CN105957077A - Detection method for foreign body in transmission lines based on visual saliency analysis - Google Patents
Detection method for foreign body in transmission lines based on visual saliency analysis Download PDFInfo
- Publication number
- CN105957077A CN105957077A CN201610269864.2A CN201610269864A CN105957077A CN 105957077 A CN105957077 A CN 105957077A CN 201610269864 A CN201610269864 A CN 201610269864A CN 105957077 A CN105957077 A CN 105957077A
- Authority
- CN
- China
- Prior art keywords
- area
- interest
- power transmission
- original image
- region
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Granted
Links
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T7/00—Image analysis
- G06T7/0002—Inspection of images, e.g. flaw detection
- G06T7/0004—Industrial image inspection
- G06T7/0008—Industrial image inspection checking presence/absence
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T7/00—Image analysis
- G06T7/60—Analysis of geometric attributes
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/30—Subject of image; Context of image processing
- G06T2207/30108—Industrial image inspection
Landscapes
- Engineering & Computer Science (AREA)
- Physics & Mathematics (AREA)
- Computer Vision & Pattern Recognition (AREA)
- General Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- Geometry (AREA)
- Quality & Reliability (AREA)
- Image Analysis (AREA)
Abstract
The invention discloses a detection method for foreign body in transmission lines based on visual saliency analysis. The method is performed through the following steps: A) dividing a plurality of transmission lines in an original image corresponding to the current frame into different parallel straight line groups; B) obtaining an initial region of interest; C) transforming the region of interest into a rectangular region of interest through the utilization of perspective transformation; D) judging whether there exists an area with abnormity; E) processing the original image after N frames; F) restoring the area with abnormity to the fault position in the original image corresponding to the current frame; and G) processing the next frame. The invention realizes the uniform consideration of each feature by judging and locating the positions of foreign bodies by the visual saliency features, and statistically increases the robustness of a detection result. In addition, the size of a region of interest can be varied for a required situation to achieve a faster detection speed, avoiding a large number of repetitive calculations and reducing the uncertainty of the results due to the diversity of the types of foreign body failures.
Description
Technical field
The present invention relates to a kind of electric line foreign matter detection method, particularly relate to the defeated of a kind of view-based access control model significance analysis
Electric line foreign matter detecting method.
Background technology
In recent years, various places are because discharging kite, balloon etc., and the event jeopardizing power grid security happens occasionally.In power system
In, the foreign body that transmission line of electricity hangs not only affects the normal power supply of circuit, and the limit arcing distance of high-tension electricity also can be made to shorten,
Jeopardize the pedestrian under electric lines of force and vehicle safety, line tripping can be caused time serious, cause section large-area power-cuts.Therefore, logical
Cross and analyze the picture that helicopter routing inspection process photographs, in time the suspension foreign body on transmission line of electricity is identified automatically, from
And take corresponding solution, working strength can be significantly reduced and improve work efficiency.
Existing electric line foreign matter fault detection method based on picture analyzing of taking photo by plane is broadly divided into two kinds: one class emphatically
In the gradient direction distribution characteristic of analysis straightway, mainly judge to deposit by the cross linear section in detection transmission line of electricity region
At foreign body.When occurring in view of the foreign body fault on transmission line of electricity, the mode of appearance of foreign body is different, only by cross linear section
Detection can cause the generation of missing inspection event.Another kind of weight analysis foreign body is special relative to the Color-spatial distribution of whole image
Property, time mainly by linearly detecting, the sudden change of color space judges whether foreign body.This type of method is in foreign body face
Having certain effect when color more highlights obvious, when foreign body is the most very thin with transmission line of electricity infall, or foreign body color is failed to understand
Time aobvious, have missing inspection event and occur.
Summary of the invention
It is an object of the invention to provide the electric line foreign matter detection method of a kind of view-based access control model significance analysis, by regarding
Feel that significant characteristics judges and location foreign body exists the characteristics such as position, Color, shape and spatial distribution, it is achieved to each feature
Unification consider, and consider the testing result being positioned at a time point, from statistics angle, add the robust of testing result
Property, significantly reduce the loss of foreign body fault.Meanwhile, the present invention can according to situation change area-of-interest size with
Detected speed faster, it is to avoid the calculating of a large amount of repetitions and reduce the knot caused because of the multiformity of foreign body failure mode
Fruit is uncertain.
The present invention uses following technical proposals:
The electric line foreign matter detection method of a kind of view-based access control model significance analysis, comprises the following steps:
Step A: using the i-th frame as present frame, i=1,2,3 ...;Utilize the power transmission line spatial parameter detected, the most defeated
The slope of electric wire and the distance with other power transmission line, be divided into a plurality of power transmission line in the original image corresponding to present frame not
Same parallel lines group;Subsequently into step B;
Step B: utilize the group result of the parallel lines group obtained in step A, obtain each parallel lines group institute respectively
The region that covers also is set to tentative area-of-interest;Then whether exist before judging present frame and sentence through abnormal conditions region
Disconnected previous frame, if there is not the previous frame through abnormal conditions region decision before present frame, then fixes tentatively interested by each
Region, as preliminary area-of-interest, enters step C;If there is upper through abnormal conditions region decision before present frame
Frame, then continue whether there is trouble point in the original image judging corresponding to this previous frame, if original corresponding to this previous frame
Image does not exist trouble point, then using each tentative area-of-interest as preliminary area-of-interest, enters step C;If on this
There is trouble point in the original image corresponding to one frame, calculate abnormal conditions in the original image corresponding to this previous frame the most respectively
Position of failure point in region and the distance of each the tentative area-of-interest in the original image corresponding to present frame, if distance
More than or equal to the distance threshold d sets, then this tentative area-of-interest is given up;If distance is less than the distance threshold d sets, then
Retain this tentative area-of-interest;And using tentative area-of-interest with a grain of salt for institute as preliminary area-of-interest, enter step
C;
Step C: utilize perspective transform, is transformed to upright by each the preliminary area-of-interest obtained in step B respectively
Rectangle area-of-interest;Subsequently into step D;
Step D: utilize significance level computation model based on human vision model, calculate each rectangle sense respectively emerging
The binaryzation visual saliency map in interest region, and carry out Image semantic classification according to binaryzation visual saliency map result, remove noise and
Non-abnormal conditions region, then judges whether abnormal conditions region;If each rectangle area-of-interest does not exists
Abnormal conditions region, then enter step E;If one or more rectangle area-of-interests exist abnormal conditions region, then enter
Step F;
Step E: obtaining the original image corresponding to the i-th+N+1 frame after present frame skips N frame, N >=1, then by the i-th+N
+ 1 frame, as present frame, returns step A and processes, the original image corresponding to present frame that is i-th+N+1 frame by present frame
Corresponding many power transmission lines in original image are divided into different parallel lines groups;Subsequently into step B;
Step F: utilize inverse perspective mapping to be reverted to corresponding to present frame from rectangle area-of-interest in abnormal conditions region
Original image in, obtain position in the original image corresponding to present frame, the abnormal conditions region, i.e. position of failure point;So
Rear entrance step G;
Step G: using i+1 frame as present frame, returns step A to the original image corresponding to present frame i.e. i+1 frame
Process, many power transmission lines in the original image corresponding to present frame i.e. i+1 frame are divided into different parallel lines
Group;Subsequently into step B.
In described step A, probability Hough straight line mapping algorithm is utilized to calculate the coordinate of each bar power transmission line two-end-point respectively,
And calculate the rectilineal interval between the slope of all power transmission lines and any two power transmission lines, by all angles each other and straight line
Spacing all restraints power transmission line less than the power transmission line of predetermined threshold value as same, and is divided into identical parallel lines group.
