CN107169401B - Rail invader detection method based on rail visual feature spectrum - Google Patents
Rail invader detection method based on rail visual feature spectrum Download PDFInfo
- Publication number
- CN107169401B CN107169401B CN201710074119.7A CN201710074119A CN107169401B CN 107169401 B CN107169401 B CN 107169401B CN 201710074119 A CN201710074119 A CN 201710074119A CN 107169401 B CN107169401 B CN 107169401B
- Authority
- CN
- China
- Prior art keywords
- track
- image
- feature
- rail
- alp
- 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.)
- Active
Links
- 230000000007 visual effect Effects 0.000 title claims abstract description 53
- 238000001228 spectrum Methods 0.000 title claims abstract description 46
- 238000001514 detection method Methods 0.000 title claims abstract description 31
- 238000000034 method Methods 0.000 claims abstract description 38
- 238000000605 extraction Methods 0.000 claims abstract description 12
- PXFBZOLANLWPMH-UHFFFAOYSA-N 16-Epiaffinine Natural products C1C(C2=CC=CC=C2N2)=C2C(=O)CC2C(=CC)CN(C)C1C2CO PXFBZOLANLWPMH-UHFFFAOYSA-N 0.000 claims abstract description 11
- 230000008859 change Effects 0.000 claims abstract description 11
- 230000006835 compression Effects 0.000 claims description 29
- 238000007906 compression Methods 0.000 claims description 29
- 239000013598 vector Substances 0.000 claims description 14
- 238000012545 processing Methods 0.000 claims description 8
- 230000009466 transformation Effects 0.000 claims description 7
- 230000000670 limiting effect Effects 0.000 claims description 6
- 230000008569 process Effects 0.000 claims description 6
- 230000006870 function Effects 0.000 claims description 5
- 238000003064 k means clustering Methods 0.000 claims description 5
- 238000010276 construction Methods 0.000 claims description 4
- 230000004044 response Effects 0.000 claims description 4
- 230000002776 aggregation Effects 0.000 claims description 3
- 238000004220 aggregation Methods 0.000 claims description 3
- 238000003708 edge detection Methods 0.000 claims description 3
- 238000005259 measurement Methods 0.000 claims description 3
- 238000004364 calculation method Methods 0.000 claims description 2
- 238000001914 filtration Methods 0.000 claims description 2
- 230000004927 fusion Effects 0.000 claims description 2
- 239000011159 matrix material Substances 0.000 claims description 2
- 238000000513 principal component analysis Methods 0.000 claims description 2
- 238000010586 diagram Methods 0.000 description 12
- 230000002457 bidirectional effect Effects 0.000 description 4
- 238000009826 distribution Methods 0.000 description 2
- 238000002474 experimental method Methods 0.000 description 2
- 238000009434 installation Methods 0.000 description 2
- 238000012544 monitoring process Methods 0.000 description 2
- 238000012805 post-processing Methods 0.000 description 2
- 238000012216 screening Methods 0.000 description 2
- 230000002411 adverse Effects 0.000 description 1
- 230000007797 corrosion Effects 0.000 description 1
- 238000005260 corrosion Methods 0.000 description 1
- 230000007547 defect Effects 0.000 description 1
- 230000008030 elimination Effects 0.000 description 1
- 238000003379 elimination reaction Methods 0.000 description 1
- 238000005516 engineering process Methods 0.000 description 1
- 230000007613 environmental effect Effects 0.000 description 1
- 230000003628 erosive effect Effects 0.000 description 1
- 238000005286 illumination Methods 0.000 description 1
- 238000012423 maintenance Methods 0.000 description 1
- 230000003287 optical effect Effects 0.000 description 1
- 230000036961 partial effect Effects 0.000 description 1
- 230000000750 progressive effect Effects 0.000 description 1
- 230000002829 reductive effect Effects 0.000 description 1
- 238000006467 substitution reaction Methods 0.000 description 1
- 238000012360 testing method Methods 0.000 description 1
- 238000000844 transformation Methods 0.000 description 1
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V20/00—Scenes; Scene-specific elements
- G06V20/10—Terrestrial scenes
- G06V20/13—Satellite images
-
- B—PERFORMING OPERATIONS; TRANSPORTING
- B61—RAILWAYS
- B61L—GUIDING RAILWAY TRAFFIC; ENSURING THE SAFETY OF RAILWAY TRAFFIC
- B61L23/00—Control, warning or like safety means along the route or between vehicles or trains
- B61L23/04—Control, warning or like safety means along the route or between vehicles or trains for monitoring the mechanical state of the route
- B61L23/041—Obstacle detection
-
- 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
Landscapes
- Engineering & Computer Science (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Multimedia (AREA)
- Theoretical Computer Science (AREA)
- Mechanical Engineering (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Astronomy & Astrophysics (AREA)
- Remote Sensing (AREA)
- Image Analysis (AREA)
Abstract
The invention provides a rail invader detection method based on a rail visual feature spectrum. Cruising and shooting images at a certain height, angle and speed along a track line, establishing a characteristic image library aiming at 6 types of image shooting and affine change by utilizing image data, and establishing a track visual characteristic spectrum based on the track characteristic image library; and performing characteristic matching by using the rail visual characteristic spectrum and the characteristics of the rail image to be detected, and judging whether a rail invader exists in the rail image to be detected according to a matching result. The orbit vision characteristic spectrum theory and the invader detection method based on the orbit vision characteristic spectrum theory can be applied to an invader detection system based on an unmanned aerial vehicle, and the problems of poor reliability and high cost of the traditional detection system are solved. The method has high feature extraction and matching speed, can effectively detect the invader in the track, and can be applied to an invader detection system based on the unmanned aerial vehicle.
