CN109492647A - A kind of power grid robot barrier object recognition methods - Google Patents
A kind of power grid robot barrier object recognition methods Download PDFInfo
- Publication number
- CN109492647A CN109492647A CN201810636201.9A CN201810636201A CN109492647A CN 109492647 A CN109492647 A CN 109492647A CN 201810636201 A CN201810636201 A CN 201810636201A CN 109492647 A CN109492647 A CN 109492647A
- Authority
- CN
- China
- Prior art keywords
- image
- pixel
- edge
- line segment
- value
- 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.)
- Withdrawn
Links
- 230000004888 barrier function Effects 0.000 title claims abstract description 47
- 238000000034 method Methods 0.000 title claims abstract description 28
- 230000005540 biological transmission Effects 0.000 claims abstract description 22
- 230000011218 segmentation Effects 0.000 claims abstract description 15
- 230000033001 locomotion Effects 0.000 claims abstract description 14
- 238000003708 edge detection Methods 0.000 claims abstract description 10
- 238000001514 detection method Methods 0.000 claims description 24
- 238000003709 image segmentation Methods 0.000 claims description 13
- 239000012212 insulator Substances 0.000 claims description 7
- 230000009466 transformation Effects 0.000 claims description 7
- 230000009977 dual effect Effects 0.000 claims description 6
- 230000001186 cumulative effect Effects 0.000 claims description 4
- 230000004044 response Effects 0.000 claims description 4
- 230000015572 biosynthetic process Effects 0.000 claims description 3
- 238000012790 confirmation Methods 0.000 claims description 3
- 238000003786 synthesis reaction Methods 0.000 claims description 3
- 230000008569 process Effects 0.000 abstract description 10
- 230000009471 action Effects 0.000 abstract description 8
- 238000012544 monitoring process Methods 0.000 abstract description 5
- 238000010276 construction Methods 0.000 abstract description 3
- 238000012545 processing Methods 0.000 description 7
- 238000004458 analytical method Methods 0.000 description 4
- 230000001235 sensitizing effect Effects 0.000 description 4
- 230000000694 effects Effects 0.000 description 3
- 230000008030 elimination Effects 0.000 description 2
- 238000003379 elimination reaction Methods 0.000 description 2
- DEYFWGXTPWNADC-UHFFFAOYSA-N 6-[2-phenylethyl(propyl)amino]-5,6,7,8-tetrahydronaphthalen-1-ol Chemical compound C1CC2=C(O)C=CC=C2CC1N(CCC)CCC1=CC=CC=C1 DEYFWGXTPWNADC-UHFFFAOYSA-N 0.000 description 1
- 230000002159 abnormal effect Effects 0.000 description 1
- 230000009286 beneficial effect Effects 0.000 description 1
- 238000004364 calculation method Methods 0.000 description 1
- 230000007547 defect Effects 0.000 description 1
- 230000007812 deficiency Effects 0.000 description 1
- 230000005611 electricity Effects 0.000 description 1
- 230000007613 environmental effect Effects 0.000 description 1
- 238000010191 image analysis Methods 0.000 description 1
- 230000003993 interaction Effects 0.000 description 1
- 238000012372 quality testing Methods 0.000 description 1
- 238000011946 reduction process Methods 0.000 description 1
- 238000012216 screening Methods 0.000 description 1
- 230000035945 sensitivity Effects 0.000 description 1
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/40—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
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/22—Matching criteria, e.g. proximity measures
-
- G06T5/70—
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T7/00—Image analysis
- G06T7/10—Segmentation; Edge detection
- G06T7/12—Edge-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
- G06T7/13—Edge 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
- G06T7/136—Segmentation; Edge detection involving thresholding
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Computer Vision & Pattern Recognition (AREA)
- General Physics & Mathematics (AREA)
- Physics & Mathematics (AREA)
- Data Mining & Analysis (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Evolutionary Biology (AREA)
- Evolutionary Computation (AREA)
- Bioinformatics & Computational Biology (AREA)
- General Engineering & Computer Science (AREA)
- Artificial Intelligence (AREA)
- Life Sciences & Earth Sciences (AREA)
- Multimedia (AREA)
- Image Analysis (AREA)
- Image Processing (AREA)
Abstract
The invention belongs to power grid robot control field more particularly to a kind of power grid robot barrier object recognition methods.The data for analytical system that this method can obtain barrier in power grid Robot power transmission line motion process is analyzed and is judged, to guarantee that robot can make quick reflection to barrier, improve power grid machine task efficiency, region segmentation successively has been done to the image of acquisition, Gray Classification and edge detection, pass through region segmentation, it has been effectively compressed the image scale of construction, the data analyzed and handled are greatly reduced, reject most of unnecessary data in image, the unnecessary data refers to the data that equipment moving will not be hindered in image, but the partial data such as acts monitoring for other actions, online shooting etc. is necessary, pass through the segmentation procedure, number of devices required for robot system can be effectively compressed, reduce action.
Description
Technical field
The invention belongs to power grid robot control field more particularly to a kind of power grid robot barrier object recognition methods.
Background technique
It in online power grid robot control program, is automatically controlled to realize, then needs robot that there is obstacle recognition
Ability allows it to active across obstacle, due to the barrier in power grid containing different type or structure, barrier
Hinder judgement and the identification action of object more complicated, data processing amount is huge, reduces the work effect of power grid robot
Rate.