Described step A comprises the following steps:
Step A1: choose the n bar power transmission line l in the original image corresponding to present framei, i=1,2 ..., n;Then profit
Every power transmission line l is calculated respectively with probability Hough straight line mapping algorithmiTwo-end-point coordinate
Step A2: calculate every power transmission line l respectivelyiSlope ki, i=1,2 ..., n, then according to every power transmission line li
Slope kiCalculate the angle between any two power transmission lines respectively;
Step A3: calculate the rectilineal interval d between any two power transmission lines in n bar power transmission line respectively, two power transmission lines it
Between rectilineal interval d refer to the minimum air line distance between two power transmission lines;
Step A4: the angle between two power transmission lines is little less than or equal to the rectilineal interval d between 5 °, and two power transmission lines
In the original image length corresponding to present frameTime, it is judged that these two power transmission lines restraint power transmission line for same, and by present frame institute
Corresponding original image is divided into identical parallel lines group with bundle power transmission line.
In described step B, utilize parallel lines group group result, travel through each group of parallel lines group, each group is put down
The point set being made up of the two-end-point of all power transmission lines in row straight line group does convex closure computing, obtains covering whole current parallel lines group
Region, the region obtained after convex closure computing is the polygonal region being made up of straight line portion end points, i.e. fixes tentatively interested
Region.
In described step B, in the original image corresponding to previous frame the position of failure point in abnormal conditions region with work as
The distance of each the tentative area-of-interest in the original image corresponding to front frame, refers to that this trouble point is to each tentative region of interest
The distance on the limit that this trouble point of distance is nearest, distance threshold d in territorysLength for two field picture
Described step C comprises the following steps:
Step C1: the summit of each the preliminary area-of-interest obtained in acquisition step B;Subsequently into step C2;
Step C2: judge the number of vertices of each preliminary area-of-interest respectively:
If this preliminary area-of-interest only has 4 summits as, then these 4 summits are elected 4 datum marks before perspective transform,
And arrange in a clockwise direction;Subsequently into step C4;
If this preliminary area-of-interest has the summit of more than 4, the most first calculate what this preliminary area-of-interest was formed
Polygonal minimum enclosed rectangle, and obtain 4 summits of this minimum enclosed rectangle;If the 4 of this minimum enclosed rectangle tops
Point is respectively positioned on inside the original image corresponding to present frame, then elect 4 datum marks before perspective transform as, and with side clockwise
To arrangement;Subsequently into step C4;If what in the 4 of this minimum enclosed rectangle summits, certain summit was positioned at corresponding to present frame is former
Beginning picture appearance, then enter step C3;
Step C3: be pointed to the summit outside the original image corresponding to present frame and be modified, is set in present frame institute
The corresponding apex coordinate outside original image is (xoutside,youtside),
If summit (xoutside,youtside) go beyond the scope in the x direction, then pass through correction formula
It is modified;
If summit (xoutside,youtside) go beyond the scope in y-direction, then pass through correction formula
It is modified;
Then obtaining the summit outside original image being pointed to corresponding to present frame is (xoutside,youtside) repair
Coordinate (x ', y ') after just, and final with this minimum within the revised original image being respectively positioned on corresponding to present frame outside
Connect 4 summits of rectangle as 4 datum marks before perspective transform, and arrange in a clockwise direction;Subsequently into step C4;
Wherein, ksFor being positioned at the slope of any one power transmission line in this preliminary area-of-interest;xborder,yborderFor distance
Summit (xoutside,youtside) nearest current frame image boundary coordinate;
Step C4: by perspective transform, the preliminary area-of-interest of 4 datum marks before having confirmed that perspective transform converts
For upright rectangle area-of-interest, implement the perspective matrix used by perspective transform by 4 datum marks before perspective transform with saturating
Determining depending on 4 datum marks of matrix upright after conversion, after perspective transform, 4 datum marks of upright matrix are by manually arranging;Perspective
Transformation matrix is 3 × 3 matrixes,
The computational methods of perspective transformation matrix are, for the datum mark before the perspective transform of each group of correspondence positionWith the datum mark after perspective transformTransformational relationWherein t
For yardstick;Subsequently into step C5;
Step C5: collect the RGB color information of all pixels in current preliminary area-of-interest, seek its meansigma methods
And as background colour;Then in current preliminary area-of-interest, there is the place of power transmission line with this background colour to carry out color and repair
Just, i.e. power transmission line position is replaced with this background colour;Subsequently into step C6;
Step C6: with the perspective transformation matrix T obtained, the current frame image after step C5 processes is carried out perspective and becomes
Changing, the rectangular image after conversion is the rectangle area-of-interest that current preliminary area-of-interest is corresponding.
Described step D comprises the following steps:
Step D1: respectively each rectangle area-of-interest is carried out Image semantic classification, i.e. utilizes gaussian filtering to rectangle
Area-of-interest is smoothed, and removes noise;Subsequently into step D2;
Step D2: calculate the binaryzation visual saliency map of each rectangle area-of-interest, the binaryzation after calculating respectively
Visual saliency map is single channel binary image;And using this single channel binary image is the mask of subsequent calculations, should
In mask, white portion i.e. non-zero pixels point marked more significantly region, and black region i.e. zero parts marked secondary notable
Region, calculate the connected domain i.e. area of white portion and the geometric center of all of non-zero pixels point in this mask, and protect
Deposit result;Subsequently into step D3;
Step D3: the connected domain situation being made up of non-zero pixels point in the mask obtained, removes area according to below step
Region less than the minimum area threshold value set, it is determined whether there is abnormal conditions region, and determine the event in abnormal conditions region
Barrier point:
Step d31: set when area sum the most total connected domain area of the connected domain of all non-zero pixels points is less than
Little area threshold ssmallTime, enter step d32;
When total connected domain area is more than maximum area threshold value s setlargeTime, enter step d32;
When total connected domain area is more than or equal to minimum area threshold value s setsmallAnd less than or equal to the maximum area set
Threshold value slargeTime, enter step d33;
Step d32: demarcate in the rectangle area-of-interest carrying out computing and do not have abnormal conditions region, will enter
The rectangle area-of-interest of row operation be labeled as foreign fault;Subsequently into step D4;
Step d33: by the rectangle area-of-interest corresponding to the connected domain of above-mentioned all non-zero pixels points from original square
Shape area-of-interest splits, for original rectangular area-of-interest remainder, collects all pictures in this remainder
The RGB color information of element, seeks its meansigma methods and as background colour;Then exist defeated in this remainder with this background colour
The place of electric wire carries out color correct, i.e. replaces power transmission line position with this background colour;Subsequently into step D1;
Step d34: judge the number of connected domain, if the number of connected domain is only one, demarcates this connected domain as correspondence
Original image in the tentative position of guilty culprit;If the number of connected domain is multiple, choose area in multiple connected domain maximum
Connected domain and the geometric center distance of the connected domain maximum with this area less than setting threshold value dtAll connected domains, and with
Connected domain and the geometric center distance of the connected domain maximum with this area that this area is maximum are less than and set threshold value dtAll
Connected domain formed bunch overall geometry center as the tentative position of guilty culprit in corresponding original image;Subsequently into step
Rapid D4;
Step D4: if carrying out there are abnormal conditions in the rectangle area-of-interest of computing, then enter step F;If just
In the rectangle area-of-interest carry out computing, there is no abnormal conditions region, then exit that currently to carry out the rectangle sense of computing emerging
Interest region also enters the next rectangle area-of-interest without computing, until all in the original image corresponding to present frame
Rectangle area-of-interest all complete identical calculating, if each rectangle area-of-interest does not the most exist abnormal conditions district
Territory, then enter step E.
In described step D:
The filtering core size dimension of Gauss model is 3 × 3, and average is μ=0.5, and variance is σ2=0.64;Binaryzation vision
Calculating of notable figure uses significance detection algorithm based on hidden markov process;The calculating district of binaryzation visual saliency map
Territory is x:0~299, y:0~59;
Minimum area threshold value ssmallFor carrying out the rectangle area-of-interest area of computingMaximum area threshold value
slargeFor carrying out the rectangle area-of-interest area of computingSet threshold value dtEmerging for carrying out the rectangle sense of computing
Interest zone length
In described step F, in the original image obtained after step d34 terminates, the coordinate of the tentative position of guilty culprit is fixed
For (xmalf,ymalf), the perspective transformation matrix T obtained in step C4 is converted to its inverse matrix T-1, utilize inverse matrix T-1To former
In beginning image, the tentative position of guilty culprit carries out perspective transform, obtains position in preliminary area-of-interest, the trouble point, becomes
The process of changing is
Wherein (xmalf',ymalf') it is the position of the trouble point of present frame,
Subsequently into step G.