Description
Technical Field
The invention relates to the technical field of rail line environment safety detection, in particular to an unmanned aerial vehicle-mounted rail invader detection method based on a rail visual feature spectrum.
Background
In recent years, rail traffic mileage, particularly railway mileage under western complex geomorphic environments, is increasing, the construction of high-speed railways in asia-europe, and severe international safety situation influence, so that a rapid identification method and technology for rail line invaders, which can be used for an unmanned airborne invader identification device, are urgently needed.
In the existing track safety detection system in China, track line conditions are mainly checked in a mode of installing monitoring cameras at fixed points and observing operators on duty in real time. In the mode, cameras need to be installed on the track line in a point-type distribution mode, the installation and maintenance cost is high, and the method cannot be popularized and applied under the influence of severe and complex geomorphic environments such as western mountain areas, unmanned areas and the like; the safety state of the current track line is judged by staring and controlling the monitoring video for a long time by the staff, and potential safety hazards are easily caused by factors such as fatigue negligence and the like.
In the intelligent identification algorithm for the rail invaders based on the fixed camera, researched by researchers at home and abroad, the existence of the invaders is detected by mostly using image pixel characteristics as templates and adopting a background difference or background modeling method. The method is easily influenced by the surrounding environment, the template needs to be updated in real time, and the method cannot be applied to an invader detection system of an unmanned aerial vehicle-mounted mobile camera platform.
Disclosure of Invention
The embodiment of the invention provides a track invader detection method based on a track visual feature spectrum, which is used for effectively detecting the track line environment safety situation through an unmanned aerial vehicle-mounted camera.
In order to achieve the purpose, the invention adopts the following technical scheme.
A rail invader detection method based on rail visual feature spectrum specifically comprises the following steps:
cruising and shooting images at a certain height, angle and speed along a track line, establishing a track characteristic diagram aiming at 6 image shooting and affine changes of rotation, brightness, visual angle, definition, scale and compression ratio through actual acquisition methods such as different weather, different angles and the like or post-processing methods such as image transformation and the like, extracting characteristic description and compressing codes based on the track characteristic diagram, and fusing geographical position information to construct a track visual characteristic spectrum;
and performing feature matching on the to-be-detected track image and a track visual feature spectrum in the same geographical position, and judging whether a track invader exists in the to-be-detected track image according to a matching result.
Further, the method comprises the steps of cruising along the track line at a certain height, angle and speed, shooting images, establishing a track characteristic diagram aiming at 6 types of image shooting and affine change of rotation, brightness, visual angle, definition, scale and compression ratio, and constructing a track visual characteristic spectrum, wherein the specific process comprises the following steps:
(1) the unmanned aerial vehicle carrying the cloud deck camera is utilized to cruise and shoot images at a certain height, angle and speed along a track line, a track characteristic template interval is defined according to the specific visual angle of the camera and the track length in the images, and track image data and positioning information are collected. By using an unmanned aerial vehicle acquisition mode under different weather and different angles and by post-processing modes such as image transformation and the like, the track characteristic diagrams under different geographic positions are established aiming at 6 kinds of shooting and affine changes of rotation, brightness, visual angle, definition, scale and compression ratio. The specific algorithm comprises the following steps:
1) computing pixel gradients
Horizontal direction template ▽ using Sobel operatorxAnd a vertical direction form ▽yCalculating a horizontal gradient value for each pixel in the orbit image data
And vertical gradient value
As shown in formula (1) and formula (2):
in the formula (2), f (x, y) is image data;
2) calculating gradient direction
Solving the horizontal square gradient from the horizontal gradient and the vertical gradient of each pixel point in the image
And square gradient in vertical direction
As shown in formula (3):
for each sub-block w × w of the image sliding window centered on pixel (i, j), the mean squared gradient value Δ is calculatedx(i, j) and Δy(i, j) as shown in formula (4):
calculating the average direction of the sub-block gradient
Is represented by formula (5):
3) rail positioning
And (3) extracting a gradient direction field from the track image according to the formulas (3), (4) and (5) to obtain a track gradient map, and performing thresholding treatment on the track gradient map, wherein the gradient direction screening range is 40-140 degrees. Finally, the method of Canny edge detection, connected region screening, image corrosion and expansion and the like is applied to obtain the position of the track in the image;
4) extraction of interest
Defining and extracting a track interested area in the track image according to the track position and the railway limit in the track image, wherein the track interested area comprises a railway limit area including a track and a roadbed and a buffer area extending outwards from the railway limit area, calculating the distance between each pixel in the track image and the nearest track pixel, and marking the area with the distance less than or equal to the track width as the railway interested area;
5) ALP feature extraction
Extracting ALP characteristics of each orbit interesting region and carrying out compression coding;
6) construction of orbit visual characteristic spectrum
And (3) taking the geographical position of each orbit interesting area as an interval index, and carrying out vector fusion on ALP features extracted from the orbit feature map between the orbits interesting areas to construct and finish the orbit visual feature spectrum.