Summary of the invention
The purpose of the invention is that image data can be effectively compressed by providing one kind, improves image recognition efficiency, subtracts
Few data processing amount, while guaranteeing to distinguish quality, facilitate the obstacle recognition method of quickly online image analysis.
To achieve the above object, the invention adopts the following technical scheme that.
A kind of power grid robot barrier object recognition methods, includes the following steps:
1) intensity segmentation: the standard picture containing target object image to be detected that robot obtains online is chosen, by it
The image for only retaining gray scale is converted to, delimited tonal range [1,2,3...L], the number of pixels that statistics grey level is i is
ni, the sum of all pixels of the standard picture is N=n1+n2+...ni+...+nL, according to the gray feature of standard picture, it is assumed that required
The gray threshold t wanted, the gray threshold refer to a certain grey level, using on the grey level or under it is all
Pixel can satisfy the accuracy requirement of detection and disturbance in judgement object target when being element;It is to distinguish with gray threshold t, obtains ash
Spend section C0=[1, t] and C1=[t+1, L];The pixel mean value in the standard picture difference section is obtained simultaneouslyAnd pixel is located to C0And C1The probability variance in sectionWherein, w0It (t) is that pixel occurs
In C0The probability in section, w1It (t) is that pixel appears in C1The probability in section,Pi=ni/ N, w1(t)=1-w0
(t),
Define internal variance:
Inter-class variance:
σB 2=w0(t)*(μ0(t)-μT)2+w1(t)*(μ1(t)-μT)2=w0(t)*w1(t)*(μ0(t)-μ1(t))2;
Population variance:
2) control data: the value of adjustment gray threshold t are obtained, so that working as t=tγOr tηOr tκWhen [1, L] ∈, λ or η
Or κ is maximized, whereinCompare t=tγ、tη、tκWhen standard image segmentation result, choosing
It takes clearest accurate t value to carry out image segmentation to standard picture as standard grayscale threshold value and obtains segmented image;Simultaneously to this
Gray level image carries out pixels statistics under t value, calculates following data, including all straight line barriers obtain marginal texture and correspond to directly
The minimum pixel spacing L of line or line segment1;The minimum image of the marginal texture respective pixel of the barrier of all non-rectilinears or line segment
Plain spacing L2;
3) image segmentation: obtaining the video image in power grid Robot power transmission line linear motion, only retains intermediate D
The pixel of pixel wide range, whereinH is plant width, and S is detection of obstacles distance,
A is coverage when image obtains, and the plant width H, which refers to, is being parallel to photograph along all entity structures of a direction
The maximum width for the projection that the face last time in egative film direction is formed;
4) power transmission line detects: the discrete Gaussian function single order local derviation template of application carries out edge detection, retains wider edge
Response, screens image according to edge gradient direction, only retains edge gradient direction and meets T1≤ θ (x, y)≤T2Figure
As edge, edge image is extracted from segmented image;T1, T2For given threshold, numerically the inclining in the picture with power transmission line
Oblique angle is identical, and when power grid robot straight line is walked, power transmission line is located at the middle position of image, and it is to prolong vertically downward that Cong Tuzhong, which is seen,
It stretches, inclination angle meets 70 < θ < 110, therefore T1=70;T2=110;
5) pixel in edge image is randomly selected to accumulator, detects the line segment in edge image, particular content
Include:
A above-mentioned edge image after treatment) is converted into image IMG, if edge image is sky, terminates algorithm;
B) if non-empty, accumulator is updated from selection pixel in IMG at random and removes pixel from IMG;
C it) checks whether the maximum summing elements BIN in accumulator is greater than preset threshold, if not having, goes to A;
D the line segment in straight line) is searched for according to the rectilinear direction that BIN is specified in IMG;
E the corresponding pixel of line segment) is removed in IMG, removes accumulator;
F) if line segment length is less than default minimum line segment length L1, then line segment is exported;
By above-mentioned cumulative transformation, straight line and line segment endpoint are extracted from image, determines the clear area of power transmission line,
The region of line segment is not included;
6) above-mentioned steps are based on, can divide and extract the edge image of barrier, the edge image from standard picture
In line segment include that the barriers such as straight-line segment corresponding to power transmission line, insulator are corresponding by a plurality of non-parallel straightway group
At closed figure normally behave as round or ellipse in the image that power grid robot obtains, to realize barrier
Detection is needed according to the line segment in edge image after determining edge image come other obstacles such as insulators in detection image
Object;Detect the round or oval pattern of pixels in edge image;After obtaining line segment endpoint, it is believed that should
Line segment corresponds to a string of circle or pattern of oval shapes in image, former based on CMHT (point transformation ellipses detection in string)
Reason, using line segment and terminal point information the confirmation oval logo obtained in step 5, (circle is considered as a major and minor axis phase
Deng ellipse), specific steps include:
I) with the coordinate (x of two endpoints of each line segment in the edge image that is obtained in step 51, y1)、 (x2,
y2) it is input parameter, calculate oval reference parameter x0, y0, a, wherein
x0=(x1+x2)/2;y0=(y1+y2)/2;
II pixel corresponding to edge in edge image) is traversed, by any one pixel (x, y) conduct on edge
A point on ellipse is tested, it is assumed that elliptical focal length is f, and semi-minor axis f then has:
Cos τ=(a2+d2-f2)/(2*a*d);
b2=(a2*d2* sin2τ)/(a2-d2*cos2τ);
The corresponding b value of each pixel is obtained after all edge pixel points are calculated;
III) the successively corresponding b value of more each pixel and L2Size, if b < L2The point is then rejected, if b > L2Then with
The record pixel;After the coordinate of the pixels for meeting condition all in image determines, if any four pixel is corresponding
B value is identical, then determines elliptic coordinates with the endpoint of four pixels and aforementioned line segment.
