CN109711400A - A kind of electric inspection process method and apparatus identifying simulated pointer formula meter reading - Google Patents
A kind of electric inspection process method and apparatus identifying simulated pointer formula meter reading Download PDFInfo
- Publication number
- CN109711400A CN109711400A CN201811314096.3A CN201811314096A CN109711400A CN 109711400 A CN109711400 A CN 109711400A CN 201811314096 A CN201811314096 A CN 201811314096A CN 109711400 A CN109711400 A CN 109711400A
- Authority
- CN
- China
- Prior art keywords
- instrument
- picture
- pointer
- svm
- camera
- 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.)
- Pending
Links
Landscapes
- Image Analysis (AREA)
Abstract
The invention discloses a kind of electric inspection process method and apparatus for identifying simulated pointer formula meter reading, position of the invention including the use of the SVM model prediction meters under test of HOG feature and optimization in the picture, completion takes figure process；Dial plate registration is read using SIFT pointer method for reconstructing.The present invention not only overcomes tired in the allotment the problem of for work upper human and material resources, financial resources of simulation meter reading in substation, and it overcomes camera photo angle and must be perpendicular to the limitation of instrument panel plane and the reading difficult problem of complicated scale instrument or non-uniform scale instrument, simultaneously for the disturbance under varying environment, different illumination conditions, this method also has very strong robustness.
Description
Technical field
The invention belongs to pointer meters fields, more particularly, to a kind of electric power for identifying simulated pointer formula meter reading
Method for inspecting and device.
Background technique
There is outstanding anti-electromagnetic interference capability in view of simulated pointer instrument, simulated pointer instrument obtains in the power system
It is widely applied.Real time monitoring work to instrument is simulated in electric system, its significance lies in that, find that certain parameters are super in time
Mark, and then avoid the security risk and huge loss of economic benefit of entire electric system.Currently, main electric inspection process means
It is manual inspection, not only low efficiency, error rate are high for this mode, it is also necessary to which very high cost goes to cultivate and employ skilled inspection
Worker.Therefore, research and development, which can voluntarily position the device to be measured simulated instrument and can read instrument registration, becomes electricity in the past 20 years
One Main Topics of Force system.Manual inspection is substituted with automatic detecting, with the theory and technical substitution of computer vision
The meter reading process of human eye, harvest is very high working efficiency, while reducing reading error rate, also reduces electric system
Artificial expense, it is ensured that safe operation of power system, and generate bigger economic interests.
Electric inspection process device realizes independent navigation using different types of sensors such as infrared, sound, light, based on different principle
And position, backstage is uploaded to by wireless transmission means in fixed position recording instrument table picture, then by instrument picture, in the background
It using the theory and method of computer vision, completes to read registration work, likewise, the registration recognition methods of simulation instrument is also each
It is different.
For simple circular instrument uniform for most of scales, current common practice is, using edge/gradient/
Colouring information, SIFT (Scale-invariant feature transform) characteristic matching or Hough loop truss obtain instrument
Epitope is set, and the angle of pointer is next obtained using the detection of Hough line, is finally carved in entire instrument board using pointer and starting
The relative angle of line calculates pointer registration.
For the number reading method of the complicated scale simulated pointer table in electric inspection process, be primarily present three limitations: instrument is fixed
Position method lacks universality always；To there are the undercorrection of pointer tilt angle in the case of parallax is rigorous；Still lack efficiently
Reliable method handles the reading work of complicated scale instrument, such as the instrument of scale uneven distribution.
Following two patent content can absolutely prove defect of the existing technology:
The patent of Patent No. CN201410074686.9 discloses a kind of readings of pointer type meters recognition methods and device.
Its main thought is to position instrument dial plate position by color in HSV space, further executes Recognition of Reading process.
This method the problem is that, under the conditions of atrocious weather, weaker illumination be may cause in the photo of shooting
The colouring information of a part is lacked, therefore, such positioning method is inaccurate, shortage robustness.Meanwhile it being read to reduce
Number error, this method require image acquisition equipment perpendicular alignmnet instrument dial plate plane, and being the installation settings of reading plotter hinders
Hinder.
The patent of Patent No. CN201210043415.8 discloses a kind of similar round pointer instrument for mobile robot
Meter reading method.Its main thought is using Canny operator extraction skirt response point, if skirt response point number is greater than some threshold
When value, it is believed that instrument is present in figure, and is gone to be fitted these points with an ellipse, obtains position of the instrument region in picture.
Number reading method is that the angle for obtaining pointer is detected with Hough, and the corresponding registration of the angle is then obtained from database.This method
The problem is that under different shooting conditions, it is very difficult to one adaptive threshold of fitting to adapt to most of situation, because
The skirt response of this Canny operator is the result is that very unstable.Meanwhile this method is not considered to produce the feelings of parallax
Under condition, pointer direction needs the problem of correcting, and can cause biggish error in reading.
Summary of the invention
In view of the drawbacks of the prior art, the purpose of the present invention is to provide a kind of electricity for identifying simulated pointer formula meter reading
Power method for inspecting and device, it is intended to solve the prior art in the positioning of simulation instrument, reading work since precision is not high, accuracy
It is low, adaptability is weak and leads to the problem of being difficult to complicated scale instrument.
To achieve the above object, on the one hand, the present invention provides a kind of electric power for identifying simulated pointer formula meter reading to patrol
Detecting method, the specific steps are as follows:
S1 extracts the HOG feature vector x of positive negative sample from training seti, use yiThe mark of ∈ { -1,1 } mark training sample
Label, respectively indicate negative example background and positive example dial plate；It is trained using the Linear SVM of belt sag variable and obtains svm classifier model.
S2, using the above-mentioned SVM model obtained, the sample of classification error is added to training by training test set sample
Collect re -training, advanced optimizes SVM model.
S3, coarse adjustment holder: according to the meters under test spatial information recorded, robot reaches the fixation in front of meters under test
Position completes to take figure D1 for the first time under the original focal length of camera.
S4, using Selective Search candidate frame policy selection area to be tested, in area to be tested, according to instruction
Practice the certain interval of the size selection of collection picture, HOG feature extraction is carried out to all rectangular window images of generation, prepares to use
SVM model adjudicates rectangular window image.
S5 carries out the judgement of SVM model to all rectangular window figures of above-mentioned generation, when SVM predicted value is greater than 0, represents pair
The rectangle frame answered is instrument；When SVM predicted value is less than 0, corresponding rectangle frame is represented as background.