Further, the extracting the ALP feature of each orbit roi includes:
①, constructing a Gaussian scale space to solve the extreme point of the scale space, wherein a Gaussian kernel is the only linear convolution kernel for realizing scale transformation, and for a scale parameter sigma, the two-dimensional Gaussian kernel is expressed as formula (6):
g(x,y,σ)＝1/(2πσ2)·exp(-(x2+y2)/(2σ2)) (6)
to construct the image pyramid, in each sub-octave, the original image is filtered using 4 increasing scale parameters σ of gaussian kernel and discrete laplacian kernel as shown in equation (7):
wherein g (x, y, σ) is a discrete Gaussian function,
for discrete laplacian, the symbol denotes convolution;
to find the extreme points in the pyramid, a third order polynomial filter result as shown in equation (8) is used;
p(x,y,σ)＝α3(x,y)σ3+α2(x,y)σ2+α1(x,y)σ+α0(x,y) (8)
wherein (A), (B), (Cx, y) are pixel coordinates, α0(x,y):α3(x, y) is calculated by weighted summation of images after Gaussian filtering, and p (x, y, sigma) is a response value of each pixel point;
within each sub-octave, calculating the response p (x, y, sigma) of each pixel of each image to obtain candidate extreme points (x, y) and corresponding scales sigma*Remapping candidate extreme points to sub-pixel precision, eliminating repeated extreme points, extracting ALP feature points, and allocating a main direction to each feature point according to a histogram of regional gradient amplitude of each ALP feature point;
② local feature selection, according to the scale σ of ALP feature points*Direction θ, distance d from the center of the image, peak D, Hessian matrix squared trace ratio ρ, second derivative of the scale space function pσσ6 feature parameters are calculated, correlation calculation is carried out as shown in a formula (9), feature points with correlation larger than a set threshold value are screened out and used as finally extracted ALP feature points;
r(σ*,θ,d,D,ρ,pσσ)＝f1(σ*)·f2(θ)·f3(d)·f4(D)·f5(ρ)·f6(pσσ) (9)
③ local feature description, in a 4 x 4 region centered on the ALP feature point, 8 directional gradient histograms are calculated for each region, these gradient histograms are concatenated to form a 128-dimensional vector, and each ALP feature point is described as a 128-dimensional vector.
Further, the compression encoding of the ALP feature of the extracted orbit roi includes:
① local feature aggregation, normalizing the local feature description of each 128-dimensional vector, and extracting 32-dimensional principal vectors through principal component analysis, extracting ALP features from 6 case distributions in the orbit feature map, and finally forming 6 x 32-dimensional feature vectors;
② local feature compression, which is divided into local feature description compression and local feature position compression, wherein in the local feature description compression, a low-complexity transform coding scheme is adopted for feature description compression, and in the local feature position compression, a position histogram coding scheme is adopted for representing features into binary bit streams;
③ global feature description, adopts a scalable compression Fisher vector coding scheme to carry out generalized description on image features.
Further, the track visual characteristic spectrum is subjected to characteristic matching with a track image to be detected, and whether a track invader exists in the track image to be detected is judged according to a matching result, and the specific process comprises the following steps:
when the track line is inspected, the track image to be detected is shot in real time according to the same shooting conditions in the process of constructing the track visual characteristic spectrum, ALP characteristics in the region of interest of the track image are extracted in each track section, and the characteristics extracted in real time are matched with the track visual characteristic spectrum at the same geographic position;
after completing feature matching between ALP features of the image shot in real time and the orbit visual feature spectrum at the same position, eliminating mismatching feature points by using a RANSAC consistency detection method. Then, determining whether an invader exists in the real-time graph by adopting a threshold method, and defining the matching rate as an equation (10);
in the formula, TotalNum represents the total number of feature points, and matchNum represents the number of feature points successfully matched. If the ratio is larger than a preset threshold value, determining that an invader exists; if the ratio is less than the threshold, then it is considered normal;
and when the existence of the invader is detected, classifying the gathered positions of most of the feature points in the unmatched residual feature point set by using a K-means clustering algorithm, and taking the gathered positions of most of the feature points as the range of the invader.
Further, the ALP feature matching of the image features extracted in real time and the orbit visual feature spectrum at the same geographic position mainly comprises:
and global description matching, namely describing ALP feature matching degree between the orbit image shot in real time and the orbit visual feature image at the geographic position by adopting similarity measurement based on Hamming distance.
And local description matching, namely performing ALP feature matching on the track image shot in real time and the track visual feature image at the geographic position by using a nearest neighbor matching algorithm.
And a bidirectional matching strategy is adopted during matching, so that the matching accuracy is improved. And bidirectional matching, namely performing characteristic matching on the characteristic description R in the characteristic spectrum and the ALP characteristic descriptor Q extracted from the image to be matched from two directions of R → Q and Q → R respectively.
According to the technical scheme provided by the embodiment of the invention, the orbit vision characteristic spectrum theory and the intruder detection method based on the orbit vision characteristic spectrum theory provided by the embodiment of the invention can be applied to an intruder detection system based on an unmanned aerial vehicle, and the problems of poor reliability and high cost of the traditional detection system are solved. The method has high feature extraction and matching speed, and can autonomously and effectively detect the invader in the track.