IV all elliptic coordinates) are successively detected and recorded, coordinate parameters are returned.
The discrete Gaussian function single order local derviation template of application carries out edge detection in the step 4), retains wider edge and rings
The particular content is answered to include:
1. being smoothed using Gaussian filter to image, noise is removed, the Gaussian filter can be expressed asF (x, y)=G (x, y) * f (x, y);
2. the first derivative and image using Gaussian function make convolution, gradient magnitude E is calculatedxWith gradient direction Ey, obtain
Edge strengthIt is respectively as follows: with direction φ (x, y)
Wherein
3. to gradient magnitude ExIt is consistent (NMS) to carry out non-maximum, retains the maximum point of partial gradient.Have in this programme
Body, which refers to, is contracted to gradient angle θ=[i, j] variation range in circular sector, and round mountain area is divided into the area of m*m
Lattice are calculated in each pixel along the gradient value of gradient line two adjacent pixels and centre of neighbourhood pixel M, if the gradient value of M is less than
Equal to the gradient value of two adjacent pixels, then M=0 is enabled.Edge pixel point is detected and connected with dual threashold value-based algorithm.It chooses and closes
Suitable threshold value to after segmentation standard picture carry out thresholding it is relatively difficult, but use dual threshold, i.e., selection two threshold value π1、
π2, obtain two threshold skirt images, by joint two therewith image carry out convergence synthesis, final edge can be obtained
Image, wherein 2 π1=π2。
The beneficial effect is that: be based on above-mentioned steps, this method can in power grid Robot power transmission line motion process,
The data for analytical system for obtaining barrier fast and accurately is analyzed and is judged, to guarantee that robot can be to obstacle
Object makes quick reflection, improves power grid machine task efficiency.In the present solution, successively having done region point to the image of acquisition
Cut, Gray Classification and edge detection by region segmentation have been effectively compressed the image scale of construction, be greatly reduced point
The data of analysis and processing reject most of unnecessary data in image, which, which refers to, will not hinder equipment in image
The data of movement, but the partial data such as acts monitoring for other actions, online shooting is necessary, by this
Barrier recognition scheme can be combined with conventional monitoring shooting work, can either guarantee routine work by segmentation procedure
It carries out, and obstacle recognition can be carried out in line image using what is obtained in routine work, be effectively compressed needed for robot system
The number of devices wanted reduces action;Segmented image is obtained by previous segmentation program, gray scale point is carried out to segmented image
After grade, image volume is further had compressed, the gray level image obtained convenient for analysis and detection, elimination pair are classified by indexing
Obstacle detection does not have influential CONSTRUCTED SPECIFICATION, protrudes the edge details of barrier, in order in rear prologue to obstacles borders
The detection and positioning of pixel utilize edge line segment and its endpoint pixel coordinate, quick obtaining in edge detection process
Different type barrier is divided in order to rear by the control to pixel spacing in relation to the position data of barrier in image
The identification of barrier is screened in continuous work, proposes intelligent effect.
Specific embodiment
It elaborates below in conjunction with specific embodiment to the invention.
Power grid robot can make corresponding movement according to known working environment and execute corresponding task, including cross over
Barrier, along cable movement, removing obstacles object etc..But in the prior art, due to the deficiency of robot functional structure itself,
Information Collection System environmental information obtained is not perfect, task and movement corresponding operation algorithm existing defects all can
Prevent robot from completing relevant work, realizes entirely autonomous behaviour control, therefore power grid robot of the invention includes master
It is dynamic to control and be remotely controlled scheme, the advantages of constraining with human-computer interaction progress is controlled in conjunction with locally autonomous and is mutually integrated, and is being operated
Journey is deposited starts manual intervention when abnormal, guarantees the normal operation of power grid robot, improves event handling efficiency.