If being judged as the number of instrument rectangle frame there are multiple, non-maxima suppression and Bounding are further executed
Box regression process fits unique rectangle frame, with the centre coordinate P of rectangle framec(xc,yc) characterize instrument in the picture
Physical location；
If all rectangle frames are judged as background, S3 is gone to, readjusts the position of holder, repeats S4 and S5,
Rectangle frame until being judged as instrument occurs.
S6, fine-adjustment tripod head: the instrument center P detected according to camera pixel imaging center P (x, y) and S5c(xc,yc), meter
It calculates on horizontal, vertical direction, Pc(xc,yc) offset relative to P (x, y).By means of transformation coefficient t, second of adjustment cloud
Platform makes instrument practical center Pc(xc,yc) and the imaging center P (x, y) of camera between offset minimize；
S7, optical amplifier, auto-focusing obtain clear scales ground dial plate picture D2 after amplification；
S8 carries out planar reconstruction to gauge pointer using SIFT:
Using SIFT thought, respectively to the master meter obtained under amplified instrument picture D2 and existing positive angle
Dial plate plane picture D3 calculates key point coordinate and principal direction.To all key points, successively on rotatable coordinate axis to principal direction, meter
Calculate SIFT description.Euclidean distance metric is used between instrument picture D2 and master meter dial plate plane picture D3, sets Europe
The threshold value of family name's distance carries out violence matching；Using RANSAC iterative algorithm, reasonable Optimum Matching point pair is filtered out, fixation is taken
The matching double points of quantity use least square method, fit amplified instrument picture D2 to master meter dial plate plane picture D3
Perspective transform relationship T1。
Using the position in Hough line testing principle positioning pointer after amplification instrument picture D2, and utilize perspective transform
Relationship T1Pointer on instrument picture D2 is reconstituted on master meter dial plate plane picture D3.
S9 seeks pixel rotation center of the intersection point of pointer and mid-scale as pointer on master meter picture D3,
And polar coordinate transform and pixel thinning algorithm is carried out to master meter picture D3, curved scale band Qu Weizhi is extracted into bone
Frame obtains instrument picture D4, further to read；
S10 counts the pixel number on instrument picture D4 by column, by the column pixel number different types of graduation mark of differentiation with
And position of the pointer in image D4, the method that simulation people reads simulation instrument are closed by pointer to the position of related graduation mark
Final reading is calculated as a result, completing primary identification in system；
S11, resets the focal length of camera, repeats S3~S10, successively handles all meters under test, until identification all to
Survey instrument.
On the other hand, the present invention provides a kind of electric inspection process robots for identifying simulated pointer formula meter reading, including
Driving wheel two, servo direct current motor two, aluminium alloy protection shell, one, programmable network camera, camera axial direction stepping electricity
One, machine, camera radial stepping motor one, holder shell, laser radar one, one piece of 12V rechargeable lithium battery, Nvidia
One piece of master control borad, one piece of ARM microcontroller core board and related Voltage stabilizing module.
Preferably, the tailstock is general-purpose interface area, comprising: booting/reboot button, the network interface of Nvidia master control borad be several, USB
Interface is several, RS232 interface is several and other general-purpose interfaces.
Preferably, two driving wheels are respectively connected with two servo direct current motors, are generated by between two servo motors
Differential realize turning；
Preferably, two servo motors and two holder stepper motors are controlled by ARM microcontroller core board, this need according to
Rely in the information exchange of Nvidia master control borad and ARM microcontroller core board, communication mode is serial communication；
Preferably, programmable network camera, laser radar, Nvidia master control borad and ARM microcontroller core board are respectively by 12V
Rechargeable lithium battery provides burning voltage after corresponding Voltage stabilizing module decompression processing；
Preferably, master controller is Nvidia master control borad, executes all main implementation procedures.Nvidia master control borad and programmable
Communication mode between network cameras is the communication of RJ45 network interface；Communication mode between Nvidia master control borad and laser radar is also
The communication of RJ45 network interface；
Programmable network camera main function is the external image obtained for identification；
Laser radar provides hardware supported for the navigation of electric inspection process robot.
Contemplated above technical scheme through the invention, compared with prior art, since the present invention uses HOG-SVM mesh
Mark detection technique and SIFT plane pointer method for reconstructing, can obtain it is following the utility model has the advantages that
(1) present invention uses HOG-SVM target detection technique, can be in inspection meter reading, fast resolution gauge field
Domain and background area are accurately positioned the position of instrument in image, effectively overcome poor anti jamming capability in instrument position fixing process
The shortcomings that.
(2) present invention using SIFT to gauge pointer carry out planar reconstruction, be shooting process instrument board generation parallax into
It has gone correction, has additionally provided the technical solution of a set of simulated pointer meter reading, made to read that result is more accurate, and reading process is more
Add reliable, stabilization, has good robustness.
Detailed description of the invention
Fig. 1 is the electric inspection process flow chart of identification meter reading provided by the invention；
Fig. 2 is the template of calibration for cameras optical center provided by the invention；
Fig. 3 is the imaging model of holder camera provided by the invention；
Fig. 4 is the regulator dial plate picture provided by the invention after parallax correction；
Fig. 5 is the extraction result of instrument board graduation mark and pointer provided by the invention；
Fig. 6 is the principal diagram of polar coordinate transform process provided by the invention；
Fig. 7 is the exterior structure of crusing robot provided by the invention.
Specific embodiment
In order to make the objectives, technical solutions, and advantages of the present invention clearer, with reference to the accompanying drawings and embodiments, right
The present invention is further elaborated.It should be appreciated that described herein, specific examples are only used to explain the present invention, not
For limiting the present invention.
Fig. 1 is the electric inspection process provided by the invention rebuild based on the detection of HOG-SVM target classification and SIFT plane pointer
The recognition methods of robot simulation's meter reading, specifically includes the following steps:
S1, preparation: (1) the location of pixels P (x, y) of calibration for cameras imaging center；(2) it chooses and measures a certain material object
Physical length H, in the case where constant and camera original focal length is constant at a distance from meters under test for camera, acquire pictorial diagram
Piece demarcates length transformation ratio between the two according to length in pixels cnt of the material object in picture and the actual length H of the material object
t。
S2 extracts HOG (Histogram of Gradient) feature vector x of positive negative sample from training seti, use yi∈
{ -1,1 } label for marking training sample, respectively indicates negative example background and positive example dial plate；Using the Linear SVM of belt sag variable
(Support Vector Machine) is trained, preliminary to obtain svm classifier model.