In described step E, N is frame number shown in a second.
The present invention judges by vision significance feature and positions foreign body to there is position, combines color, shape or space
The characteristics such as distribution, it is achieved the unification to each feature is considered, consider on the time testing result before more simultaneously, statistically increase
The robustness of testing result.And can according to situation change area-of-interest size to be detected speed faster,
Avoid the calculating of a large amount of repetition and reduce the result uncertainty caused because of the multiformity of foreign body failure mode.
Accompanying drawing explanation
Fig. 1 is the schematic flow sheet of the present invention.
Detailed description of the invention
Below in conjunction with drawings and Examples, the present invention is made with detailed description:
As it is shown in figure 1, the electric line foreign matter detection method of view-based access control model significance analysis of the present invention, including with
Lower step:
Step A: using the i-th frame as present frame, i=1,2,3 ...;Utilize the power transmission line spatial parameter detected, the most defeated
The slope of electric wire and the distance with other power transmission line, be divided into a plurality of power transmission line in the original image corresponding to present frame not
Same parallel lines group;Subsequently into step B;
The present invention utilizes probability Hough straight line mapping algorithm to calculate the coordinate of each bar power transmission line two-end-point respectively, and calculates institute
There is the rectilineal interval between the slope of power transmission line and any two power transmission lines, by all angles each other and rectilineal interval the most not
The power transmission line exceeding predetermined threshold value restraints power transmission line as same, and is divided into identical parallel lines group.
Step A includes step in detail below:
Step A1: choose the n bar power transmission line l in the original image corresponding to present framei, i=1,2 ..., n;Then profit
Every power transmission line l is calculated respectively with probability Hough straight line mapping algorithmiTwo-end-point coordinate
Step A2: calculate every power transmission line l respectivelyiSlope ki, i=1,2 ..., n, then according to every power transmission line li
Slope kiCalculate the angle between any two power transmission lines respectively;
Step A3: calculate the rectilineal interval d between any two power transmission lines in n bar power transmission line respectively, two power transmission lines it
Between rectilineal interval d refer to the minimum air line distance between two power transmission lines;
Step A4: the angle between two power transmission lines is little less than or equal to the rectilineal interval d between 5 °, and two power transmission lines
In the original image length corresponding to present frameTime, it is judged that these two power transmission lines restraint power transmission line for same, and by present frame institute
Corresponding original image is divided into identical parallel lines group with bundle power transmission line.
Step B: utilize the group result of the parallel lines group obtained in step A, obtain each parallel lines group institute respectively
The region that covers also is set to tentative area-of-interest;Then whether exist before judging present frame and sentence through abnormal conditions region
Disconnected previous frame, if there is not the previous frame through abnormal conditions region decision before present frame, then fixes tentatively interested by each
Region, as preliminary area-of-interest, enters step C, and this step is mainly used in judging whether present frame is the first frame, if the
The most there is not previous frame in one frame;If there is the previous frame through abnormal conditions region decision before present frame, then continue to judge this
Whether the original image corresponding to previous frame exists trouble point, if the original image corresponding to this previous frame does not exist fault
Point, then using each tentative area-of-interest as preliminary area-of-interest, enter step C;If original corresponding to this previous frame
Image exists trouble point, calculates the position, trouble point in abnormal conditions region in the original image corresponding to this previous frame the most respectively
Put and the distance of each the tentative area-of-interest in the original image corresponding to present frame, if distance more than or equal to set away from
From threshold value ds, then this tentative area-of-interest is given up;If distance is less than the distance threshold d sets, then retain this and fix tentatively interested
Region;And using tentative area-of-interest with a grain of salt for institute as preliminary area-of-interest, enter step C;
In the present invention, utilize parallel lines group group result, travel through each group of parallel lines group, to each group of parallel lines
The point set being made up of the two-end-point of all power transmission lines in group does convex closure computing, obtains covering the district of whole current parallel lines group
Territory, the region obtained after convex closure computing is the polygonal region being made up of straight line portion end points, i.e. fixes tentatively area-of-interest.
In original image corresponding to previous frame the position of failure point in abnormal conditions region with in original image corresponding to present frame
The distance of each tentative area-of-interest, refer to that this trouble point is nearest to this trouble point of distance in each tentative area-of-interest
Article one, the distance on limit, distance threshold dsLength for two field picture
Step C: utilize perspective transform, is transformed to upright by each the preliminary area-of-interest obtained in step B respectively
Rectangle area-of-interest;Subsequently into step D;
In the present invention, step C includes step in detail below:
Step C1: the summit of each the preliminary area-of-interest obtained in acquisition step B;Subsequently into step C2;
Step C2: judge the number of vertices of each preliminary area-of-interest respectively:
If this preliminary area-of-interest only has 4 summits as, then these 4 summits are elected 4 datum marks before perspective transform,
And arrange in a clockwise direction;Subsequently into step C4;
If this preliminary area-of-interest has the summit of more than 4, the most first calculate what this preliminary area-of-interest was formed
Polygonal minimum enclosed rectangle, and obtain 4 summits of this minimum enclosed rectangle;If the 4 of this minimum enclosed rectangle tops
Point is respectively positioned on inside the original image corresponding to present frame, then elect 4 datum marks before perspective transform as, and with side clockwise
To arrangement;Subsequently into step C4;If what in the 4 of this minimum enclosed rectangle summits, certain summit was positioned at corresponding to present frame is former
Beginning picture appearance, then enter step C3;
Step C3: be pointed to the summit outside the original image corresponding to present frame and be modified, is set in present frame institute
The corresponding apex coordinate outside original image is (xoutside,youtside),
If summit (xoutside,youtside) go beyond the scope in the x direction, then pass through correction formula
It is modified;
If summit (xoutside,youtside) go beyond the scope in y-direction, then pass through correction formula
It is modified;
Then obtaining the summit outside original image being pointed to corresponding to present frame is (xoutside,youtside) repair
Coordinate (x ', y ') after just, and final with this minimum within the revised original image being respectively positioned on corresponding to present frame outside
Connect 4 summits of rectangle as 4 datum marks before perspective transform, and arrange in a clockwise direction;Subsequently into step C4;
Wherein, ksFor being positioned at the slope of any one power transmission line in this preliminary area-of-interest;xborder,yborderFor distance
Summit (xoutside,youtside) nearest current frame image boundary coordinate;
Step C4: by perspective transform, the preliminary area-of-interest of 4 datum marks before having confirmed that perspective transform converts
For upright rectangle area-of-interest, implement the perspective matrix used by perspective transform by 4 datum marks before perspective transform with saturating
Determining depending on 4 datum marks of matrix upright after conversion, after perspective transform, 4 datum marks of upright matrix are by manually arranging;Perspective
Transformation matrix is 3 × 3 matrixes,
The computational methods of perspective transformation matrix are, for the datum mark before the perspective transform of each group of correspondence positionWith the datum mark after perspective transformTransformational relation
Wherein t is yardstick;Subsequently into step C5;
Step C5: collect the RGB color information of all pixels in current preliminary area-of-interest, seek its meansigma methods
And as background colour;Then in current preliminary area-of-interest, there is the place of power transmission line with this background colour to carry out color and repair
Just, i.e. power transmission line position is replaced with this background colour;Subsequently into step C6;
Step C6: with the perspective transformation matrix T obtained, the current frame image after step C5 processes is carried out perspective and becomes
Changing, the rectangular image after conversion is the rectangle area-of-interest that current preliminary area-of-interest is corresponding.