A kind of power grid robot barrier object recognition methods of the invention, specific steps include raw based on local feature recognition
At target feature point, main contents include: 1) to detect local feature region;2) target signature is extracted;The contents of the section is used for
The image characteristic point of barrier that may be present is obtained before movement, provides matching mould to carry out obstacle recognition for robot
Plate, for the accuracy for improving the online image recognition of power grid robot, the matching stencil for carrying out local feature region detection should
It is being extracted based on line image of being directly acquired in the robot operative scenario, specific work process includes,
It is normal data that the power network line containing most of or there may be barrier is selected in operation interval, uses power grid
Robot or other picture pick-up devices obtain normal video or image on the power network line;It is wrapped in normal video or image
Containing a large amount of inessential data, including all kinds of background elements and other elements, and in the online walking process of power grid robot
The a few types such as device structure along the barrier encountered usually only power grid, it is therefore desirable to the source data of normal data
It carries out simplifying processing, simplifies data handling procedure, reduce data volume when robot On-line matching, concrete principle are as follows:
1. the image in screening criteria data chooses several standard pictures comprising important goal object;
2. choosing a secondary standard image, according to the gray value of the standard picture, delimit tonal range [1,2,3...L], system
Counting the number of pixels that grey level is l is nl, the sum of all pixels of the standard picture is N=n1+n2+...nl+...+nL, according to mark
The gray feature of quasi- image, it is assumed that required gray threshold t, the gray threshold refer to a certain grey level, utilize the ash
On degree rank or under all pixels can satisfy the demand of detection and disturbance in judgement object target when being element;With ash
Spending threshold value t is to distinguish, and obtains gray scale interval C0=[1, t] and C1=[t+1, L];The standard picture difference section is obtained simultaneously
Pixel mean valueAnd pixel is located to C0And C1The probability in section
VarianceWherein, w0It (t) is pixel
Point appears in C0The probability in section, w1It (t) is that pixel appears in C1The probability in section,Pi=ni/ N, easily
Know, w1(t)=1-w0(t),
Define internal varianceInter-class variance
σB 2=w0(t)*(μ0(t)-μT)2+w1(t)*(μ1(t)-μT)2=w0(t)*w1(t)*(μ0(t)-μ1(t))2;
Population varianceAbove-mentioned variance meets σT 2=σW 2+σB 2;
3. the value of gray threshold t is adjusted, so that working as t=tγOr tηOr tκWhen [1, L] ∈, λ or η or κ are maximized,
InCompare t=tγ、tη、tκWhen standard image segmentation result, choose it is clearest accurately
T value carries out image segmentation to standard picture as standard grayscale threshold value.
Above by tonal gradation and gray threshold as standard criteria for classifying image, reduce in the processing of image
Hold, at the same existing colored image can be convenient be converted to high quality gray level image, can be effectively treated existing obtained
Image deletes unwanted pixel, can be obtained required detection after carrying out image segmentation by above-mentioned gray threshold
The pixel set of element reduces the pixel quantity in standard picture, reduces data volume, improves data-handling efficiency.Before being based on
Stating method, we can obtain the standard picture containing substantially objective contour, since image simplification process and image itself are former
Because will lead to edge blurry, be not easy to recognize, it is therefore desirable to denoising carried out to image, in the present invention, noise reduction process principle
Are as follows: it should be pointed out that image is aforementioned using same picture format and identical camera shooting scheme acquisition data in the present invention
The interval endpoint and control coefrficient of use delimited according to the precision that the quality of standard picture and needs distinguish, above-mentioned area
Between endpoint and control coefrficient with using different-format image or camera shooting scheme might not be identical, concrete principle is such as
Under:
4. being smoothed using Gaussian filter to image, noise is removed, the Gaussian filter can be expressed asF (x, y)=G (x, y) * f (x, y);
5. the first derivative and image using Gaussian function make convolution, gradient magnitude E is calculatedxWith gradient direction Ey, obtain
Edge strengthIt is respectively as follows: with direction φ (x, y)
Wherein
6. to gradient magnitude ExIt is consistent (NMS) to carry out non-maximum, retains the maximum point of partial gradient.Have in this programme
Body, which refers to, is contracted to gradient angle θ=[i, j] variation range in circular sector, and round mountain area is divided into the area of m*m
Lattice are calculated in each pixel along the gradient value of gradient line two adjacent pixels and centre of neighbourhood pixel M, if the gradient value of M is less than
Equal to the gradient value of two adjacent pixels, then M=0 is enabled.
7. edge pixel point is detected and connected with dual threashold value-based algorithm.Suitable threshold value is chosen to the standard picture after segmentation
Carry out thresholding it is relatively difficult, but use dual threshold, i.e., selection two threshold value π1、π2, two threshold skirt images are obtained, are led to
Crossing joint two, image carries out convergence synthesis therewith, final edge image can be obtained, wherein 2 π1=π2.In the step, lead to
It crosses using the image of high threshold and reduces false edge, using the edge of Low threshold image as the intermittent of high threshold image
Connection end point, and then closed high threshold image is obtained, the validity of image border is improved, picture quality is improved.
After the division of above-mentioned threshold value and image procossing, the edge element of specific image in standard picture can be obtained,
The edge element is that the gray value based on standard picture is chosen, thus while the edge element in the image obtained, but still
Different object elements directly cannot be differentiated from the edge factor data, such as include all kinds of stockbridge dampers, pendency in power transmission line
The barriers such as wire clamp, strain clamp and insulator need to realize that different obstacle detourings acts for different types of barrier
Accurately to distinguish above-mentioned different obstacle identities.It realizes simultaneously and accurately identifies, positions and track.