S3, using the above-mentioned SVM model obtained, the sample of classification error is added to training by validation test collection sample
Collect re -training, advanced optimizes SVM model.
S4, according to the meters under test spatial information recorded, robot is reached solid in front of meters under test using airmanship
Set a distance is completed to take figure D1 for the first time under the original focal length of camera.
S5, using Selective Search candidate frame policy selection area to be tested, in area to be tested, according to instruction
The size for practicing collection picture, selects certain interval, carries out HOG feature extraction to all rectangular window images of generation, prepares to utilize
SVM model adjudicates rectangular window image.
S6 carries out the judgement of SVM model to all rectangular window figures of above-mentioned generation, when SVM predicted value is greater than 0, represents pair
The rectangle frame answered is instrument；When SVM predicted value is less than 0, corresponding rectangle frame is represented as background.
If being judged as instrument rectangle frame number, there are multiple, further execution non-maxima suppression and Bounding
Box regression process fits unique rectangle frame, with rectangle frame center Pc(xc,yc) characterize the actual bit of instrument in the picture
It sets；
If all rectangle frames are judged as background, S4 is gone to, readjusts the position of holder, repeats S5 and S6,
Rectangle frame until being judged as instrument occurs.
S7, according to the center P of the imaging center location of pixels P (x, y) of camera and the S5 actual instrument detectedc(xc,
yc), it can calculate on horizontal, vertical direction, Pc(xc,yc) offset relative to P (x, y).By means of transformation coefficient t, second
Secondary adjustment holder makes the imaging center location of pixels P (x, y) of camera and the center P of actual instrumentc(xc,yc) between offset most
Smallization；
S8 amplifies camera optics, and auto-focusing obtains the dial plate picture D2 of clear scales after amplification；
S9 carries out planar reconstruction to gauge pointer using SIFT:
Using SIFT thought, respectively to the master meter plane picture D3 under amplified instrument picture D2 and positive angle
Key point coordinate and principal direction are calculated, to all key points, successively on rotatable coordinate axis to its principal direction, calculates SIFT description
Son.And between picture D2 and D3, using euclidean distance metric, given threshold carries out violence matching；It is calculated using RANSAC iteration
Method filters out reasonable Optimum Matching point pair, the matching double points of fixed quantity is taken, using least square method, after fitting amplification
Instrument picture D2 to master meter dial plate plane picture D3 perspective transform relationship T1。
It is closed using the position in Hough transform principle positioning pointer after amplification instrument picture D3, and using perspective transform
It is T1Pointer on amplified instrument picture D2 is reconstituted on master meter dial plate plane picture D3.
S10, the intersection point that pointer and mid-scale are sought on master meter dial plate picture D3 are rotated as the pixel of pointer
Center, and polar coordinate transform and pixel thinning algorithm are carried out to master meter dial plate picture D3, by curved scale band Qu Weizhi,
Skeleton is extracted, instrument picture D4 is obtained, further to read；
S11, statistical pixel is counted by column, distinguishes different types of graduation mark and pointer in image D4 by column pixel number
In position, simulation people read simulation instrument method, by the positional relationship of pointer and related graduation mark, be calculated finally
Reading is as a result, complete primary identification；
S12 resets the focal length of camera, repeats S5~S12, successively handles the meters under test of all acquisitions, until identification is complete
Portion's meters under test.
The scaling method of the location of pixels P (x, y) at camera imaging center is as follows in the step S1:
As shown in Fig. 2, calibration process requires the distance between scaling board and camera constant and does not change the position of camera.
When calibrating template is when closer apart from camera, since scaling board background is black, and intermediate filled circles are white,
Center location and radius of the white filled circles in Fig. 2 can be quickly and accurately found using Hough circle detection method, and
By pinhole camera model it is known that the zoom center of camera is just located at the imaging center P (x, y) of camera, therefore:
Under identical shooting distance, different imaging magnifications, it is assumed that the white filled circles that certain detection process detects
The center of circle is the P in Fig. 21(x1,y1), radius R1；Another secondary white filled circles center of circle detected is the P in Fig. 22(x2,y2),
Radius is R2, then there is following relationship:
It is possible thereby to acquire the coordinate of P (x, y).For the sake of accurate, the above process is executed repeatedly, the side averaged is taken
Formula obtains the optimal imaging center pixel coordinate P (x, y) of camera.
In the step S1 keep camera between meters under test at a distance from constant and situation that camera original focal length is constant
Under, it is as follows to the scaling method of the transformation ratio t of the length in pixels of picture in kind in camera to material object actual length:
According to the two-dimension vector map containing all meters under test location parameters constructed, by the position coordinates of robot
It is denoted as C (xR,yR).Two-dimensional coordinate R (x' of certain meters under test in map is knownR,yR') and setting robot with it is to be measured
The distance between instrument D.
First find the pillar of a non-reflective material upright, black, being of moderate size, a height of H.Adjust holder pitching
Angle appears in whole pillar can completely in the image of shooting.Assuming that the mean pixel length warp of this root pillar in the picture
Crossing statistics is cnt, then equation: the objective establishment of H=t*cnt.
And problem is that there are position error, C (x for navigation procedureR,yR) and R (x'R,y'R) actual range reason between two o'clock
By can above float at random within the scope of some in the positive and negative of D.The way for copying calibration imaging center, if C (xR,yR) and R (x'R,y
'R) the distance between two o'clock differs too big with D, then gives up this transformation coefficient t obtained by calibrating, it is on the contrary then retain.Finally, more
Averaged after secondary calibration, it is believed that t is objective reality value.
The HOG feature acquisition modes of training set picture described in step S2 are as follows:
(1) it cuts m from the existing shooting picture library of camera and opens and open the picture for including background comprising complete instrument and n, this m is a
Positive example is consistent with n negative example sizes, and the ratio in training set is about 1:10, and marks positive example and negative example respectively with 1 and -1,
It is hereby achieved that a scene training set abundant；
(2) gray processing processing is carried out to the training data image of all fixed sizes, and executes Gamma correction, adjustment pair
Degree of ratio；
(3) point seeks gradient in the horizontal and vertical directions pixel-by-pixel；
(4) rectangle HOG description block is used, HOG feature extraction is carried out.Each cell element is 8 ﹡, 8 pixels, and a block is by 2 ﹡
2 cell element compositions, the interval that block slides in the transverse and longitudinal coordinate of picture is 8 pixels；
(5) in each cell element, based on the gradient direction of undirected histogram and pixel, by nine sections to each born of the same parents
All pixels in member click through column hisgram statistics, the weight of the amplitude vote by proxy of each pixel in cell element；Count one
In a block after the histogram of gradients of all cell elements, corresponding 9 dimensional vectors of each cell element are successively unfolded in order；
(6) possess the sliding window of identical size with block, after the image for traversing whole picture HOG feature to be extracted, obtain a height
The vector of dimension obtains the HOG feature vector of positive negative sample.