Step D: utilize significance level computation model based on human vision model, this model at Bowen Jiang,
Lihe Zhang, Huchuan Lu, Chuan Yang, and Ming-Hsuan Yang in 2013 at " International
Conference on Computer Vision " (international machine vision meeting) paper " Saliency of issuing above
Detection viaAbsorbing Markov Chain " in be suggested, calculate the two of each rectangle area-of-interest respectively
Value visual saliency map, and carry out Image semantic classification according to binaryzation visual saliency map result, remove noise and non-abnormal conditions
Region, then judges whether abnormal conditions region;If the most there is not abnormal conditions district in each rectangle area-of-interest
Territory, then enter step E;If one or more rectangle area-of-interests exist abnormal conditions region, then enter step F;
In the present invention, step D includes step in detail below:
Step D1: respectively each rectangle area-of-interest is carried out Image semantic classification, i.e. utilizes gaussian filtering to rectangle
Area-of-interest is smoothed, and removes noise;Subsequently into step D2;
Step D2: calculate the binaryzation visual saliency map of each rectangle area-of-interest, the binaryzation after calculating respectively
Visual saliency map is single channel binary image;And using this single channel binary image is the mask of subsequent calculations, should
In mask, white portion i.e. non-zero pixels point marked more significantly region, and black region i.e. zero parts marked secondary notable
Region, calculate the connected domain i.e. area of white portion and the geometric center of all of non-zero pixels point in this mask, and protect
Deposit result;Subsequently into step D3;
Step D3: the connected domain situation being made up of non-zero pixels point in the mask obtained, removes area according to below step
Region less than the minimum area threshold value set, it is determined whether there is abnormal conditions region, and determine the event in abnormal conditions region
Barrier point:
Step d31: set when area sum the most total connected domain area of the connected domain of all non-zero pixels points is less than
Little area threshold ssmallTime, enter step d32;
When total connected domain area is more than maximum area threshold value s setlargeTime, enter step d32;
When total connected domain area is more than or equal to minimum area threshold value s setsmallAnd less than or equal to the maximum area set
Threshold value slargeTime, enter step d33;
Step d32: demarcate in the rectangle area-of-interest carrying out computing and do not have abnormal conditions region, will enter
The rectangle area-of-interest of row operation be labeled as foreign fault;Subsequently into step D4;
Step d33: by the rectangle area-of-interest corresponding to the connected domain of above-mentioned all non-zero pixels points from original square
Shape area-of-interest splits, for original rectangular area-of-interest remainder, collects all pictures in this remainder
The RGB color information of element, seeks its meansigma methods and as background colour;Then exist defeated in this remainder with this background colour
The place of electric wire carries out color correct, i.e. replaces power transmission line position with this background colour;Subsequently into step D1;
Step d34: judge the number of connected domain, if the number of connected domain is only one, demarcates this connected domain as correspondence
Original image in the tentative position of guilty culprit;If the number of connected domain is multiple, choose area in multiple connected domain maximum
Connected domain and the geometric center distance of the connected domain maximum with this area less than setting threshold value dtAll connected domains, and with
Connected domain and the geometric center distance of the connected domain maximum with this area that this area is maximum are less than and set threshold value dtAll
Connected domain formed bunch overall geometry center as the tentative position of guilty culprit in corresponding original image;Subsequently into step
Rapid D4;
Step D4: if carrying out there are abnormal conditions in the rectangle area-of-interest of computing, then enter step F;If just
In the rectangle area-of-interest carry out computing, there is no abnormal conditions region, then exit that currently to carry out the rectangle sense of computing emerging
Interest region also enters the next rectangle area-of-interest without computing, until all in the original image corresponding to present frame
Rectangle area-of-interest all complete identical calculating, if each rectangle area-of-interest does not the most exist abnormal conditions district
Territory, then enter step E.
In the present invention, in step D1, the filtering core size dimension of Gauss model is 3 × 3, and average is μ=0.5, and variance is σ2
=0.64;
In step D2, the calculating of binaryzation visual saliency map uses based on BowenJiang, Lihe Zhang, Huchuan
Lu, Chuan Yang, and Ming-Hsuan Yang in 2013 at " International Conference on
Computer Vision " (international machine vision meeting) paper of issuing above
In " SaliencyDetectionviaAbsorbing Markov Chain ", design is based on hidden markov process aobvious
Work property detection algorithm;The zoning of binaryzation visual saliency map is x:0~299, y:0~59;
In step D3, minimum area threshold value ssmallFor carrying out the rectangle area-of-interest area of computingMaximum
Area threshold slargeFor carrying out the rectangle area-of-interest area of computingSet threshold value dtFor carrying out computing
Rectangle region of interest length of field
Step E: obtaining the original image corresponding to the i-th+N+1 frame after present frame skips N frame, N >=1, then by the i-th+N
+ 1 frame, as present frame, returns step A and processes, the original image corresponding to present frame that is i-th+N+1 frame by present frame
Corresponding many power transmission lines in original image are divided into different parallel lines groups;Subsequently into step B;
In the present invention, N is frame number shown in a second.
Step F: utilize inverse perspective mapping to be reverted to corresponding to present frame from rectangle area-of-interest in abnormal conditions region
Original image in, obtain position in the original image corresponding to present frame, the abnormal conditions region, i.e. position of failure point;So
Rear entrance step G;
In the present invention, in the original image obtained after step d34 terminates, the coordinate of the tentative position of guilty culprit is set to
(xmalf,ymalf), the perspective transformation matrix T obtained in step C4 is converted to its inverse matrix T-1, utilize inverse matrix T-1To original
In image, the tentative position of guilty culprit carries out perspective transform, obtains position in preliminary area-of-interest, the trouble point, conversion
Process is
Wherein (xmalf',ymalf') it is the position of the trouble point of present frame,
Subsequently into step G.
Step G: using i+1 frame as present frame, returns step A to the original image corresponding to present frame i.e. i+1 frame
Process, many power transmission lines in the original image corresponding to present frame i.e. i+1 frame are divided into different parallel lines
Group;Subsequently into step B.
Embodiment 1:
Hunan is sent power transformation monitoring unmanned video sequence (1920 × 1080 pixels, 30fps) to process by the present embodiment.
Have chosen two kinds of foreign bodies to test, one is kite, and another is branch.This video Scene background is in dynamically change
In, jolt time during video camera steadily, background complexity foreign material are more and foreign material shape and color regime complex, illumination becomes
Change scope is the biggest.Program is write based on OpenCV2.44, and the image coordinate initial point in video is positioned at the upper left corner.The present embodiment bag
Include following steps:
Step A: in the video-frequency band chosen, the first frame of selecting video section is acquired, and obtains the original image of correspondence
In 9 power transmission line li, i=1,2 ..., 9;Then probability Hough straight line mapping algorithm is utilized to calculate every power transmission line l respectivelyi
Two-end-point coordinate
Step A2: calculate every power transmission line l respectivelyiSlope ki, i=1,2 ..., 9, then according to every power transmission line li
Slope kiCalculate any two power transmission line l respectivelyiWith ljBetween angle;
Step A3: calculate any two power transmission line l in 9 power transmission lines respectivelyiWith ljBetween rectilineal interval d, due at frame
The length of image cathetus section is limited, two power transmission line liWith ljBetween rectilineal interval dijUse between two power transmission lines
Short lines distance;
Step A4: the angle between two power transmission lines is little less than or equal to the rectilineal interval d between 5 °, and two power transmission lines
In the original image length corresponding to present frameTime, it is judged that these two power transmission lines restraint power transmission line for same, and by present frame institute
Corresponding original image is divided into identical parallel lines group with bundle power transmission line.The present embodiment obtains in the two field picture chosen
To line grouping result be four groups, the straight line number often organized is (n1,n2,n3,n4)=(3,2,2,2).
In step B, the present frame owing to choosing in the present embodiment is the first frame, there is not previous frame, the most directly determines
Each tentative area-of-interest is as preliminary area-of-interest.