Since above-mentioned barrier is usually with the object of specific shape, including straight line, rotated image, circle in standard picture
Arc etc., therefore can use the Hough transform of image to be identified and positioned, the shift theory in the present invention is as follows:
1. constructing cumulative array H and element being initialized as 0;
2. each of for image pixel (x, y)
3. finding the corresponding parameter (θ, ρ) of accumulator H (θ, ρ) local maximum;
Traditional Hough transform algorithm is improved by the way that step is 2. middle, above-mentioned transformation can reduce multiple edge point standard picture
Data processing amount improves efficiency of algorithm, reduces marginal point and repeats statistic processes.Based on being needed in actually detected work to each disconnected
Electricity is into accurate positionin, it is therefore desirable to converted with PPHT, therefore, in the present invention to barrier marginal edge in standard picture
Position fixing process specifically includes:
7) intensity segmentation: the standard picture for containing target object image to be detected is chosen, is converted into and only retains gray scale
Image, delimit tonal range [1,2,3...L], statistics grey level be i number of pixels be ni, the picture of the standard picture
Plain sum is N=n1+n2+...ni+...+nL, according to the gray feature of standard picture, it is assumed that and required gray threshold t,
The gray threshold refers to a certain grey level, using on the grey level or under all pixels be element when can
Meets the needs of detection and disturbance in judgement object target;It is to distinguish with gray threshold t, obtains gray scale interval C0=[1, t] and C1=
[t+1, L];The pixel mean value in the standard picture difference section is obtained simultaneously And pixel is located to C0And C1The probability variance in sectionWherein, w0 (t) is that pixel is pointed out
Present C0The probability in section, w1It (t) is that pixel appears in C1The probability in section,It is apparent from,
w1(t)=1-w0(t),
Define internal varianceInter-class variance
σB 2=w0(t)*(μ0(t)-μT)2+w1(t)*(μ1(t)-μT)2=w0(t)*w1(t)*(μ0(t)-μ1(t))2;
Population varianceAbove-mentioned variance meets σT 2=σW 2+σB 2;
8) control data: the value of adjustment gray threshold t are obtained, so that working as t=tγOr tηOr tκWhen [1, L] ∈, λ or η
Or κ is maximized, whereinCompare t=tγ、tη、tκWhen standard image segmentation result, choosing
It takes clearest accurate t value to carry out image segmentation to standard picture as standard grayscale threshold value and obtains segmented image;
9) image segmentation: when determining power grid robot normal walking, machine is influenced in the video image that robot front obtains
The sensitizing range of device people future-action and instruction, such as to common domestic power grid, can determine its direct picture center 40
A pixel wide is sensitizing range;It is easy to know, when the remoter the distance or width analyzed the wider, need to improve sensitivity
The pixel wide and picture quality in region obtains deeper pictorial element;In this step, by delimiting sensitizing range, contracting
The volume of small standard picture only considers that the barrier of power grid robot motion can be hindered, to reduce data processing amount, at raising
Manage efficiency.
In power grid Robot power transmission line linear motion, it is assumed that the plant width of its a direction is H, obstacle quality testing
Ranging is from for S, and coverage when image obtains is α, and the corresponding angular range of H width is β at S distance, then in α
Barrier A, B, C in angular range, the B and C being only located in β angular range influence robot motion in this direction
Practical obstacle object, the A between angular range beta and α will not impact the movement of robot.The β angular range is being schemed
As upper corresponding pixel region is sensitizing range, during actual analysis, approximate calculation, i.e. tan (β/2) can be carried out
=H/ (2*S),When performing image segmentation, only retain the pixel of intermediate β angular range, pixel
WidthD0For the total pixel wide of direction video camera, under normal circumstances, D is
40。
10) power transmission line detects: the discrete Gaussian function single order local derviation template of application carries out edge detection, retains wider edge
Response, screens image according to edge gradient direction, only retains edge gradient direction and meets T1≤ θ (x, y)≤T2Figure
As edge, edge image is extracted from segmented image;T1, T2For given threshold, numerically the inclining in the picture with power transmission line
Oblique angle is identical, and when power grid robot straight line is walked, power transmission line is located at the middle position of image, and it is to prolong vertically downward that Cong Tuzhong, which is seen,
It stretches, inclination angle meets 70 < θ < 110, therefore T1=70;T2=110;
11) pixel in edge image is randomly selected to accumulator, detects the line segment in edge image, particular content
Include:
G above-mentioned edge image after treatment) is converted into image IMG, if edge image is sky, terminates algorithm;
H) if non-empty, accumulator is updated from selection pixel in IMG at random and removes pixel from IMG;
I it) checks whether the maximum summing elements BIN in accumulator is greater than preset threshold, if not having, goes to A;
J the line segment in straight line) is searched for according to the rectilinear direction that BIN is specified in IMG;
K the corresponding pixel of line segment) is removed in IMG, removes accumulator;
L) if line segment length is less than default minimum line segment length L1, then line segment is exported;The minimum line segment length L1, pass through
The minimum pixel spacing that target obstacle can be embodied in detection conventional images determines;
By above-mentioned cumulative transformation, straight line and line segment endpoint are extracted from image, determines the clear area of power transmission line,
The region of line segment is not included;
12) above-mentioned steps are based on, can divide and extract the edge image of barrier, the edge image from standard picture
In line segment include that the barriers such as straight-line segment corresponding to power transmission line, insulator are corresponding by a plurality of non-parallel straightway group
At closed figure normally behave as round or ellipse in the image that power grid robot obtains, to realize barrier
Detection is needed according to the line segment in edge image after determining edge image come other obstacles such as insulators in detection image
Object;Detect the round or oval pattern of pixels in edge image;Based on the content in step 5, the extremity of segment is being obtained
After point, it is believed that the line segment corresponds to a string of circle or pattern of oval shapes in image, based on CMHT (in string
Point transformation ellipses detection) principle, using line segment and terminal point information the confirmation oval logo obtained in step 5, (circle can be seen
The ellipse equal as a major and minor axis), specific steps include:
V) with the coordinate (x of two endpoints of each line segment in the edge image that is obtained in step 51, y1)、 (x2,
y2) it is input parameter, calculate oval reference parameter x0, y0, a, wherein
x0=(x1+x2)/2;y0=(y1+y2)/2;
VI pixel corresponding to edge in edge image) is traversed, by any one pixel (x, y) conduct on edge
A point on ellipse is tested, it is assumed that elliptical focal length is f, and semi-minor axis f then has:
Cos τ=(a2+d2-f2)/(2*a*d);
b2=(a2*d2* sin2τ)/(a2-d2*cos2τ);
The corresponding b value of each pixel is obtained after all edge pixel points are calculated;
VII) in standard edge image, the minimum pixel spacing L of obstacle recognition is extracted2, successively more each pixel
Corresponding b value and L2Size, if b < L2The point is then rejected, if b > L2Then with the record pixel;To all full in image
After the coordinate of the pixel of sufficient condition determines, if the corresponding b value of any four pixel is identical, with four pixels with
And the endpoint of aforementioned line segment determines elliptic coordinates.