The greyscale transformation formula of image in above-mentioned training set are as follows:
Gray=IR*0.299+IG*0.587+IB*0.114
Wherein Gray represents the gray value of some pixel, IR、IG、IBRespectively represent the corresponding pixel of color image
The value in three channels.
Gamma corrects transformation for mula are as follows:
Pout=(Pin)1/gamma
Wherein PoutFor value of certain pixel after Gamma correction, PinFor the pixel initial value, gamma is correction coefficient,
Gamma often takes 0.5.
The following formula of the gradient magnitude M and gradient direction θ of pixel (i, j) obtains:
Gx(i, j)=I (i+1, j)-I (i-1, j)
Gy(i, j)=I (i, j+1)-I (i, j-1)
θ=arctan (Gy/Gx)
Wherein Gx(i, j) and Gy(i, j) is the gradient of pixel (i, j) horizontal direction and vertical direction, and I (i, j) is point
(i, j) corresponding pixel value.
Svm classifier model described in step S2 is obtained by using the HOG feature vector training of training set picture, specifically
Acquisition methods are as follows:
(1) form of svm classifier function is belt sag variable, and kernel function is the soft svm classifier function f (x) of linear kernel；
(2) the multidimensional HOG feature vector for obtaining K training set image zooming-out is denoted as D as the training data of SVMs=
{(x1,y1),...,(xL,yL)},yi∈ { -1,1 } wherein DsIndicate training sample set, yi∈ { -1,1 } indicates training sample set
The corresponding label of middle feature vector；
(3) kernel function of svm classifier function is linear kernel, and selected substitution loss function simultaneously initializes relevant parameter, will be asked
The SVM problem of solution largest interval classifying face is converted into lagrange duality problem, by Karash-Kuhn-Tucker (KKT) item
Part solves optimal svm classifier function using SMO (Sequential Minimal Optimization) algorithm iteration.
Svm classifier function formula are as follows:
ysv=1
Wherein sgn is sign function, and b is the displacement item of classification function, αiIt is the optimal Lagrange of classifying face bound term
Multiplier vector α=(α1,...,αK) one-component, whether what it was represented is constrained effective to some specific sample.αi=0
When indicate invalid constraint；The effective constraint of on the contrary then expression, while corresponding (xsv,ysv) sample, i.e. supporting vector (SV).
The general description of the soft support vector machines of belt sag variable are as follows:
s.t.yi(ωTxi+b)≥1-εi, i=1,2,3 ..., K
Wherein C is the penalty factor of slack variable, usually takes 0.01；εiFor slack variable, substitutes and damage shown herein as hinge
Lose function: εhinge(z)=max (0,1-z)；μiIndicate the Lagrange multiplier of slack variable.
Belt sag variable, the dual problem for solving α are described as follows:
0≤αi≤ C, i=0,1,2 ..., K
WhereinFor xiTransposition, s.t. expression submit to constrain.Due to using inequality constraints, KKT in constraint
Condition are as follows:
The SMO algorithm for solving α, is described in detail below:
The arbitrarily selected two bound term factor-alphas of each round iterative processiAnd αj, initialization other parameters (including other are all
The constraint factor of item), solve the α of epicycleiAnd αj.That is, each round iteration only updates two parameters, through excessively taking turns iteration,
It goes to approach optimal SVM division hyperplane.
Based on Selective Search described in step S5, the method for extraction apparatus entry mark candidate frame is as follows:
(1) in Felzenszwalb connected area segmentation result images D1, a series of similar connected domains of characterization are tentatively extracted
The characteristics of rectangle frame, gained rectangle frame correspondence image region is: all areas top left pixel point has identical with bottom right pixel point
Pixel value；
(2) the multiple frames of similarity combination are overlapped according to color, texture, size and space；
(3) rectangle frame after above-mentioned merging can characterize to a certain extent, if block of pixels and surrounding that rectangle frame occurs are carried on the back
There are notable differences in contrast and gradient for scape, it is generally recognized that in the presence of the meters under test for needing to detect in these rectangle frames.
During the merging rectangle frame, similarity measurement formula is as follows:
s(ri,rj)=α1scolor(ri,rj)+α2stexture(ri,rj)+α3ssize(ri,rj)+α4sfill(ri,rj)
Wherein these regions are denoted as a set R={ r1,…,rn}；scolorFor color similarity, stextureFor texture phase
Like property, ssizeFor size similitude, sfillSpace overlaps similitude, and s is comprehensive similitude；α1、α2、α3And α4Respectively color phase
The constant factor that similitude is overlapped like property, texture paging, size similitude and space indicates the weight of each similitude, and needs
Meet α1+α2+α3+α4=1.s(ri,rj) value is bigger, just represent riAnd rjIt is more similar.Given threshold TsIf s (ri,rj) > Ts, then
Merge；Conversely, then nonjoinder.TsGenerally take 0.5.
In the step S7, the foundation of second of adjustment holder is as follows:
(1) such as Fig. 3, imaging plane coordinate center O in imaging center coordinate P (x, y) corresponding diagram 3 of cameraI；Target detection
The instrument center P that process detectsc(xc,yc) P in corresponding diagram 3I, instrument center pixel PIRelative to imaging center OILevel
Biasing and vertical biasing respectively correspond the x in Fig. 3IAnd yI, holder is further adjusted, instrument center P is madeIWith imaging center OIMore
It is close；
After first time adjusts, instrument center PIWith imaging center OIThere are still the reason of obvious deviation to be: this is machine
The series of factors such as the position error of device people, the position error of holder stepper motor and rounding error when calculating are comprehensive
Caused by；
(2) according to object detection results, instrument pixel center P is calculated separatelyIRelative to imaging center pixel coordinate OITwo
Deviation x on a directionIAnd yI, using transformation ratio t respectively by deviation xIAnd yIIt is scaled true instrument center and camera imaging
Horizontal departure Δ x and vertical deflection Δ y between center；
(3) the camera imaging model based on Fig. 3 calculates holder in angle, θ that is horizontal, needing to adjust on vertical directionh' and
θv'。
(4) by stepper motor, holder is made to rotate θ in horizontal and vertical directionh' and θv' angle.Complete second of holder
Adjustment.