Utilize parallel lines group group result, travel through each group of parallel lines group, in each group of parallel lines group by institute
The point set being made up of the two-end-point of power transmission line does convex closure computing, obtains covering the region of whole current parallel lines group, through excess convexity
The region that contracted affreightment obtains after calculating is the polygonal region being made up of straight line portion end points, i.e. fixes tentatively area-of-interest, is also preliminary
Area-of-interest.In the present invention, often all two Extreme points sets of group straight line group are (xi,yi), i=1,2 ..., 2n, wherein n
Bar number for this group straight line.In order to not lose tentative area-of-interest, can cover by these end points are carried out convex closure computing
In view of the precision problem of straight-line detection while whole tentative area-of-interest.Convex closure computing selects the API of OpenCV to carry out
Calculating, in the present embodiment, calculated four initial irregular polygon area-of-interests are
The preliminary area-of-interest obtained in step B is irregular polygon, gathers this polygonal all tops
Point, in the present embodiment, the number of vertices of four area-of-interests is respectively (n1,n2,n3,n4)=(5,4,4,4).
Step C1: the summit of each the preliminary area-of-interest obtained in acquisition step B;Subsequently into step C2;
Step C2: judge the number of vertices of each preliminary area-of-interest respectively:
If this preliminary area-of-interest only has 4 summits as, then these 4 summits are elected 4 datum marks before perspective transform
(pt1,pt2,pt3,pt4), and arrange in a clockwise direction;Subsequently into step C4;
If this preliminary area-of-interest has the summit of more than 4, the most first calculate what this preliminary area-of-interest was formed
Polygonal minimum enclosed rectangle Rectmin, and obtain 4 summits of this minimum enclosed rectangleIf the 4 of this minimum enclosed rectangle summits are respectively positioned in the original image corresponding to present frame
Portion, then elect 4 datum marks before perspective transform as, and arrange in a clockwise direction;Subsequently into step C4;If outside this minimum
Connect certain summit in 4 summits of rectangle and be positioned at outside the original image corresponding to present frame, then enter step C3;
Step C3: be pointed to the summit outside the original image corresponding to present frame and be modified, is set in present frame institute
The corresponding apex coordinate outside original image is (xoutside,youtside),
If summit (xoutside,youtside) go beyond the scope in the x direction, then pass through correction formula
It is modified;
If summit (xoutside,youtside) go beyond the scope in y-direction, then pass through correction formula
It is modified;
Then obtaining the summit outside original image being pointed to corresponding to present frame is (xoutside,youtside) repair
Coordinate (x ', y ') after just, and final with this minimum within the revised original image being respectively positioned on corresponding to present frame outside
Connect 4 summits of rectangle as 4 datum marks before perspective transform, and arrange in a clockwise direction;Subsequently into step C4;
Wherein, ksFor being positioned at the slope of any one power transmission line in this preliminary area-of-interest;xborder,yborderFor distance
The current frame image boundary coordinate that summit is nearest.
Step C4: by perspective transform, the preliminary area-of-interest of 4 each and every one datum marks before having confirmed that perspective transform becomes
Be changed to upright rectangle area-of-interest, implement perspective matrix used by perspective transform by 4 datum marks before perspective transform and
After perspective transform, 4 datum marks of upright matrix determine, after perspective transform, 4 datum marks of upright matrix are by manually arranging this reality
Executing datum mark after 4 selected in example convert is (pt1',pt2',pt3',pt4')=((0,0), (300,0), (0,60),
(300,60)).Perspective transformation matrix is 3 × 3 matrixes,
The computational methods of perspective transformation matrix are, for the datum mark before the perspective transform of each group of correspondence positionWith the datum mark after perspective transformTransformational relationIts
Middle t is yardstick;Subsequently into step C5;
Step C5: collect the RGB color information of all pixels in current preliminary area-of-interest, seek its meansigma methods
And as background colour;Then in current preliminary area-of-interest, there is the place of power transmission line with this background colour to carry out color and repair
Just, i.e. power transmission line position is replaced with this background colour;Subsequently into step C6;
Step C6: with the perspective transformation matrix T obtained, the current frame image after step C5 processes is carried out perspective and becomes
Changing, the rectangular image after conversion is upright rectangle area-of-interest.
Step D: utilize significance level computation model based on human vision model, calculate each rectangle sense respectively emerging
The binaryzation visual saliency map in interest region, and carry out Image semantic classification according to binaryzation visual saliency map result, remove noise and
Non-abnormal conditions region, then judges whether abnormal conditions region;If each rectangle area-of-interest does not exists
Abnormal conditions region, then enter step E;If one or more rectangle area-of-interests exist abnormal conditions region, then enter
Step F;
In the present invention, step D includes step in detail below:
Step D1: respectively each rectangle area-of-interest is carried out Image semantic classification, i.e. utilizes gaussian filtering to rectangle
Area-of-interest is smoothed, and removes noise;The filtering core size dimension of Gauss model used in the present embodiment is 3 ×
3, average is μ=0.5, and variance is σ2=0.64.Subsequently into step D2;
Step D2: calculate the binaryzation visual saliency map of each rectangle area-of-interest, the binaryzation after calculating respectively
Visual saliency map is single channel binary image;Arranging the visual saliency map after calculating is matrix mask during subsequent calculations
Mask, according to this mask image, calculates the area S of the i.e. white portion of connected domain at the most all of non-zero pixels pointiAnd
Geometric center (xi,yi), i=1,2 ..., n, and preserve result;Subsequently into step D3;
The present embodiment calculates visual saliency map used at algorithm based on Bowen Jiang, Lihe Zhang,
Huchuan Lu, Chuan Yang, and Ming-Hsuan Yang in 2013 at " International Conference
On Computer Vision " (international machine vision meeting) paper " Saliency Detection of issuing above
ViaAbsorbing Markov Chain " the middle significance detection algorithm based on hidden markov process designed
Step D3: the connected domain situation being made up of non-zero pixels point in the mask obtained, removes all according to below step
Area sum the most total connected domain area area of the connected domain of non-zero pixels point is less than the region of the minimum area threshold value set, really
Determine whether there is abnormal conditions region, and determine the guilty culprit point in abnormal conditions region:
Step d31: when total connected domain area is less than minimum area threshold value S setsmallWhen=200, enter step d32;
When total connected domain area is more than maximum area threshold value S setlargeWhen=6000, enter step d32;
When total connected domain area is more than or equal to minimum area threshold value S setsmall=200 and less than or equal to set maximum
Area threshold SlargeWhen=6000, enter step d33;
Step d32: demarcate in the rectangle area-of-interest carrying out computing and do not have abnormal conditions region, will enter
The rectangle area-of-interest of row operation be labeled as foreign fault;Subsequently into step D4;
Step d33: by the rectangle area-of-interest corresponding to the connection of above-mentioned all non-zero pixels points from original rectangle
Area-of-interest splits, for original rectangular area-of-interest remainder, collects all pixels in this remainder
RGB color information, seek its meansigma methods and as background colour;Then in this remainder, there is transmission of electricity with this background colour
The place of line carries out color correct, i.e. replaces power transmission line position with this background colour;Subsequently into step D1;
Step d34: judge the number of connected domain, if the number of connected domain is only one, demarcates this connected domain as correspondence
Original image in the tentative position of guilty culprit;If the number of connected domain is multiple, choose area in multiple connected domain maximum
Connected domain and the geometric center distance of the connected domain maximum with this area less than setting threshold value dtAll connected domains of=30,
And be less than set threshold value d with the connected domain of this area maximum and the geometric center distance of the connected domain maximum with this areat=
All connected domains of 30 formed bunch overall geometry center as the tentative position of guilty culprit in corresponding original image;So
Rear entrance step D4;
Step D4: if carrying out there are abnormal conditions in the rectangle area-of-interest of computing, then enter step F;If just
In the rectangle area-of-interest carry out computing, there is no abnormal conditions region, then exit that currently to carry out the rectangle sense of computing emerging
Interest region also enters the next rectangle area-of-interest without computing, until all in the original image corresponding to present frame
Rectangle area-of-interest all complete identical calculating, if each rectangle area-of-interest does not the most exist abnormal conditions district
Territory, then enter step E.