VIII all elliptic coordinates) are successively detected and recorded, coordinate parameters are returned;
Based on above-mentioned steps, this method can obtain fast and accurately in power grid Robot power transmission line motion process
The data for analytical system of barrier is taken to be analyzed and judged, to guarantee that robot can make quick reflection to barrier,
Improve power grid machine task efficiency.In the present solution, successively to the image of acquisition done region segmentation, Gray Classification and
Edge detection has been effectively compressed the image scale of construction by region segmentation, and the data analyzed and handled are greatly reduced,
Most of unnecessary data in image is rejected, which refers to the data that equipment moving will not be hindered in image, but should
Partial data such as acts monitoring for other actions, online shooting is necessary, and by the segmentation procedure, can incite somebody to action
Barrier recognition scheme is combined with conventional monitoring shooting work, can either guarantee the progress of routine work, and can be using often
What is obtained in rule work carries out obstacle recognition in line image, has been effectively compressed number of devices required for robot system, has subtracted
Few action;Segmented image is obtained by previous segmentation program, after carrying out Gray Classification to segmented image, is further pressed
Contracted image volume, obtains the gray level image convenient for analysis and detection by indexing classification, elimination does not influence obstacle detection
CONSTRUCTED SPECIFICATION, protrude the edge details of barrier, in order in rear prologue to the detection of obstacles borders pixel and fixed
Position, in edge detection process, using edge line segment and its endpoint pixel coordinate, related barrier in quick obtaining image
Position data, by the control to pixel spacing, divide different type barrier in order in follow-up work to barrier
Identification screen, propose intelligent effect.
Finally it should be noted that above embodiments are only to illustrate the technical solution of the invention, rather than to this hair
It is bright create protection scope limitation, although being explained in detail referring to preferred embodiment to the invention, this field it is general
Lead to it will be appreciated by the skilled person that can be modified or replaced equivalently to the technical solution of the invention, without departing from this
The spirit and scope of innovation and creation technical solution.
Claims (2)
1. a kind of power grid robot barrier object recognition methods, which comprises the steps of:
1) intensity segmentation: the standard picture containing target object image to be detected that robot obtains online is chosen, is converted
It for the image for only retaining gray scale, delimit tonal range [1,2,3...L], the number of pixels that statistics grey level is i is ni, the mark
The sum of all pixels of quasi- image is N=n1+n2+...ni+...+nL, according to the gray feature of standard picture, it is assumed that required ash
Spend threshold value t, the gray threshold refers to a certain grey level, using on the grey level or under all pixels be to want
It can satisfy the accuracy requirement of detection and disturbance in judgement object target when plain;It is to distinguish with gray threshold t, obtains gray scale interval C0=
[1, t] and C1=[t+1, L];The pixel mean value in the standard picture difference section is obtained simultaneouslyAnd pixel is located to C0And C1The probability variance in sectionWherein, w0It (t) is that pixel occurs
In C0The probability in section, w1It (t) is that pixel appears in C1The probability in section,Pi=ni/ N, w1(t)=1-w0
(t),
Define internal variance:
Inter-class variance:
σB 2=w0(t)*(μ0(t)-μT)2+w1(t)*(μ1(t)-μT)2=w0(t)*w1(t)*(μ0(t)-μ1(t))2;
Population variance:
2) control data: the value of adjustment gray threshold t are obtained, so that working as t=tγOr tηOr tκWhen [1, L] ∈, λ or η or k are taken
Maximum value, whereinCompare t=tγ、tη、tκWhen standard image segmentation result, choose it is most clear
Clear accurate t value carries out image segmentation to standard picture as standard grayscale threshold value and obtains segmented image;Simultaneously to grey under the t value
It spends image and carries out pixels statistics, calculate following data, including all straight line barriers obtain marginal texture and correspond to straight line or line segment