Holder adjusts angle, θ on horizontal, vertical directionh' and θv' calculation formula it is as follows:
Δ x=txI
Δ y=tyI
The step S9, instrument picture D2 and regulator dial plate plane D3 carry out the matched method of SIFT such as:
(1) the master meter dial plate plane picture D3 under amplified instrument picture D2 and positive angle is established respectively high
This scale pyramid.Pyramid is made of three sub- octaves (Octave), and every sub- octave includes 6 original images in different Gauss rulers
The lower filtering image of degree, and the size relationship between sub- octave is, upper layer for the length and width of lower layer be 1/2 it is down-sampled；Gauss
In pyramid, from bottom to top layer, Gauss scale parameter is continuous；
(2) the every difference of Gaussian (DoG) of sub- octave between layers is calculated, Gaussian difference scale space is constituted；
(3) in difference of Gaussian domain, all pixels point under each scale is once traversed, if a certain pixel is in its 8 neighbour
It is extreme point, then it is assumed that the pixel is corresponding Gauss in domain and the layering of upper and lower two Gaussian differences in corresponding 3 region 3 ﹡
A characteristic point under scale.
(4) all pixels point under the space DoG is traversed one by one, extracts all characteristic points；And it is bent to screen out some parts
Rate is lower or gradient responds weaker pixel；
(5) in a certain size the region centered on each characteristic point, under corresponding Gauss scale, node-by-node algorithm
The gradient magnitude and gradient direction of all pixels point carry out omnidirectional's statistics with histogram in 8 sections, and ballot weight is corresponding pixel points
Gradient magnitude；In histogram, principal direction of the most angle of poll as this feature point；
(6) it is being with it by the image rotation to its principal direction under its corresponding Gauss scale for some characteristic point
In each of 16 ﹡, 16 window at center 4 ﹡, 4 seed point, using the method similar with (5), 8 sections weighted with Gauss are calculated
Omnidirectional's histogram；The SIFT feature description vectors of one 128 dimension are formed altogether；
(7) violence matches: successively calculating the corresponding description subcharacter vector of all characteristic points and standard in instrument picture D2
The corresponding Euclidean distance for describing subcharacter vector of all characteristic points, sets the threshold value of Euclidean distance in instrument board picture D3
TSIFT1If the Euclidean distance being calculated is less than TSIFT1, then it is assumed that corresponding Feature Points Matching success in figure；Conversely, then matching
Failure；
(8) RANSAC is screened: i.e. Random Sample Consensus, it is the sample according to one group comprising abnormal data
Notebook data collection iterates to calculate out the mathematical model parameter of data, obtains the algorithm of effective sample data.In single iteration, with
The match point matched from above-mentioned violence to machine, which is concentrated, extracts 4 pairs of match points, calculates a homography matrix；It is answered using the list
Property matrix calculate under the mathematical model, all match points on instrument picture D2 are transformed on regulator dial plate picture D3, and
Calculate the Euclidean distance on instrument picture D2 on match point and regulator dial plate picture D3 between match point.Given threshold TSIFT2,
If the Euclidean distance of true match point and fitting match point is greater than TSIFT2, then it is assumed that error hiding is produced, is rejected；It is on the contrary then
Retain.
The gaussian filtering formula of Gaussian scale-space are as follows:
L (x, y, σ)=G (x, y, σ) * I (x, y)
Wherein I (x, y) indicates that the single pixel gray value in the image of feature to be extracted, L (x, y, σ) indicate Gauss scale
Pixel under space, G (x, y, σ) indicate gaussian kernel function, and σ is Gauss scale factor, when scale factor value is bigger, Gauss
The result images of filtering are fuzzyyer.
The mathematical form of difference of Gaussian are as follows:
D (x, y, σ)=[G (x, y, k σ)-G (x, y, σ)] * I (x, y)=L (x, y, k σ)-L (x, y, σ)
Wherein k is Gauss scale coefficient, characterizes the size of Gauss scale.
During calculating characteristic point principal direction, the calculation formula of pixel gradient magnitude and gradient direction in neighborhood are as follows:
L (x, y, σ ')=G (x, y, σ ') * I (x, y)
Wherein, DoG scale locating for certain characteristic point is σ ', then centered on this feature point, and radius is the area of 3 × 1.5 σ '
In domain, the amplitude and argument of pixel L (x, y, σ ') is expressed as M (x, y) and θ (x, y).
During RANSAC, the corresponding homography conversion matrix T of iteration each timelrAre as follows:
Wherein a11To a33For homography matrix TlrTransformation coefficient, and a33=1, TlrFreedom degree is 8.
The Corresponding matching point of instrument picture and regulator dial plate picture is to may be expressed as:
Wherein [xl yl 1]T[xr yr 1]TRespectively a pair of of match point exists in instrument picture and regulator dial plate picture
Homogeneous coordinates in respective coordinate system.
In the step S10, the method that converts slide-rule dial band for curved scale band are as follows:
Fig. 4 provides a kind of simulated pointer dial plate, which has the curved scale line of uneven distribution.As in SIFT
Regulator dial plate in pointer reconstruction process, if without serious matching error in SIFT pointer reconstruction process, it is believed that calculate
To pointer rotation center be randomly distributed near true rotation center.Based on the above thought, in different multiple identifications
The dial plate picture D2 that the different registrations shot under multiple camera original focal lengths are acquired in period, by pointer in regulator dial plate D3
Reconstruction process after, successively find out the intersection point of pointer Yu dial mid-scale；The transverse and longitudinal of the multiple rotation centers found out is sat
Mark successively seeks arithmetic mean of instantaneous value, as objective pointer pixel rotation center；
According to rotation center, suitable scale size is selected, by the picture on the regulator dial plate picture D3 for having rebuild pointer
Vegetarian refreshments successively transforms in polar coordinate system from eulerian coordinate system；
Pixel thinning algorithm is executed, the skeleton of different type scale and pointer, convenient further identification, such as Fig. 5 are extracted
It is shown.