Step E: obtaining the original image corresponding to the i-th+N+1 frame after present frame skips N frame, N >=1, then by the i-th+N
+ 1 frame, as present frame, returns step A and processes, the original image corresponding to present frame that is i-th+N+1 frame by present frame
And many power transmission lines in the original image that i-th corresponding to+N+1 frame are divided into different parallel lines groups;Subsequently into step
B;In the present invention, N is frame number shown in a second.The present embodiment is according to frame rate 30fps of experiment video, the jump of setting
Frame number be 20~30 frames, i.e. process again after 1s.This step can utilize former frame to obtain result, and currently
Preliminary area-of-interest detected by frame, skips some region substantially not havinging fault, improves judging efficiency.
Step F: utilize inverse perspective mapping to be reverted to corresponding to present frame from rectangle area-of-interest in abnormal conditions region
Original image in, obtain position in the original image corresponding to present frame, the abnormal conditions region, i.e. position of failure point;So
Rear entrance step G;
In the present invention, in the original image obtained after step d34 terminates, the coordinate of the tentative position of guilty culprit is set to
(xmalf,ymalf), the perspective transformation matrix T obtained in step C4 is converted to its inverse matrix T-1, utilize inverse matrix T-1To artwork
In Xiang, the tentative position of guilty culprit carries out perspective transform and obtains position in preliminary area-of-interest, the trouble point, conversion process
For
Wherein (xmalf',ymalf') it is the position of the trouble point of present frame.
The present embodiment is centered by this point, and the length of side is 100 work squares, and this square area indicates the region of guilty culprit;Then
Enter step G.
Step G: using i+1 frame that is second frame as present frame, returns former to corresponding to present frame that is second frame of step A
Beginning image processes, and many power transmission lines in the original image corresponding to present frame i.e. i+1 frame are divided into different putting down
Row straight line group;Subsequently into step B.
Claims (10)
1. the electric line foreign matter detection method of a view-based access control model significance analysis, it is characterised in that comprise the following steps:
Step A: using the i-th frame as present frame, i=1,2,3 ...;Utilize the power transmission line spatial parameter detected, i.e. power transmission line
Slope and with the distance of other power transmission line, a plurality of power transmission line in the original image corresponding to present frame is divided into different
Parallel lines group;Subsequently into step B;
Step B: utilize the group result of the parallel lines group obtained in step A, obtain each parallel lines group respectively and covered
Region and be set to tentative area-of-interest;Then whether exist through abnormal conditions region decision before judging present frame
, if there is not the previous frame through abnormal conditions region decision before present frame, then by each tentative area-of-interest in previous frame
As preliminary area-of-interest, enter step C;If there is the previous frame through abnormal conditions region decision before present frame, then
Continue to judge whether the original image corresponding to this previous frame exists trouble point, if in the original image corresponding to this previous frame
There is not trouble point, then using each tentative area-of-interest as preliminary area-of-interest, enter step C;If this previous frame institute
There is trouble point in corresponding original image, calculate in the original image corresponding to this previous frame in abnormal conditions region the most respectively
The distance of position of failure point and each the tentative area-of-interest in original image corresponding to present frame, if distance is more than
In the distance threshold d sets, then this tentative area-of-interest is given up;If distance is less than the distance threshold d sets, then retaining should
Tentative area-of-interest;And using tentative area-of-interest with a grain of salt for institute as preliminary area-of-interest, enter step C;
Step C: utilize perspective transform, is transformed to upright rectangle respectively by each the preliminary area-of-interest obtained in step B
Area-of-interest;Subsequently into step D;
Step D: utilize significance level computation model based on human vision model, calculate each rectangle region of interest respectively
The binaryzation visual saliency map in territory, and carry out Image semantic classification according to binaryzation visual saliency map result, removes noise and non-different
Reason condition region, then judges whether abnormal conditions region;If the most there is not exception in each rectangle area-of-interest
Situation region, then enter step E;If one or more rectangle area-of-interests exist abnormal conditions region, then enter step
F;
Step E: obtaining the original image corresponding to the i-th+N+1 frame after present frame skips N frame, N >=1, then by the i-th+N+1 frame
As present frame, return step A and the original image corresponding to present frame that is i-th+N+1 frame is processed, by right for present frame institute
Many power transmission lines in the original image answered are divided into different parallel lines groups;Subsequently into step B;
Step F: utilize inverse perspective mapping by abnormal conditions region from rectangle area-of-interest revert to corresponding to present frame former
In beginning image, obtain position in the original image corresponding to present frame, the abnormal conditions region, i.e. position of failure point;Then enter
Enter step G;
Step G: using i+1 frame as present frame, returns step A and carries out the original image corresponding to present frame i.e. i+1 frame
Process, many power transmission lines in the original image corresponding to present frame i.e. i+1 frame are divided into different parallel lines groups;So
Rear entrance step B.
The electric line foreign matter detection method of view-based access control model significance analysis the most according to claim 1, it is characterised in that:
In described step A, utilize probability Hough straight line mapping algorithm to calculate the coordinate of each bar power transmission line two-end-point respectively, and calculate institute
There is the rectilineal interval between the slope of power transmission line and any two power transmission lines, by all angles each other and rectilineal interval the most not
The power transmission line exceeding predetermined threshold value restraints power transmission line as same, and is divided into identical parallel lines group.
The electric line foreign matter detection method of view-based access control model significance analysis the most according to claim 1 and 2, its feature exists
In, described step A comprises the following steps:
Step A1: choose the n bar power transmission line l in the original image corresponding to present framei, i=1,2 ..., n;Then utilize general
Rate Hough straight line mapping algorithm calculates every power transmission line l respectivelyiTwo-end-point coordinate
Step A2: calculate every power transmission line l respectivelyiSlope ki, i=1,2 ..., n, then according to every power transmission line liOblique
Rate kiCalculate the angle between any two power transmission lines respectively;
Step A3: calculate the rectilineal interval d between any two power transmission lines in n bar power transmission line respectively, between two power transmission lines
Rectilineal interval d refers to the minimum air line distance between two power transmission lines;
Step A4: the angle between two power transmission lines is less than less than or equal to the rectilineal interval d between 5 °, and two power transmission lines works as
Original image length corresponding to front frameTime, it is judged that these two power transmission lines restraint power transmission line for same, and by corresponding to present frame
Original image in be divided into identical parallel lines group with bundle power transmission line.
The electric line foreign matter detection method of view-based access control model significance analysis the most according to claim 1, it is characterised in that:
In described step B, utilize parallel lines group group result, travel through each group of parallel lines group, to each group of parallel lines group
In the point set that is made up of the two-end-point of all power transmission lines do convex closure computing, obtain covering the region of whole current parallel lines group,
The region obtained after convex closure computing is the polygonal region being made up of straight line portion end points, i.e. fixes tentatively area-of-interest.