Minimum pixel spacing L1;The minimum pixel spacing L of the marginal texture respective pixel of the barrier of all non-rectilinears or line segment2;
3) image segmentation: obtaining the video image in power grid Robot power transmission line linear motion, only retains intermediate D pixel
The pixel of width range, whereinH is plant width, and S is detection of obstacles distance, and a is figure
Coverage when as obtaining, the plant width H, which refers to, is being parallel to photographic negative side along all entity structures of a direction
To face last time formed projection maximum width;
4) power transmission line detects: the discrete Gaussian function single order local derviation template of application carries out edge detection, retains wider skirt response,
Image is screened according to edge gradient direction, only retains edge gradient direction and meets T1≤ θ (x, y)≤T2Image border,
Edge image is extracted from segmented image;T1, T2It is numerically identical as the inclination angle of power transmission line in the picture for given threshold,
When power grid robot straight line is walked, power transmission line is located at the middle position of image, and it is to extend vertically downward that Cong Tuzhong, which is seen, inclination angle
Meet 70 < θ < 110, therefore T1=70;T2=110;
5) pixel in edge image is randomly selected to accumulator, detects the line segment in edge image, particular content includes:
A above-mentioned edge image after treatment) is converted into image IMG, if edge image is sky, terminates algorithm;
B) if non-empty, accumulator is updated from selection pixel in IMG at random and removes pixel from IMG;
C it) checks whether the maximum summing elements BIN in accumulator is greater than preset threshold, if not having, goes to A;
D the line segment in straight line) is searched for according to the rectilinear direction that BIN is specified in IMG;
E the corresponding pixel of line segment) is removed in IMG, removes accumulator;
F) if line segment length is less than default minimum line segment length L1, then line segment is exported;
By above-mentioned cumulative transformation, straight line and line segment endpoint are extracted from image, determines the clear area of power transmission line, i.e., not
Region comprising line segment;
6) above-mentioned steps are based on, can divide and extract the edge image of barrier from standard picture, in the edge image
Line segment includes the corresponding envelope being made of a plurality of non-parallel straightway of the barriers such as straight-line segment corresponding to power transmission line, insulator
Figure is closed, in the image that power grid robot obtains, normally behaves as round or ellipse, to realize detection of obstacles,
After determining edge image, need according to the line segment in edge image come other barriers such as insulators in detection image;Examine
Survey the round or oval pattern of pixels in edge image;After obtaining line segment endpoint, it is believed that the line segment is corresponding
One string of circle or pattern of oval shapes in image is based on CMHT (point transformation ellipses detection in string) principle, utilizes step
Line segment and terminal point information the confirmation oval logo (circle is considered as the ellipse equal for a major and minor axis) obtained in rapid 5,
Its specific steps includes:
I) with the coordinate (x of two endpoints of each line segment in the edge image that is obtained in step 51, y1)、(x2, y2) it is defeated
Enter parameter, calculates oval reference parameter x0, y0, a, wherein
x0=(x1+x2)/2;y0=(y1+y2)/2;
II pixel corresponding to edge in edge image) is traversed, by any one pixel (x, y) on edge as on ellipse
A point test, it is assumed that elliptical focal length is f, and semi-minor axis f then has:
Cos τ=(a2+d2-f2)/(2*a*d);
b2=(a2*d2*sin2τ)/(a2-d2*cos2τ);
The corresponding b value of each pixel is obtained after all edge pixel points are calculated;
III) the successively corresponding b value of more each pixel and L2Size, if b < L2The point is then rejected, if b > L2Then with the note
Video recording vegetarian refreshments;After the coordinate of the pixels for meeting condition all in image determines, if the corresponding b value phase of any four pixel
Together, then elliptic coordinates are determined with the endpoint of four pixels and aforementioned line segment;
IV all elliptic coordinates) are successively detected and recorded, coordinate parameters are returned.