The mathematical formulae of polar coordinate transform are as follows:
ρ=(x-xa)·sinθ+(y-ya)·cosθ
Wherein, as shown in fig. 6, left figure is signal of the regulator dial plate picture D3 in x-y image coordinate system, Pa(xa,ya)
The pointer rotation center of regulator dial plate picture D3 is indicated, according to the actual situation, when pointer is directed toward certain root on circular arc scale band
When graduation mark, pointer direction is generally overlapped with graduation mark direction, zero graduation line and the corresponding pointer pole of maximum range graduation mark
Limit pivot angle is θ1And θ2；Right figure is with Pa(xa,ya) centered on, left figure transforms to the result signal of ρ-θ coordinate system, all scales
Line and pointer are generally perpendicular to θ axis, and minimum, the corresponding θ value of maximum scale is also about exactly θ1And θ2。
The step S11, the number reading method of instrument dial are as follows:
(1) to the instrument board polar coordinate transform-refinement for having rebuild pointer as a result, by column statistical pixel point, according to different lines
The black pixel point number of accumulation, extracts the position where different types of graduation mark and pointer；Among these, where pointer
Column have most accumulation numbers.
(2) as shown in figure 5, there are three types of scales in total.Wherein longest is the graduation mark for identifying whole scale, is denoted as 1 type, right
The registration value answered is 0.2；The graduation mark of the followed by slightly short whole scale of mark half, is denoted as 2 types, and 2 type graduation marks are located at adjacent two
Between 1 type graduation mark of item, corresponding registration value is 0.1；It is finally shortest unit scales line, is denoted as 3 types, corresponding registration value
It is 0.02.
(3) a point situation is identified, as shown in Figure 4 and Figure 5, is distributed on more complicated instrument board in this scale, when
Registration be greater than 1.0 and registration less than 1.0 when, it is uneven due to dial plate, number reading method be it is different, need point situation to carry out
Identification；Meanwhile the case where being located between two graduation marks for pointer, it should position according to pointer and two neighboring scale
The position of line, is calculated.The calculation formula of instrument dial registration are as follows:
C=C1+C2
Wherein C is final reading as a result, C1And C2Respectively skim result and precision result；SpntFor where pointer in Fig. 5
Columns, S1.0、S2.0And S5.0For 1.0 graduation marks, 2.0 graduation marks and the corresponding columns of 5.0 graduation marks in Fig. 5, A is to work as Spnt≤
S1.0When skimming reading result；O, p, q respectively represent S in Fig. 5pntArrange 1 type of (pointer position) left side, 2 types, 3 type graduation marks
Item number, γ1、γ2、γ3It respectively corresponds and is equal to 0.2,0.1,0.02；γ ' is the unit scales value for reading carefully and thoroughly part, Spnt1For Fig. 5
SpntArrange the columns where the nearest graduation mark in (pointer position) left side, Spnt2For Fig. 5 Spnt(pointer institute is in place for column
Set) columns where the nearest graduation mark in the right.
Fig. 7 is the structure chart of electric inspection process robot provided in this embodiment, as shown in Figure 7 the identification meter reading
Crusing robot, including two driving wheel two, servo direct current motor aluminium alloys protect shells, programmable network camera one
Platform, camera axial direction stepper motor one, camera radial stepping motor one, holder shell, laser radar one, 12V are chargeable
One piece of lithium battery, one piece of Nvidia master control borad, one piece of ARM microcontroller core board and related Voltage stabilizing module.
Preferably, the tailstock is general-purpose interface area, comprising: booting/reboot button, the network interface of Nvidia master control borad be several, USB
Interface is several, RS232 interface is several and other general-purpose interfaces.
Preferably, two driving wheels 1 are respectively connected with two servo direct current motors, are generated by between two servo motors
Differential realize turning；
Preferably, two servo motors and two holder stepper motors are controlled by ARM microcontroller core board, this need according to
Rely in the information exchange of Nvidia master control borad and ARM microcontroller core board, communication mode is serial communication；
Preferably, programmable network camera, laser radar, Nvidia master control borad and ARM microcontroller core board are respectively by 12V
Rechargeable lithium battery provides burning voltage after corresponding Voltage stabilizing module decompression processing；
Preferably, master controller is Nvidia master control borad, executes all main implementation procedures.Nvidia master control borad and programmable
Communication mode between network cameras is the communication of RJ45 network interface；Communication mode between Nvidia master control borad and laser radar is also
The communication of RJ45 network interface；
Programmable network camera main function is the external image obtained for identification；
Laser radar builds figure for SLAM and navigation provides hardware supported.
In conclusion the present invention solves universal simulation instrument localization method and holds by HOG-SVM target detection technique
The shortcomings that vulnerable to illumination, background interference, and the precision of target detection is very high；By SIFT plane pointer method for reconstructing, effectively
And bring parallax under tilt angle is accurately eliminated, keep reading result more accurate；By polar coordinate transform method,
In conjunction with pixel thinning algorithm, curved scale is transformed to slide-rule dial.Realize the curved scale mould of graduation mark uneven distribution
The accurate reading of quasi- instrument registration.
As it will be easily appreciated by one skilled in the art that the foregoing is merely illustrative of the preferred embodiments of the present invention, not to
The limitation present invention, any modifications, equivalent substitutions and improvements made within the spirit and principles of the present invention should all include
Within protection scope of the present invention.
Claims (8)
1. a kind of electric inspection process method for identifying simulated pointer formula meter reading characterized by comprising
(1) by extracting the HOG feature of picture in training set and test set, optimal svm classifier model is constructed；
(2) extract meters under test candidate rectangle frame, and to all candidate rectangle frames carry out HOG feature extraction after, use is above-mentioned
SVM model identifies instrument rectangle frame, and then obtains the instrument picture D2 of amplified clear scales；
(3) pointer position on instrument picture D2 is reconstituted on master meter plane picture D3 using SIFT method, identification to
Survey the reading of instrument.
2. the method according to claim 1, wherein above-mentioned steps (1) comprise the following specific steps that:
(1.1) the HOG feature vector x of positive negative sample is extracted from training seti, use yiThe mark of ∈ { -1,1 } mark training sample
Label: -1 indicates the negative example of background, and 1 indicates dial plate positive example；
(1.2) sample in training set is trained using the Linear SVM of belt sag variable, obtains initial svm classifier mould
Type；
(1.3) using the svm classifier model testing test set sample of above-mentioned acquisition, the sample of classification error in test set is added
To training set, the SVM model of final optimization pass is obtained.