The electric line foreign matter detection method of view-based access control model significance analysis the most according to claim 1, it is characterised in that:
In described step B, in the original image corresponding to previous frame, the position of failure point in abnormal conditions region is right with present frame institute
The distance of each the tentative area-of-interest in the original image answered, refers to that this trouble point is to distance in each tentative area-of-interest
The distance on the limit that this trouble point is nearest, distance threshold dsLength for two field picture
The electric line foreign matter detection method of view-based access control model significance analysis the most according to claim 1, it is characterised in that
Described step C comprises the following steps:
Step C1: the summit of each the preliminary area-of-interest obtained in acquisition step B;Subsequently into step C2;
Step C2: judge the number of vertices of each preliminary area-of-interest respectively:
If this preliminary area-of-interest only has 4 summits as, then these 4 summits are elected 4 datum marks before perspective transform, and
Arrange in a clockwise direction;Subsequently into step C4;
If this preliminary area-of-interest has the summit of more than 4, the most first calculate that this preliminary area-of-interest formed is polygon
The minimum enclosed rectangle of shape, and obtain 4 summits of this minimum enclosed rectangle;If the 4 of this minimum enclosed rectangle summits are equal
It is positioned at inside the original image corresponding to present frame, then elects 4 datum marks before perspective transform as, and arrange in a clockwise direction
Cloth;Subsequently into step C4;If certain summit is positioned at the original graph corresponding to present frame in the 4 of this minimum enclosed rectangle summits
As outside, then enter step C3;
Step C3: be pointed to the summit outside the original image corresponding to present frame and be modified, be set in corresponding to present frame
The apex coordinate outside original image be (xoutside,youtside),
If summit (xoutside,youtside) go beyond the scope in the x direction, then pass through correction formula
It is modified;
If summit (xoutside,youtside) go beyond the scope in y-direction, then pass through correction formula
It is modified;
Then obtaining the summit outside original image being pointed to corresponding to present frame is (xoutside,youtside) be modified after
Coordinate (x ', y '), and final with this minimum external square within the revised original image being respectively positioned on corresponding to present frame
4 summits of shape are as 4 datum marks before perspective transform, and arrange in a clockwise direction;Subsequently into step C4;
Wherein, ksFor being positioned at the slope of any one power transmission line in this preliminary area-of-interest;xborder,yborderFor distance summit
(xoutside,youtside) nearest current frame image boundary coordinate;
Step C4: by perspective transform, the preliminary area-of-interest of 4 datum marks before having confirmed that perspective transform is transformed to directly
Vertical rectangle area-of-interest, implements the perspective matrix used by perspective transform and is become by 4 datum marks before perspective transform and perspective
After changing, 4 datum marks of upright matrix determine, after perspective transform, 4 datum marks of upright matrix are by manually arranging;Perspective transform
Matrix is 3 × 3 matrixes,
The computational methods of perspective transformation matrix are, for the datum mark before the perspective transform of each group of correspondence positionWith the datum mark after perspective transformTransformational relationWherein t
For yardstick;Subsequently into step C5;
Step C5: collect the RGB color information of all pixels in current preliminary area-of-interest, seek its meansigma methods and make
For background colour;Then in current preliminary area-of-interest, there is the place of power transmission line with this background colour and carry out color correct, i.e.
Power transmission line position is replaced with this background colour;Subsequently into step C6;
Step C6: with the perspective transformation matrix T obtained, the current frame image after step C5 processes is carried out perspective transform, become
Rectangular image after alternatively is the rectangle area-of-interest that current preliminary area-of-interest is corresponding.
The electric line foreign matter detection method of view-based access control model significance analysis the most according to claim 1, it is characterised in that
Described step D comprises the following steps:
Step D1: respectively each rectangle area-of-interest is carried out Image semantic classification, i.e. utilizes gaussian filtering emerging to rectangle sense
Interest region is smoothed, and removes noise;Subsequently into step D2;
Step D2: calculate the binaryzation visual saliency map of each rectangle area-of-interest respectively, the binaryzation vision after calculating
Notable figure is single channel binary image;And using this single channel binary image is the mask of subsequent calculations, in this mask
White portion i.e. non-zero pixels point marked more significantly region, and black region i.e. zero parts marked time significant district
Territory, calculates the connected domain i.e. area of white portion and the geometric center of all of non-zero pixels point in this mask, and preserves knot
Really;Subsequently into step D3;
Step D3: the connected domain situation being made up of non-zero pixels point in the mask obtained, removes area according to below step and is less than
The region of the minimum area threshold value set, it is determined whether there is abnormal conditions region, and determine the fault institute in abnormal conditions region
Point:
Step d31: when area sum the most total connected domain area of the connected domain of all non-zero pixels points is less than the minimal face set
Long-pending threshold value ssmallTime, enter step d32;
When total connected domain area is more than maximum area threshold value s setlargeTime, enter step d32;
When total connected domain area is more than or equal to minimum area threshold value s setsmallAnd less than or equal to the maximum area threshold value set
slargeTime, enter step d33;
Step d32: demarcate in the rectangle area-of-interest carrying out computing and do not have abnormal conditions region, will transport
Calculate rectangle area-of-interest be labeled as foreign fault;Subsequently into step D4;
Step d33: by the rectangle area-of-interest corresponding to the connected domain of above-mentioned all non-zero pixels points from original rectangle sense
Interest splits in region, for original rectangular area-of-interest remainder, collects all pixels in this remainder
RGB color information, seeks its meansigma methods and as background colour;Then in this remainder, there is power transmission line with this background colour
Place carry out color correct, i.e. replace power transmission line position with this background colour;Subsequently into step D1;
Step d34: judge the number of connected domain, if the number of connected domain is only one, demarcates this connected domain as corresponding former
The tentative position of guilty culprit in image;If the number of connected domain is multiple, choose the company that in multiple connected domain, area is maximum
The geometric center distance of logical territory and the connected domain maximum with this area is less than setting threshold value dtAll connected domains, and with this face
The geometric center distance of long-pending maximum connected domain and the connected domain maximum with this area is less than setting threshold value dtAll connections
Territory formed bunch overall geometry center as the tentative position of guilty culprit in corresponding original image;Subsequently into step
D4;
Step D4: if carrying out there are abnormal conditions in the rectangle area-of-interest of computing, then enter step F;If entering
The rectangle area-of-interest of row operation does not has abnormal conditions region, then exits the rectangle region of interest currently carrying out computing
Territory also enters the next rectangle area-of-interest without computing, until all of square in the original image corresponding to present frame
Shape area-of-interest all completes identical calculating, if the most there is not abnormal conditions region in each rectangle area-of-interest, then
Enter step E.
The electric line foreign matter detection method of view-based access control model significance analysis the most according to claim 7, it is characterised in that:
In described step D:
The filtering core size dimension of Gauss model is 3 × 3, and average is μ=0.5, and variance is σ2=0.64;Binaryzation vision is notable
Calculating of figure uses significance detection algorithm based on hidden markov process;The zoning of binaryzation visual saliency map is
X:0~299, y:0~59;
Minimum area threshold value ssmallFor carrying out the rectangle area-of-interest area of computingMaximum area threshold value slarge
For carrying out the rectangle area-of-interest area of computingSet threshold value dtFor carrying out the rectangle region of interest of computing
Length of field
The electric line foreign matter detection method of view-based access control model significance analysis the most according to claim 8, it is characterised in that
In described step F, in the original image obtained after step d34 terminates, the coordinate of the tentative position of guilty culprit is set to (xmalf,
ymalf), the perspective transformation matrix T obtained in step C4 is converted to its inverse matrix T-1, utilize inverse matrix T-1To in original image
The tentative position of guilty culprit carries out perspective transform, obtains position in preliminary area-of-interest, the trouble point, and conversion process is
Wherein (xmalf',ymalf') it is the position of the trouble point of present frame, then
Enter step G.
The electric line foreign matter detection method of view-based access control model significance analysis the most according to claim 1, its feature exists
In, in described step E, N is frame number shown in a second.
