low-rank decomposition |
CN112488056A (en) * | 2020-12-17 | 2021-03-12 | 上海媒智科技有限公司 | Linear track foreign matter intrusion detection method and device based on computer vision |
CN112989931A (en) * | 2021-02-05 | 2021-06-18 | 广州华微明天软件技术有限公司 | Intelligent identification method for foreign matters in subway rail |
Citations (3)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN102737252A (en) * | 2012-06-01 | 2012-10-17 | 西南交通大学 | Method for detecting faults caused by foreign body pollution between electrified railway insulator plates based on affine invariant moment |
CN103745224A (en) * | 2013-12-24 | 2014-04-23 | 西南交通大学 | Image-based railway contact net bird-nest abnormal condition detection method |
WO2018104460A1 (en) * | 2016-12-07 | 2018-06-14 | Siemens Aktiengesellschaft | Method, device and track-bound vehicle, in particular a rail vehicle, for identifying dangerous situations in the track-bound traffic system, in particular in the railway traffic system |
-
2019
- 2019-02-28 CN CN201910151851.9A patent/CN109934135B/en active Active
Patent Citations (3)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN102737252A (en) * | 2012-06-01 | 2012-10-17 | 西南交通大学 | Method for detecting faults caused by foreign body pollution between electrified railway insulator plates based on affine invariant moment |
CN103745224A (en) * | 2013-12-24 | 2014-04-23 | 西南交通大学 | Image-based railway contact net bird-nest abnormal condition detection method |
WO2018104460A1 (en) * | 2016-12-07 | 2018-06-14 | Siemens Aktiengesellschaft | Method, device and track-bound vehicle, in particular a rail vehicle, for identifying dangerous situations in the track-bound traffic system, in particular in the railway traffic system |
Non-Patent Citations (1)
Title |
---|
朱正平等: "基于机器视觉的轨道异物侵限检测方法研究", 《甘肃高师学报》 * |
Cited By (6)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN111582084A (en) * | 2020-04-24 | 2020-08-25 | 北京航空航天大学 | Weak supervision learning-based rail foreign matter detection method and system under empty base view angle |
CN111582084B (en) * | 2020-04-24 | 2022-07-08 | 北京航空航天大学 | Weak supervision learning-based rail foreign matter detection method and system under empty base view angle |
CN112329604A (en) * | 2020-11-03 | 2021-02-05 | 浙江大学 | Multi-modal emotion analysis method based on multi-dimensional low-rank decomposition |
CN112488056A (en) * | 2020-12-17 | 2021-03-12 | 上海媒智科技有限公司 | Linear track foreign matter intrusion detection method and device based on computer vision |
CN112989931A (en) * | 2021-02-05 | 2021-06-18 | 广州华微明天软件技术有限公司 | Intelligent identification method for foreign matters in subway rail |
CN112989931B (en) * | 2021-02-05 | 2022-10-18 | 广州华微明天软件技术有限公司 | Intelligent identification method for foreign matters in subway rail |
Also Published As
Publication number | Publication date |
---|---|
CN109934135B (en) | 2020-04-28 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN109934135A (en) | A kind of rail foreign matter detecting method decomposed based on low-rank matrix | |
Aytekin et al. | Railway fastener inspection by real-time machine vision | |
DE102011117585B4 (en) | Systems and methods for tracking objects | |
CN108549864B (en) | Vehicle-mounted thermal imaging pedestrian detection-oriented region-of-interest filtering method and device | |
Adu-Gyamfi et al. | Automated vehicle recognition with deep convolutional neural networks | |
CN109101924B (en) | Machine learning-based road traffic sign identification method | |
US11132563B2 (en) | Method for identifying objects in an image of a camera | |
CN104951784A (en) | Method of detecting absence and coverage of license plate in real time | |
KR101977052B1 (en) | System for road surface condition investigation using unmanned air vehicle and method using the same | |
CN105488453A (en) | Detection identification method of no-seat-belt-fastening behavior of driver based on image processing | |
CN111768417B (en) | Railway wagon overrun detection method based on monocular vision 3D reconstruction technology | |
CN106548131A (en) | A kind of workmen's safety helmet real-time detection method based on pedestrian detection | |
Ma et al. | Texture classification for rail surface condition evaluation | |
CN111091551A (en) | Method for detecting loss fault of brake beam strut opening pin of railway wagon | |
Suchitra et al. | Detection & classification of arrow markings on roads using signed edge signatures | |
Du et al. | Change detection: The framework of visual inspection system for railway plug defects | |
CN105469099B (en) | Pavement crack detection and identification method based on sparse representation classification | |
CN107341455A (en) | A kind of detection method and detection means to the region multiple features of exotic on night airfield runway road surface | |
CN110728214B (en) | Weak and small figure target detection method based on scale matching | |
Kraisin et al. | Accuracy improvement of a province name recognition on Thai license plate | |
Mery et al. | Automated object recognition in baggage screening using multiple X-ray views | |
CN113639685B (en) | Displacement detection method, device, equipment and storage medium | |
Zheng et al. | Automatic online vision-based inspection system of coupler yoke for freight trains. | |
CN113658089A (en) | Double-data-stream fusion object identification method based on depth camera | |
Bian et al. | Optimal operator space pursuit: A framework for video sequence data analysis |
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 |