3. method according to claim 1 or 2, which is characterized in that above-mentioned steps (2) include step specific as follows:
(2.1) according to existing meters under test spatial information, the angle of imaging device is adjusted, obtains the acquisition of first picture D1；
(2.2) area to be tested is selected in picture D1 using Selective Search method；
(2.3) HOG feature extraction is carried out to all rectangular window images for including in above-mentioned area to be tested and SVM category of model is sentenced
Certainly, until the rectangle frame for being judged as instrument occurs；
(2.4) imaging device is finely tuned again, makes the meters under test center P in above-mentioned instrument rectangle framec(xc,yc) with camera at
After offset between inconocenter P (x, y) minimizes, the instrument picture D2 of focused acquisition amplification.
4. method as claimed in claim 3, which is characterized in that the SVM category of model adjudicates mode are as follows:
(2.3.1) is greater than 0 when SVM predicted value, and corresponding rectangular window is instrument rectangle frame；It is corresponding when SVM predicted value is less than 0
Rectangular window is background rectangle frame；
When (2.3.2) judgement instrument rectangle frame number is not unique, non-maxima suppression and Bounding Box regression process are executed,
Fit unique instrument rectangle frame；When judgement instrument rectangle frame number is 0, return step (2.1), until ruling out instrument square
Shape frame.
5. method as described in claim 1 or 4, which is characterized in that the step (3) includes step specific as follows:
(3.1) SIFT thought is utilized, key point coordinate and the master of the instrument picture D2 and master meter plane picture D3 are calculated
Direction obtains SIFT description；
(3.2) violence matching is carried out to above-mentioned instrument picture D2 and master meter plane picture D3 according to SIFT description；
(3.3) RANSAC iterative algorithm is used, optimal between instrument picture D2 and master meter plane picture D3 is filtered out
With point pair, it is fitted the perspective transform relationship T of amplified instrument picture D2 to master meter dial plate plane picture D31；
(3.4) using the position in Hough line testing principle positioning pointer after amplification instrument picture D2, and pointer is reconstituted in
On master meter dial plate plane picture D3；
(3.5) polar coordinate transform and pixel thinning algorithm is carried out to dial plate picture D3 to extract curved scale band Qu Weizhi
Skeleton obtains instrument picture D4, further to read；
(3.6) different types of graduation mark and pointer are distinguished in instrument picture D4 by the column pixel number of above-mentioned instrument picture D4
In position, calculate meter reading；
(3.7) imaging device is reset, step (2.1)~(3.6) are repeated, until identifying whole meters under test.
6. a kind of electric inspection process device for identifying simulated pointer formula meter reading characterized by comprising power supply, programmable net
Network camera, camera stepper motor, laser radar, Nvidia master control borad and ARM microcontroller core board；
Laser radar (6), programmable network camera (4) and the ARM microcontroller core board and Nvidia master control borad information are handed over
Mutually；
The Nvidia master control borad is master controller, for executing all main implementation procedures；
The programmable network camera (4) is for obtaining external image to be identified；
The laser radar (6) provides hardware supported to build figure and navigation；
The ARM microcontroller core board is for controlling camera stepper motor；
The power supply is used for programmable network camera (4), laser radar (6), Nvidia master control borad and ARM microcontroller core
Plate provides burning voltage.
7. device as claimed in claim 6, which is characterized in that
Communication mode between the programming networks camera (4) and laser radar (6) and Nvidia master control borad is RJ45 network interface
Communication；
Communication mode between the ARM microcontroller core board and Nvidia master control borad is serial communication.
8. device as claimed in claims 6 or 7, which is characterized in that the device further includes Voltage stabilizing module, network interface and switch
Button.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201811314096.3A CN109711400A (en) | 2018-11-06 | 2018-11-06 | A kind of electric inspection process method and apparatus identifying simulated pointer formula meter reading |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201811314096.3A CN109711400A (en) | 2018-11-06 | 2018-11-06 | A kind of electric inspection process method and apparatus identifying simulated pointer formula meter reading |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN109711400A true CN109711400A (en) | 2019-05-03 |
Family
ID=66254863
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201811314096.3A Pending CN109711400A (en) | 2018-11-06 | 2018-11-06 | A kind of electric inspection process method and apparatus identifying simulated pointer formula meter reading |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN109711400A (en) |
Cited By (7)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110084842A (en) * | 2019-05-05 | 2019-08-02 | 广东电网有限责任公司 | A kind of secondary alignment methods of machine user tripod head servo and device |
| CN110909738A (en) * | 2019-11-15 | 2020-03-24 | 杭州远鉴信息科技有限公司 | Automatic reading method of pointer instrument based on key point detection |
| CN112344980A (en) * | 2019-08-09 | 2021-02-09 | 浙江中控技术股份有限公司 | Full-automatic verification system and method for pointer instrument |
| CN113077518A (en) * | 2021-03-15 | 2021-07-06 | 中移(上海)信息通信科技有限公司 | Camera parameter calibration method, device and storage medium |
| CN113536895A (en) * | 2021-05-28 | 2021-10-22 | 立得空间信息技术股份有限公司 | Disc pointer meter identification method |
| CN113610094A (en) * | 2021-08-27 | 2021-11-05 | 四川中电启明星信息技术有限公司 | Distribution room pointer instrument reading method based on rotation projection calibration |
| CN116343232A (en) * | 2023-04-03 | 2023-06-27 | 内蒙古师范大学 | Ancient book mathematical symbol recognition method based on pre-classification |
Citations (8)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN103927507A (en) * | 2013-01-12 | 2014-07-16 | 山东鲁能智能技术有限公司 | Improved multi-instrument reading identification method of transformer station inspection robot |
| CN105045262A (en) * | 2014-11-27 | 2015-11-11 | 马金环 | Inspection robot in transformer substation able to fulfill automatic meter reading |
| CN105913095A (en) * | 2016-05-17 | 2016-08-31 | 杭州申昊科技股份有限公司 | Instrument recognition method for transformer substation patrol inspection robot |
| CN106041877A (en) * | 2016-06-28 | 2016-10-26 | 张学衡 | Substation patrol robot |