Applications Claiming Priority (2)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN2015102104419 | 2015-04-29 | ||
CN201510210441 | 2015-04-29 |
Publications (2)
Publication Number | Publication Date |
---|---|
CN105957077A true CN105957077A (en) | 2016-09-21 |
CN105957077B CN105957077B (en) | 2019-01-15 |
Family
ID=56915756
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201610269864.2A Active CN105957077B (en) | 2015-04-29 | 2016-04-27 | The electric line foreign matter detection method of view-based access control model significance analysis |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN105957077B (en) |
Cited By (10)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN107680095A (en) * | 2017-10-25 | 2018-02-09 | 哈尔滨理工大学 | The electric line foreign matter detection of unmanned plane image based on template matches and optical flow method |
CN108710883A (en) * | 2018-06-04 | 2018-10-26 | 国网辽宁省电力有限公司信息通信分公司 | A kind of complete conspicuousness object detecting method using contour detecting |
CN109145905A (en) * | 2018-08-29 | 2019-01-04 | 河海大学常州校区 | A kind of transmission line of electricity accessory detection method of view-based access control model conspicuousness |
CN110414308A (en) * | 2019-05-16 | 2019-11-05 | 南京理工大学 | A kind of target identification method for dynamic foreign matter on transmission line of electricity |
CN110889328A (en) * | 2019-10-21 | 2020-03-17 | 大唐软件技术股份有限公司 | Method, device, electronic equipment and storage medium for detecting road traffic condition |
CN112184816A (en) * | 2020-09-28 | 2021-01-05 | 珠海博明视觉科技有限公司 | Lithium battery flexible positioning method based on battery tabs |
CN112991375A (en) * | 2021-02-08 | 2021-06-18 | 上海通办信息服务有限公司 | Method and system for reshaping arbitrary-shaped image area into N rectangular areas |
CN113743151A (en) * | 2020-05-27 | 2021-12-03 | 顺丰科技有限公司 | Method and device for detecting road surface sprinkled object and storage medium |
CN113762252A (en) * | 2017-08-18 | 2021-12-07 | 深圳市道通智能航空技术股份有限公司 | Unmanned aerial vehicle intelligent following target determination method, unmanned aerial vehicle and remote controller |
CN113763224A (en) * | 2020-06-03 | 2021-12-07 | 阿里巴巴集团控股有限公司 | Image processing method and device |
Citations (5)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US4871972A (en) * | 1987-07-13 | 1989-10-03 | Roy Jacques T | Apparatus for detecting faulty power line insulator |
JP2004258003A (en) * | 2003-02-28 | 2004-09-16 | Techno Success Kk | System and method for detecting line-to-ground fault accident of transmission line |
CN103020611A (en) * | 2012-12-30 | 2013-04-03 | 信帧电子技术(北京)有限公司 | Method for detecting fighting behaviors |
CN103714342A (en) * | 2013-12-20 | 2014-04-09 | 华北电力大学(保定) | An aerial-photo insulator chain automatic positioning method based on binary image shape features |
CN104281857A (en) * | 2013-07-08 | 2015-01-14 | 中国科学院沈阳自动化研究所 | Insulator detecting method based on orthogonal direction projection |
-
2016
- 2016-04-27 CN CN201610269864.2A patent/CN105957077B/en active Active
Patent Citations (5)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US4871972A (en) * | 1987-07-13 | 1989-10-03 | Roy Jacques T | Apparatus for detecting faulty power line insulator |
JP2004258003A (en) * | 2003-02-28 | 2004-09-16 | Techno Success Kk | System and method for detecting line-to-ground fault accident of transmission line |
CN103020611A (en) * | 2012-12-30 | 2013-04-03 | 信帧电子技术(北京)有限公司 | Method for detecting fighting behaviors |
CN104281857A (en) * | 2013-07-08 | 2015-01-14 | 中国科学院沈阳自动化研究所 | Insulator detecting method based on orthogonal direction projection |
CN103714342A (en) * | 2013-12-20 | 2014-04-09 | 华北电力大学(保定) | An aerial-photo insulator chain automatic positioning method based on binary image shape features |
Non-Patent Citations (3)
Title |
---|
BOBO DUAN等: "A study on hybrid approaches in image recognition system", 《IEEE INTELLIGENT VEHICLES SYMPOSIUM》 * |
SWADA J等: "A mobile robot for inspectpower transimission linesion of", 《IEEE TRANSACTIONS ON POWER DELIVERY》 * |
谢鑫: "输电线路除冰机器人视觉系统的研究", 《中国优秀硕士学位论文全文数据库 信息科技辑》 * |
Cited By (14)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN113762252B (en) * | 2017-08-18 | 2023-10-24 | 深圳市道通智能航空技术股份有限公司 | Unmanned aerial vehicle intelligent following target determining method, unmanned aerial vehicle and remote controller |
CN113762252A (en) * | 2017-08-18 | 2021-12-07 | 深圳市道通智能航空技术股份有限公司 | Unmanned aerial vehicle intelligent following target determination method, unmanned aerial vehicle and remote controller |
CN107680095A (en) * | 2017-10-25 | 2018-02-09 | 哈尔滨理工大学 | The electric line foreign matter detection of unmanned plane image based on template matches and optical flow method |
CN108710883B (en) * | 2018-06-04 | 2021-08-24 | 国网辽宁省电力有限公司信息通信分公司 | Complete salient object detection method adopting contour detection |
CN108710883A (en) * | 2018-06-04 | 2018-10-26 | 国网辽宁省电力有限公司信息通信分公司 | A kind of complete conspicuousness object detecting method using contour detecting |
CN109145905A (en) * | 2018-08-29 | 2019-01-04 | 河海大学常州校区 | A kind of transmission line of electricity accessory detection method of view-based access control model conspicuousness |
CN110414308A (en) * | 2019-05-16 | 2019-11-05 | 南京理工大学 | A kind of target identification method for dynamic foreign matter on transmission line of electricity |
CN110889328A (en) * | 2019-10-21 | 2020-03-17 | 大唐软件技术股份有限公司 | Method, device, electronic equipment and storage medium for detecting road traffic condition |
CN113743151A (en) * | 2020-05-27 | 2021-12-03 | 顺丰科技有限公司 | Method and device for detecting road surface sprinkled object and storage medium |
CN113763224A (en) * | 2020-06-03 | 2021-12-07 | 阿里巴巴集团控股有限公司 | Image processing method and device |
CN112184816A (en) * | 2020-09-28 | 2021-01-05 | 珠海博明视觉科技有限公司 | Lithium battery flexible positioning method based on battery tabs |
CN112184816B (en) * | 2020-09-28 | 2024-03-22 | 珠海博明视觉科技有限公司 | Lithium battery flexible positioning method based on battery lugs |
CN112991375A (en) * | 2021-02-08 | 2021-06-18 | 上海通办信息服务有限公司 | Method and system for reshaping arbitrary-shaped image area into N rectangular areas |
CN112991375B (en) * | 2021-02-08 | 2024-01-23 | 上海通办信息服务有限公司 | Method and system for remolding image area with arbitrary shape into N rectangular areas |
Also Published As
Publication number | Publication date |
---|---|
CN105957077B (en) | 2019-01-15 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN105957077A (en) | Detection method for foreign body in transmission lines based on visual saliency analysis | |
CN107451965B (en) | Distort face image correcting method, device, computer equipment and storage medium | |
CN109215082B (en) | Camera parameter calibration method, device, equipment and system | |
Py et al. | Plankton classification with deep convolutional neural networks | |
CN104361314B (en) | Based on infrared and transformer localization method and device of visual image fusion | |
CN112950667A (en) | Video annotation method, device, equipment and computer readable storage medium | |
CN109544497A (en) | Image interfusion method and electronic equipment for transmission line faultlocating | |
CN110163193A (en) | Image processing method, device, computer readable storage medium and computer equipment | |
CN101739690B (en) | Method for detecting motion targets by cooperating multi-camera | |
CN110414308B (en) | Target identification method for dynamic foreign matters on power transmission line | |
CN104376575B (en) | A kind of pedestrian counting method and device based on multi-cam monitoring | |
CN110458858A (en) | A kind of detection method of cross drone, system and storage medium | |
CN105957073B (en) | Transmission line of electricity broken lot fault detection method | |
CN105205459B (en) | A kind of recognition methods of characteristics of image vertex type and device | |
CN111462057B (en) | Transmission line glass insulator self-explosion detection method based on deep learning | |
Li et al. | High-level visual features for underwater place recognition | |
CN108710837A (en) | Cigarette smoking recognition methods, device, computer equipment and storage medium | |
Siheng et al. | Power equipment recognition method based on mask R-CNN and bayesian context network | |
CN116245871A (en) | Power transmission line abnormal target detection method based on improved YOLOX algorithm | |
CN116128820A (en) | Pin state identification method based on improved YOLO model | |
CN114445615A (en) | Rotary insulator target detection method based on scale invariant feature pyramid structure | |
CN101561316B (en) | On-line test visual data processing system based on region of interest (ROI) | |
Manninen et al. | Multi-stage deep learning networks for automated assessment of electricity transmission infrastructure using fly-by images | |
CN111292331B (en) | Image processing method and device | |
Lee et al. | FRED: Towards a Full Rotation-Equivariance in Aerial Image Object Detection |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
C06 | Publication | ||
PB01 | Publication | ||
C10 | Entry into substantive examination | ||
SE01 | Entry into force of request for substantive examination | ||
GR01 | Patent grant | ||
GR01 | Patent grant |