Additional aspects and advantages of the invention will be set forth in part in the description which follows, and in part will be obvious from the description, or may be learned by practice of the invention.
Drawings
In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings needed to be used in the description of the embodiments will be briefly introduced below, and it is obvious that the drawings in the following description are only some embodiments of the present invention, and it is obvious for those skilled in the art to obtain other drawings based on these drawings without inventive labor.
Fig. 1 is a processing flow chart of a rail invader detecting method based on rail visual feature spectrum according to an embodiment of the present invention;
fig. 2 is a schematic diagram illustrating a track feature pattern according to an embodiment of the present invention;
FIG. 3 is a schematic diagram of a track positioning system according to an embodiment of the present invention;
FIG. 4 is a diagram of a region of interest of a track according to an embodiment of the present invention;
FIG. 5 is a diagram illustrating the result of ALP feature extraction according to an embodiment of the present invention;
fig. 6 is an image of rail intrusion according to an embodiment of the present invention;
fig. 7 is a schematic diagram of an intruder range extraction according to an embodiment of the present invention.
Detailed Description
Reference will now be made in detail to embodiments of the present invention, examples of which are illustrated in the accompanying drawings, wherein like reference numerals refer to the same or similar elements or elements having the same or similar function throughout. The embodiments described below with reference to the accompanying drawings are illustrative only for the purpose of explaining the present invention, and are not to be construed as limiting the present invention.
As used herein, the singular forms "a", "an", "the" and "the" are intended to include the plural forms as well, unless the context clearly indicates otherwise. It will be further understood that the terms "comprises" and/or "comprising," when used in this specification, specify the presence of stated features, integers, steps, operations, elements, and/or components, but do not preclude the presence or addition of one or more other features, integers, steps, operations, elements, components, and/or groups thereof. It will be understood that when an element is referred to as being "connected" or "coupled" to another element, it can be directly connected or coupled to the other element or intervening elements may also be present. Further, "connected" or "coupled" as used herein may include wirelessly connected or coupled. As used herein, the term "and/or" includes any and all combinations of one or more of the associated listed items.
It will be understood by those skilled in the art that, unless otherwise defined, all terms (including technical and scientific terms) used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention belongs. It will be further understood that terms, such as those defined in commonly used dictionaries, should be interpreted as having a meaning that is consistent with their meaning in the context of the prior art and will not be interpreted in an idealized or overly formal sense unless expressly so defined herein.
For the convenience of understanding the embodiments of the present invention, the following description will be further explained by taking several specific embodiments as examples in conjunction with the drawings, and the embodiments are not to be construed as limiting the embodiments of the present invention.
Example one
The invention provides a track visual characteristic spectrum theory and a construction method thereof, and provides a characteristic matching intruder detection method based on the theory, aiming at the defects that the existing track intruder detection system can only carry out fixed-point detection, is greatly influenced by environmental change and has low detection speed, and the characteristics of small load of an unmanned aerial vehicle and high real-time performance of the detection system.
The processing flow of the rail invader detection method based on the rail visual feature spectrum provided by the embodiment of the invention is shown in fig. 1, and mainly comprises the following processing procedures:
(1) the method comprises the following steps of constructing an orbit vision characteristic spectrum, and defining the orbit vision characteristic spectrum as follows: and taking the geographic information as an index, respectively extracting image characteristics of image rotation, brightness, visual angle, definition, scale, compression ratio and the like and affine change at the same position of the track, and coding and compressing the image characteristics. The method mainly comprises the steps of carrying out various affine transformation processing on an orbit image, extracting an orbit region of interest, extracting ALP characteristics, compressing and generating a visual characteristic spectrum;
(2) the intrusion object detection part mainly adopts ALP feature matching, RANSAC-based mismatching point elimination and K-means clustering methods. Experimental results show that the method provided by the invention has high feature extraction and matching speed, and can autonomously and effectively extract the range of the rail invader.
The feature matching stage mainly comprises global description matching and local description matching, and the specific implementation process is as follows:
1) global description matching is achieved by quickly calculating the weighted correlation of the 2 image global description features. The method adopts similarity measurement based on Hamming distance to describe the matching degree of two images. The similarity definition is shown in (1).
In the formula
Is a feature binary vector. Ha (,) represents the hamming distance. w is aHaAre coefficients learned from a database. Setting a threshold τ if sQ,RIf the value is greater than tau, the features are matched, otherwise the features are not matched.
2) In the local description matching, a nearest neighbor matching algorithm is used for performing feature matching on the image to be matched and the reference image. After matching is completed, RANSAC consistency check is carried out, and wrong matching points are removed.
3) And a bidirectional matching strategy is adopted during matching, so that the matching accuracy is improved. And bidirectional matching, namely performing characteristic matching on the characteristic description R in the characteristic spectrum and the ALP characteristic descriptor Q extracted from the image to be matched from two directions of R → Q and Q → R respectively.
And after matching of the ALP characteristics of the real-time shot image and the ALP characteristics of the orbit visual characteristic image at the same position, determining whether an invader exists in the real-time image by adopting a threshold method. Defining the matching rate ratio as formula (2);
in the formula, TotalNum represents the total number of feature points, and matchNum represents the number of feature points successfully matched. If the ratio is larger than a preset threshold value, determining that an invader exists; if ratio is less than the threshold, then it is considered normal.