| CN107590498A (en) * | 2017-09-27 | 2018-01-16 | 哈尔滨工业大学 | A kind of self-adapted car instrument detecting method based on Character segmentation level di- grader |
| CN107729896A (en) * | 2017-10-24 | 2018-02-23 | 武汉科技大学 | A kind of automatic identifying method suitable for the wide tuning drive gear formula instrument of transformer station |
| CN108491838A (en) * | 2018-03-08 | 2018-09-04 | 南京邮电大学 | Pointer-type gauges registration read method based on SIFT and HOUGH |
| CN108564085A (en) * | 2018-03-13 | 2018-09-21 | 南京大学 | A kind of method of automatic reading pointer type instrument reading |
-
2018
- 2018-11-06 CN CN201811314096.3A patent/CN109711400A/en active Pending
Patent Citations (8)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN103927507A (en) * | 2013-01-12 | 2014-07-16 | 山东鲁能智能技术有限公司 | Improved multi-instrument reading identification method of transformer station inspection robot |
| CN105045262A (en) * | 2014-11-27 | 2015-11-11 | 马金环 | Inspection robot in transformer substation able to fulfill automatic meter reading |
| CN105913095A (en) * | 2016-05-17 | 2016-08-31 | 杭州申昊科技股份有限公司 | Instrument recognition method for transformer substation patrol inspection robot |
| CN106041877A (en) * | 2016-06-28 | 2016-10-26 | 张学衡 | Substation patrol robot |
| CN107590498A (en) * | 2017-09-27 | 2018-01-16 | 哈尔滨工业大学 | A kind of self-adapted car instrument detecting method based on Character segmentation level di- grader |
| CN107729896A (en) * | 2017-10-24 | 2018-02-23 | 武汉科技大学 | A kind of automatic identifying method suitable for the wide tuning drive gear formula instrument of transformer station |
| CN108491838A (en) * | 2018-03-08 | 2018-09-04 | 南京邮电大学 | Pointer-type gauges registration read method based on SIFT and HOUGH |
| CN108564085A (en) * | 2018-03-13 | 2018-09-21 | 南京大学 | A kind of method of automatic reading pointer type instrument reading |
Non-Patent Citations (3)
| Title |
|---|
| 朱柏林: "巡检机器人云台结构设计及其指针仪表识别技术", 《中国优秀硕士论文全文数据库（电子期刊）信息科技辑》 * |
| 李学聪: "基于机器视觉指针表检测的关键技术研究", 《中国优秀博士论文全文数据库（电子期刊）信息科技辑》 * |
| 王浚哲等: "Automatic Reading System for Analog Instruments Based on Computer Vision and Inspection Robot for Power Plant", 《IEEE，10TH INTERNATIONAL CONFERENCE ON MODELLING, IDENTIFICATION AND CONTROL (ICMIC)》 * |
Cited By (10)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110084842A (en) * | 2019-05-05 | 2019-08-02 | 广东电网有限责任公司 | A kind of secondary alignment methods of machine user tripod head servo and device |
| CN110084842B (en) * | 2019-05-05 | 2024-01-26 | 广东电网有限责任公司 | Servo secondary alignment method and device for robot holder |
| CN112344980A (en) * | 2019-08-09 | 2021-02-09 | 浙江中控技术股份有限公司 | Full-automatic verification system and method for pointer instrument |
| CN110909738A (en) * | 2019-11-15 | 2020-03-24 | 杭州远鉴信息科技有限公司 | Automatic reading method of pointer instrument based on key point detection |
| CN113077518A (en) * | 2021-03-15 | 2021-07-06 | 中移(上海)信息通信科技有限公司 | Camera parameter calibration method, device and storage medium |
| CN113077518B (en) * | 2021-03-15 | 2022-02-11 | 中移(上海)信息通信科技有限公司 | Camera parameter calibration method, device and storage medium |
| CN113536895A (en) * | 2021-05-28 | 2021-10-22 | 立得空间信息技术股份有限公司 | Disc pointer meter identification method |
| CN113610094A (en) * | 2021-08-27 | 2021-11-05 | 四川中电启明星信息技术有限公司 | Distribution room pointer instrument reading method based on rotation projection calibration |
| CN113610094B (en) * | 2021-08-27 | 2023-06-02 | 四川中电启明星信息技术有限公司 | Distribution room pointer instrument reading method based on rotation projection calibration |
| CN116343232A (en) * | 2023-04-03 | 2023-06-27 | 内蒙古师范大学 | Ancient book mathematical symbol recognition method based on pre-classification |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN109711400A (en) | A kind of electric inspection process method and apparatus identifying simulated pointer formula meter reading | |
| Gatziolis et al. | 3D tree dimensionality assessment using photogrammetry and small unmanned aerial vehicles | |
| Biskup et al. | A stereo imaging system for measuring structural parameters of plant canopies | |
| CN111223133B (en) | Registration method of heterogeneous images | |
| CN110142785A (en) | A kind of crusing robot visual servo method based on target detection | |
| CN106651942A (en) | Three-dimensional rotation and motion detecting and rotation axis positioning method based on feature points | |
| CN102986372B (en) | Based on the identification of the harvesting object of full-view stereo vision, classification and space positioning apparatus and method | |
| CN108107462B (en) | RTK and high-speed camera combined traffic sign post attitude monitoring device and method | |
| CN111553245A (en) | Vegetation classification method based on machine learning algorithm and multi-source remote sensing data fusion | |
| CN109269430A (en) | The more plants of standing tree diameter of a cross-section of a tree trunk 1.3 meters above the ground passive measurement methods based on depth extraction model | |
| CN106125744A (en) | The Intelligent Mobile Robot cloud platform control method of view-based access control model servo | |
| Stürzl et al. | Three-dimensional models of natural environments and the mapping of navigational information | |
| CN113267258B (en) | Infrared temperature measurement method, device, equipment, intelligent inspection robot and storage medium | |
| CN105118055A (en) | Camera positioning correction calibration method and system | |
| CN107560592B (en) | Precise distance measurement method for photoelectric tracker linkage target | |
| Mi et al. | A vision-based displacement measurement system for foundation pit | |
| CN107192350A (en) | A kind of three-dimensional laser scanner intrinsic parameter scaling method and device | |
| CN104574406A (en) | Joint calibration method between 360-degree panorama laser and multiple visual systems | |
| CN112949478A (en) | Target detection method based on holder camera | |
| CN113674402B (en) | Plant three-dimensional hyperspectral point cloud model generation method, correction method and device thereof | |
| CN106892133A (en) | Load the aircraft head stability test device and method of camera head | |
| CN110175954A (en) | The quick joining method of improved ICP point cloud, device, electronic equipment and storage medium | |
| CN116758234A (en) | Mountain terrain modeling method based on multipoint cloud data fusion | |
| CN109308714A (en) | Camera and laser radar information method for registering based on classification punishment | |
| CN115861448A (en) | System calibration method and system based on angular point detection and characteristic point extraction |
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 | ||
| RJ01 | Rejection of invention patent application after publication | ||
| RJ01 | Rejection of invention patent application after publication |
Application publication date: 20190503 |