2. a kind of power grid robot barrier object recognition methods according to claim 1, which is characterized in that answered in the step 4)
Edge detection is carried out with discrete Gaussian function single order local derviation template, retaining wider skirt response particular content includes,
1. being smoothed using Gaussian filter to image, noise is removed, the Gaussian filter can be expressed asF (x, y)=G (x, y) * f (x, y);
2. the first derivative and image using Gaussian function make convolution, gradient magnitude E is calculatedxWith gradient direction Ey, it is strong to obtain edge
DegreeIt is respectively as follows: with direction φ (x, y)Wherein
3. to gradient magnitude ExIt is consistent (NMS) to carry out non-maximum, retains the maximum point of partial gradient;It is specifically referred in this programme
Gradient angle θ=[i, j] variation range is contracted in circular sector, and round mountain area is divided into area's lattice of m*m, is calculated
Each pixel along gradient line two adjacent pixels and centre of neighbourhood pixel M gradient value, if the gradient value of M be less than or equal to two phases
The gradient value of adjacent pixel, then enable M=0;Edge pixel point is detected and connected with dual threashold value-based algorithm;Choose suitable threshold value to point
It is relatively difficult that standard picture after cutting carries out thresholding, but uses dual threshold, i.e. two threshold value π of selection1、π2, obtain two threshold values
Edge image, by joint two therewith image carry out convergence synthesis, final edge image can be obtained, wherein 2 π1=π2。
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201810636201.9A CN109492647A (en) | 2018-06-20 | 2018-06-20 | A kind of power grid robot barrier object recognition methods |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201810636201.9A CN109492647A (en) | 2018-06-20 | 2018-06-20 | A kind of power grid robot barrier object recognition methods |
Publications (1)
Publication Number | Publication Date |
---|---|
CN109492647A true CN109492647A (en) | 2019-03-19 |
Family
ID=65689172
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201810636201.9A Withdrawn CN109492647A (en) | 2018-06-20 | 2018-06-20 | A kind of power grid robot barrier object recognition methods |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN109492647A (en) |
Cited By (5)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN110705545A (en) * | 2019-09-09 | 2020-01-17 | 云南电网有限责任公司曲靖供电局 | Obstacle identification method and device, storage medium and inspection robot |
CN113043399A (en) * | 2020-12-31 | 2021-06-29 | 湘潭大学 | Bamboo joint removing intelligent device and method for interrupted bamboo tableware production |
CN113238560A (en) * | 2021-05-24 | 2021-08-10 | 珠海市一微半导体有限公司 | Robot map rotating method based on line segment information |
CN116540776A (en) * | 2023-06-05 | 2023-08-04 | 深圳市华赛睿飞智能科技有限公司 | Unmanned aerial vehicle vision obstacle avoidance method and system |
CN116651847A (en) * | 2023-06-07 | 2023-08-29 | 江苏省人民医院(南京医科大学第一附属医院) | Ultrasonic cleaning effect detection system for medical instrument |
-
2018
- 2018-06-20 CN CN201810636201.9A patent/CN109492647A/en not_active Withdrawn
Cited By (7)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN110705545A (en) * | 2019-09-09 | 2020-01-17 | 云南电网有限责任公司曲靖供电局 | Obstacle identification method and device, storage medium and inspection robot |
CN110705545B (en) * | 2019-09-09 | 2023-03-10 | 云南电网有限责任公司曲靖供电局 | Obstacle identification method and device, storage medium and inspection robot |
CN113043399A (en) * | 2020-12-31 | 2021-06-29 | 湘潭大学 | Bamboo joint removing intelligent device and method for interrupted bamboo tableware production |
CN113238560A (en) * | 2021-05-24 | 2021-08-10 | 珠海市一微半导体有限公司 | Robot map rotating method based on line segment information |
CN116540776A (en) * | 2023-06-05 | 2023-08-04 | 深圳市华赛睿飞智能科技有限公司 | Unmanned aerial vehicle vision obstacle avoidance method and system |
CN116540776B (en) * | 2023-06-05 | 2023-11-07 | 深圳市华赛睿飞智能科技有限公司 | Unmanned aerial vehicle vision obstacle avoidance method and system |
CN116651847A (en) * | 2023-06-07 | 2023-08-29 | 江苏省人民医院(南京医科大学第一附属医院) | Ultrasonic cleaning effect detection system for medical instrument |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN109492647A (en) | A kind of power grid robot barrier object recognition methods | |
CN109033950B (en) | Vehicle illegal parking detection method based on multi-feature fusion cascade depth model | |
CN109145803B (en) | Gesture recognition method and device, electronic equipment and computer readable storage medium | |
CN105930822A (en) | Human face snapshot method and system | |
CN103824070B (en) | A kind of rapid pedestrian detection method based on computer vision | |
CN107909081B (en) | Method for quickly acquiring and quickly calibrating image data set in deep learning | |
CN110443128B (en) | Finger vein identification method based on SURF feature point accurate matching | |
CN103699905B (en) | Method and device for positioning license plate | |
CN110309781B (en) | House damage remote sensing identification method based on multi-scale spectrum texture self-adaptive fusion | |
US20130243343A1 (en) | Method and device for people group detection | |
CN109871763A (en) | A kind of specific objective tracking based on YOLO | |
CN114241548A (en) | Small target detection algorithm based on improved YOLOv5 | |
CN107657209B (en) | Template image registration mechanism based on finger vein image quality | |
CN110415208A (en) | A kind of adaptive targets detection method and its device, equipment, storage medium | |
CN109685045A (en) | A kind of Moving Targets Based on Video Streams tracking and system | |
CN111507426A (en) | No-reference image quality grading evaluation method and device based on visual fusion characteristics | |
CN109190456B (en) | Multi-feature fusion overlook pedestrian detection method based on aggregated channel features and gray level co-occurrence matrix | |
CN109886170B (en) | Intelligent detection, identification and statistics system for oncomelania | |
CN105894037B (en) | A kind of full supervised classification method of remote sensing images extracted based on SIFT training sample | |
CN109255336A (en) | Arrester recognition methods based on crusing robot | |
CN112287802A (en) | Face image detection method, system, storage medium and equipment | |
CN113449606A (en) | Target object identification method and device, computer equipment and storage medium | |
CN115841633A (en) | Power tower and power line associated correction power tower and power line detection method | |
CN110458019B (en) | Water surface target detection method for eliminating reflection interference under scarce cognitive sample condition | |
CN111683221A (en) | Real-time video monitoring method and system for natural resources embedded with vector red line data |
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 | ||
WW01 | Invention patent application withdrawn after publication |
Application publication date: 20190319 |
|
WW01 | Invention patent application withdrawn after publication |