When the existence of the invader is detected, the unmatched feature point set comprises the feature points of the suspicious invader and the noise feature points existing in the surrounding environment irrelevant to the invader. The characteristic points on the invader are dense, the noise points are scattered, and the interference characteristic points are removed by utilizing a threshold value method through calculating the Euclidean distance between each characteristic point and the rest points.
And finally, classifying the positions of most feature point aggregates in the residual feature point sets by using a K-means clustering algorithm, namely extracting the approximate range of the suspicious invaders.
Example two
Fig. 2 is a schematic diagram of a track feature sample constructed by acquiring an image of a certain track section through an actual scene or performing various affine transformations, wherein a rotation feature image (a) is obtained by a camera at a fixed position through self-rotation shooting; the characteristic images (b) (d) of the luminance change and the sharpness change are obtained by taking images by changing the aperture and the focal length of the camera; the characteristic images (c) and (e) of the visual angle change and the scale change are respectively obtained by shooting images at different visual angles and different shooting distances in the same place; the compression ratio change characteristic image (f) is obtained by processing the JPEG2000 image compression standard at different compression ratios. Fig. 3 is an image of positioning a track, where (a) is an original track image captured by an unmanned aerial vehicle, (b) is a gradient image obtained by calculating a Sobel gradient direction field and performing threshold processing, and (c) is a track position image obtained by image erosion, Canny edge detection, and image area filling. Fig. 4 is an orbit image after region of interest extraction, and ALP descriptor extraction is performed on the basis of the image. ALP feature descriptor schematic the ALP feature descriptor is shown in fig. 5, where the center of each circle represents the feature point and the radius of the circle represents the feature point dimension, where only a portion of the ALP feature points are shown. Feature vector aggregation and compression are carried out on the ALP features extracted after various shooting and affine transformation, geographic position information is fused, and finally an orbit visual feature spectrum is constructed. Fig. 6 is an orbit image of an intruding object in the presence of the unmanned aerial vehicle. FIG. 7 is a schematic diagram of the range of an intruding object, which is detected after feature extraction, feature matching and K-means clustering. The experimental result shows that the method provided by the invention can effectively detect the rail invader. Under test experiment environments of Intel (R) core (TM) i5-3470 CPU, 8GB memory, Visual Studio 2013 and OpenCV 3.0, the time used by the experiment operation result is about 1.2s, and the requirement of unmanned aerial vehicle cruise detection real-time performance at the speed of 60km/h is met.
In summary, the orbit visual characteristic spectrum theory and the invader detection method based on the orbit visual characteristic spectrum theory provided by the embodiment of the invention take the ALP compression characteristic of the orbit image as the detection template, thereby greatly reducing the storage capacity and meeting the requirements of an embedded detection system; ALP feature matching speed is fast, satisfies unmanned aerial vehicle speed requirement of cruising, consequently can apply to in the invader detecting system based on unmanned aerial vehicle. Meanwhile, the ALP characteristics are stable, the influence of illumination, scale and the like is small, and the detection reliability is high. Only need an unmanned aerial vehicle can detect whole road section, need not fixed point installation surveillance camera along the line, greatly reduced the cost has very important meaning to track detection under adverse circumstances such as western mountain area, unmanned area.
Those of ordinary skill in the art will understand that: the figures are merely schematic representations of one embodiment, and the blocks or flow diagrams in the figures are not necessarily required to practice the present invention.
From the above description of the embodiments, it is clear to those skilled in the art that the present invention can be implemented by software plus necessary general hardware platform. Based on such understanding, the technical solutions of the present invention may be embodied in the form of a software product, which may be stored in a storage medium, such as ROM/RAM, magnetic disk, optical disk, etc., and includes instructions for causing a computer device (which may be a personal computer, a server, or a network device, etc.) to execute the method according to the embodiments or some parts of the embodiments.
The embodiments in the present specification are described in a progressive manner, and the same and similar parts among the embodiments are referred to each other, and each embodiment focuses on the differences from the other embodiments. In particular, for apparatus or system embodiments, since they are substantially similar to method embodiments, they are described in relative terms, as long as they are described in partial descriptions of method embodiments. The above-described embodiments of the apparatus and system are merely illustrative, and the units described as separate parts may or may not be physically separate, and the parts displayed as units may or may not be physical units, may be located in one place, or may be distributed on a plurality of network units. Some or all of the modules may be selected according to actual needs to achieve the purpose of the solution of the present embodiment. One of ordinary skill in the art can understand and implement it without inventive effort.
The above description is only for the preferred embodiment of the present invention, but the scope of the present invention is not limited thereto, and any changes or substitutions that can be easily conceived by those skilled in the art within the technical scope of the present invention are included in the scope of the present invention. Therefore, the protection scope of the present invention shall be subject to the protection scope of the claims.
Claims (5)
1. A rail invader detection method based on rail visual feature spectrum is characterized by comprising the following steps:
cruising along a track line at a certain height, angle and speed, shooting images, establishing a track characteristic image library aiming at 6 types of image shooting and affine change of rotation, brightness, visual angle, definition, scale and compression ratio by using image data, and establishing a track visual characteristic spectrum based on the track characteristic image library;
performing feature matching on the rail visual feature spectrum and a rail image to be detected, and judging whether a rail invader exists in the rail image to be detected according to a matching result;
the method comprises the following steps of cruising along a track line at a certain height, angle and speed, shooting images, establishing a track characteristic image library aiming at 6 types of image shooting and affine change of rotation, brightness, visual angle, definition, scale and compression ratio by utilizing image data, and establishing a track visual characteristic spectrum based on the track characteristic image library, wherein the track visual characteristic spectrum comprises the following steps:
the method comprises the following steps of cruising and shooting images at a certain height, angle and speed along a track line by using an unmanned aerial vehicle carrying a cloud platform camera, dividing a track characteristic template interval according to the length of a track in the images, establishing a track characteristic image library aiming at 6 shooting and affine changes of rotation, brightness, a visual angle, definition, a scale and a compression ratio according to image data, and establishing a track visual characteristic spectrum based on the track characteristic image library, wherein the processing process comprises the following 6 processing steps:
1) computing pixel gradients
Horizontal direction template applying Sobel operator
And a vertical direction template
Calculating a horizontal gradient value for each pixel in the orbit image data
And vertical gradient value
As shown in formula (1) and formula (2):
in the formula (2), f (x, y) is image data;
2) calculating the directional field
Solving horizontal square gradient by gradient operator of each pixel point in image data
And square gradient in vertical direction
As shown in formula (3):
for each one withThe image sliding window sub-block w x w, centered on pixel (i, j), calculates the mean squared gradient value Δx(i, j) and Δy(i, j) as shown in formula (4):
calculating the average direction of the sub-block gradient
Is represented by formula (5):
3) rail positioning
Extracting a direction field from the track image according to the formulas (3), (4) and (5), obtaining a gradient image and a gradient value of the track image according to the direction field, and performing thresholding on the track image by taking the gradient value of the track image as a threshold value;
removing background influence, and obtaining the position of the track in the track image by Canny edge detection;
4) boundary extraction
Extracting a rail bound area in the rail image according to the position of the rail in the rail image and the railway bound definition, wherein the rail bound area comprises the rail, a railway bed and a buffer area extending outwards from the rail bound area, calculating the distance between each pixel in the rail image and the nearest rail pixel, and marking the area with the distance smaller than or equal to the width of the rail as the rail bound area;
5) ALP feature extraction
Extracting an ALP feature of each track limit area, and carrying out compression coding on the extracted ALP feature of the track limit area;
6) construction of orbit visual characteristic spectrum
And taking the geographical position of each track limiting area as an interval index, fusing the geographical position of each track limiting area and the ALP characteristic of each track limiting interval, and integrating the fusion characteristics of all track limiting intervals of the track segment to construct a finished track visual characteristic spectrum.
2. The method of claim 1, wherein said extracting ALP features for each track bounding area comprises:
①, constructing a Gaussian scale space to solve the extreme point of the scale space, wherein a Gaussian kernel is the only linear convolution kernel for realizing scale transformation, and for a scale parameter delta, the two-dimensional Gaussian kernel is expressed as shown in formula (6):
g(x,y,σ)＝1/(2πσ2)·exp(-(x2+y2)/(2σ2)) (6)
to construct the image pyramid, the original image is filtered using 4 increasing scale parameters δ of gaussian kernel and discrete laplacian kernel in each sub-octave, as shown in equation (7):
wherein g (x, y, delta) is a discrete Gaussian function,
for discrete laplacian, the symbol denotes convolution;
to find the extreme points in the pyramid, a third order polynomial filter result as shown in equation (8) is used;
p(x,y,δ)＝α3(x,y)δ3+α2(x,y)δ2+α1(x,y)δ+α0(x,y) (8)
where (x, y) is the pixel coordinate, δ is the image scale, α0(x,y):α3(x, y) is calculated by weighted summation of images after Gaussian filtering, and p (x, y, delta) is a response value of each pixel point;
within each sub octave, calculating the response p (x, y, delta) of each pixel of each image to obtain candidate extreme points (x, y) and corresponding scales delta*Remapping candidate extreme points to sub-imagesPixel precision, eliminating repeated extreme points, extracting ALP feature points, and distributing a main direction to each feature point according to a histogram of regional gradient amplitude of each ALP feature point;
② local feature selection, based on the ALP feature point dimension delta*Direction θ, distance d from the center of the image, peak D, Hessian matrix squared trace ratio ρ, second derivative of the scale space function pδδ6 feature parameters are calculated, correlation calculation is carried out as shown in a formula (9), feature points with correlation larger than a set threshold value are screened out and used as finally extracted ALP feature points;
r(δ*,θ,d,D,ρ,pδδ)＝f1(δ*)·f2(θ)·f3(d)·f4(D)·f5(ρ)·f6(pδδ) (9)
wherein f isnThe (—) represents a quantized correlation coefficient, and n is set in advance to 1 to 6;
③ local feature description, in a 4 x 4 region centered on the ALP feature point, 8 directional gradient histograms are calculated for each region, these gradient histograms are concatenated to form a 128-dimensional vector, and each ALP feature point is described as a 128-dimensional vector.
3. The method of claim 2, wherein said compressively encoding the ALP characteristics of the extracted track bounding area comprises:
① local feature aggregation, normalizing the local feature description of each 128-dimensional vector, and extracting a 32-dimensional principal vector through principal component analysis;
② local feature compression, which is divided into local feature description compression and local feature position compression, wherein in the local feature description compression, a low-complexity transform coding scheme is adopted for feature description compression, and in the local feature position compression, a position histogram coding scheme is adopted for representing features into binary bit streams;
③ global feature description, adopts a scalable compression Fisher vector coding scheme to carry out generalized description on image features.
4. The method according to any one of claims 1 to 3, wherein the step of performing feature matching on the rail visual feature spectrum and the rail image to be detected and judging whether a rail invader exists in the rail image to be detected according to a matching result comprises the steps of:
when the track line is inspected, the track image to be detected is shot in real time according to the same shooting conditions in the process of constructing the track visual characteristic spectrum, the ALP characteristics of the track image shot in real time in the track limit range are extracted in each track interval, and the ALP characteristics of the track image shot in real time and the track visual characteristic spectrum at the geographic position are matched;
after completing feature matching between ALP features of a real-time shot image and orbit visual feature spectrums at the same position in a template feature library, determining whether an invader exists in the real-time image by adopting a threshold method, and defining a matching rate as an equation (10):
in the formula, TotalNum represents the total number of feature points, and matchNum represents the number of feature points successfully matched;
if the ratio is larger than a preset threshold value, determining that an invader exists; if the ratio is less than the threshold, then it is considered normal;
and when the existence of the invader is detected, classifying the gathered positions of most of the feature points in the unmatched residual feature point set by using a K-means clustering algorithm, and taking the gathered positions of most of the feature points as the range of the invader.
5. The method of claim 4, wherein ALP feature matching the real-time captured orbit image with the orbit vision feature spectrum at the geographic location comprises:
global description matching, namely describing ALP feature matching degree between the orbit image shot in real time and the orbit visual feature image at the geographic position by adopting similarity measurement based on Hamming distance;
and local description matching, namely performing ALP feature matching on the track image shot in real time and the track visual feature image at the geographic position by using a nearest neighbor matching algorithm.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201710074119.7A CN107169401B (en) | 2017-02-10 | 2017-02-10 | Rail invader detection method based on rail visual feature spectrum |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201710074119.7A CN107169401B (en) | 2017-02-10 | 2017-02-10 | Rail invader detection method based on rail visual feature spectrum |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN107169401A CN107169401A (en) | 2017-09-15 |
| CN107169401B true CN107169401B (en) | 2020-05-05 |
Family
ID=59849476
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201710074119.7A Active CN107169401B (en) | 2017-02-10 | 2017-02-10 | Rail invader detection method based on rail visual feature spectrum |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN107169401B (en) |
Families Citing this family (11)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US11014667B2 (en) * | 2017-10-18 | 2021-05-25 | Bombardier Transportation Gmbh | Rail vehicle and on-board safety drone |
| JP6776213B2 (en) * | 2017-11-01 | 2020-10-28 | 株式会社東芝 | Orbit identification device |
| AU2019218475B2 (en) * | 2018-02-06 | 2022-11-24 | Matisa Materiel Industriel S.A. | Methods for locating points or lines of interest on a railway track, positioning and driving an intervention machine on a railway track |
| CN108550139A (en) * | 2018-03-19 | 2018-09-18 | 北京航空航天大学 | Rail foreign matter detecting method and device based on multi-level depth characteristic |
| CN109145715B (en) * | 2018-07-02 | 2020-10-30 | 北京航空航天大学 | Air-based pedestrian boundary-crossing detection method, device and system for rail transit |
| CN110712658A (en) * | 2019-10-18 | 2020-01-21 | 宋秀梅 | Warning platform for abnormal data of front road |
| CN112526154B (en) * | 2020-11-21 | 2021-09-03 | 西安交通大学 | Unmarked measuring method for motion of circular template matching rotating structure under computer vision |
| CN114220044B (en) * | 2021-11-23 | 2022-07-29 | 慧之安信息技术股份有限公司 | River course floater detection method based on AI algorithm |
| CN114241280B (en) * | 2022-02-23 | 2022-05-27 | 广东电网有限责任公司东莞供电局 | Dynamic identification method for background picture image of master station scheduling system |
| CN115052109B (en) * | 2022-08-16 | 2022-11-29 | 北京理工大学 | Target positioning method and system based on multiple types of cameras |
| CN117011322B (en) * | 2023-09-26 | 2024-02-23 | 天津七一二移动通信有限公司 | Railway intrusion object identification method based on image |
Citations (8)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN101587619A (en) * | 2008-05-21 | 2009-11-25 | 上海新联纬讯科技发展有限公司 | Detection method for intrusion object based on video monitoring |
| CN102496232A (en) * | 2011-12-12 | 2012-06-13 | 杭州市电力局 | Transmission facility monitoring method and system |
| CN103280052A (en) * | 2013-05-15 | 2013-09-04 | 重庆大学 | Intrusion detection method applied in intelligent video monitoring of long-distance railway lines |
| CN103978996A (en) * | 2014-05-15 | 2014-08-13 | 南京理工大学 | High speed railway environmental information collection, assessment and early warning device and method |
| CN104331910A (en) * | 2014-11-24 | 2015-02-04 | 沈阳建筑大学 | Track obstacle detection system based on machine vision |
| CN105426894A (en) * | 2015-11-06 | 2016-03-23 | 中国铁道科学研究院 | Image detection method and device for railroad plug nails |
| CN105799740A (en) * | 2016-03-08 | 2016-07-27 | 浙江大学 | Automatic detecting and early warning method for track foreign matter invasion based on Internet of Things technology |
| CN105915846A (en) * | 2016-04-26 | 2016-08-31 | 成都通甲优博科技有限责任公司 | Monocular and binocular multiplexed invading object monitoring method and system |
-
2017
- 2017-02-10 CN CN201710074119.7A patent/CN107169401B/en active Active
Patent Citations (8)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN101587619A (en) * | 2008-05-21 | 2009-11-25 | 上海新联纬讯科技发展有限公司 | Detection method for intrusion object based on video monitoring |
| CN102496232A (en) * | 2011-12-12 | 2012-06-13 | 杭州市电力局 | Transmission facility monitoring method and system |
| CN103280052A (en) * | 2013-05-15 | 2013-09-04 | 重庆大学 | Intrusion detection method applied in intelligent video monitoring of long-distance railway lines |
| CN103978996A (en) * | 2014-05-15 | 2014-08-13 | 南京理工大学 | High speed railway environmental information collection, assessment and early warning device and method |
| CN104331910A (en) * | 2014-11-24 | 2015-02-04 | 沈阳建筑大学 | Track obstacle detection system based on machine vision |
| CN105426894A (en) * | 2015-11-06 | 2016-03-23 | 中国铁道科学研究院 | Image detection method and device for railroad plug nails |
| CN105799740A (en) * | 2016-03-08 | 2016-07-27 | 浙江大学 | Automatic detecting and early warning method for track foreign matter invasion based on Internet of Things technology |
| CN105915846A (en) * | 2016-04-26 | 2016-08-31 | 成都通甲优博科技有限责任公司 | Monocular and binocular multiplexed invading object monitoring method and system |
Non-Patent Citations (4)
| Title |
|---|
| 《Affine scale space for viewpoint invariant keypoint detection》;Biao Zhao等;《2015 IEEE 17th International Workshop on Multimedia Signal Processing (MMSP)》;20151231;第1-6页 * |
| 《基于智能视频技术的铁路入侵检测》;习可;《中国优秀硕士学位论文全文数据库 信息科技辑》;20110315(第3期);第4章第1段，第4.2节第2段，第4.4.2节第1段 * |
| 《基于高斯混合模型的铁路入侵物体目标识别方法》;董宏辉等;《中国铁道科学》;20110331;第32卷(第2期);第131-135页 * |
| 《视频监控铁路限界内逗留物体的检测方法》;于革等;《中国铁道科学》;20130731;第34卷(第4期);第105-109页 * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN107169401A (en) | 2017-09-15 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN107169401B (en) | Rail invader detection method based on rail visual feature spectrum | |
| Dhiman et al. | Pothole detection using computer vision and learning | |
| Liu et al. | A review of applications of visual inspection technology based on image processing in the railway industry | |
| Luvizon et al. | A video-based system for vehicle speed measurement in urban roadways | |
| CN106683119B (en) | Moving vehicle detection method based on aerial video image | |
| Lin et al. | Vaid: An aerial image dataset for vehicle detection and classification | |
| CN106845364B (en) | Rapid automatic target detection method | |
| Bedruz et al. | Real-time vehicle detection and tracking using a mean-shift based blob analysis and tracking approach | |
| Yang et al. | Vehicle detection methods from an unmanned aerial vehicle platform | |
| CN117492026B (en) | Railway wagon loading state detection method and system combined with laser radar scanning | |
| Liu et al. | Vehicle detection from aerial color imagery and airborne LiDAR data | |
| CN113012215A (en) | Method, system and equipment for space positioning | |
| CN110636248B (en) | Target tracking method and device | |
| Ryu et al. | Feature-based pothole detection in two-dimensional images | |
| CN113158954A (en) | Automatic traffic off-site zebra crossing area detection method based on AI technology | |
| Liu et al. | Automated vehicle wheelbase measurement using computer vision and view geometry | |
| CN111339824A (en) | Road surface sprinkled object detection method based on machine vision | |
| CN116520351A (en) | Train state monitoring method, system, storage medium and terminal | |
| CN113628251B (en) | Smart hotel terminal monitoring method | |
| Guan et al. | A visual saliency based railway intrusion detection method by UAV remote sensing image | |
| Khuc | Computer vision based structural identification framework for bridge health mornitoring | |
| Liu et al. | An improved subpixel-level registration method for image-based fault diagnosis of train bodies using SURF features | |
| Kahaki et al. | Incident and traffic-bottleneck detection algorithm in high-resolution remote sensing imagery | |
| WO2015178001A1 (en) | Image matching system, image matching method, and recording medium storing program | |
| Long et al. | Automatic Vehicle Speed Estimation Method for Unmanned Aerial Vehicle Images. |
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 | ||
| EE01 | Entry into force of recordation of patent licensing contract | ||
| EE01 | Entry into force of recordation of patent licensing contract |
Application publication date: 20170915 Assignee: GUANGZHOU HI-TARGET SURVEYING INSTRUMENT Co.,Ltd. Assignor: Beijing Jiaotong University Contract record no.: X2021990000807 Denomination of invention: Orbit intrusion detection method based on orbit visual feature spectrum Granted publication date: 20200505 License type: Exclusive License Record date: 20211222 |