CN107145905A - The image recognizing and detecting method that elevator fastening nut loosens - Google Patents
The image recognizing and detecting method that elevator fastening nut loosens Download PDFInfo
- Publication number
- CN107145905A CN107145905A CN201710301044.1A CN201710301044A CN107145905A CN 107145905 A CN107145905 A CN 107145905A CN 201710301044 A CN201710301044 A CN 201710301044A CN 107145905 A CN107145905 A CN 107145905A
- Authority
- CN
- China
- Prior art keywords
- image
- bolt
- candidate
- pixel
- value
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Granted
Links
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
- G06F18/214—Generating training patterns; Bootstrap methods, e.g. bagging or boosting
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/20—Image preprocessing
- G06V10/30—Noise filtering
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/40—Extraction of image or video features
- G06V10/44—Local feature extraction by analysis of parts of the pattern, e.g. by detecting edges, contours, loops, corners, strokes or intersections; Connectivity analysis, e.g. of connected components
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Physics & Mathematics (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Data Mining & Analysis (AREA)
- Multimedia (AREA)
- Bioinformatics & Computational Biology (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Evolutionary Biology (AREA)
- Evolutionary Computation (AREA)
- Artificial Intelligence (AREA)
- General Engineering & Computer Science (AREA)
- Life Sciences & Earth Sciences (AREA)
- Image Analysis (AREA)
Abstract
The present invention proposes the image recognizing and detecting method that a kind of elevator fastening nut loosens, including：S1, allows threshold value and best alignment to match candidate shape by training enough bumper bolt positive sample images tightened to obtain optimum distance between each nut；S2, goes out some candidate frames, and filter out unnecessary candidate frame according to color value by BING algorithms selections；S3, carries out rim detection to the candidate frame of reservation, obtains the edge closure shape of bolt color mark part；Whether S4, loosen according to edge geometry and Distance Judgment bolt.
Description
Technical field
The present invention relates to computer control field, more particularly to the image recognition detection side that a kind of elevator fastening nut loosens
Method.
Background technology
Elevator is a kind of elevator gear using motor as power, in modernization and today of high speed, wide range of services
The people between tier building or cargo transport in human society.In recent years, the electrification frequently occurred causes social concerns, protects
Card elevator is normally run, and is concerned about that rider's personal safety is imperative, therefore is examined with regard to the operating bumper bolt looseness of elevator
Survey research, it is ensured that the safe and reliable operation of equipment is significant.
Conventional bolt detection includes artificial detection and the major class of automatic detection two.Artificial detection refers to be equipped with special work people
Whether member periodically has loosening with naked-eye observation bolt.Such a method is simple and easy to apply, it is not necessary to by complicated equipment.But the method
Shortcoming one be that efficiency is low, labor intensity is big, testing staff's fatiguability；Two be significantly rely on staff specialized capability and
Working attitude, detection quality is difficult to be protected.
Second major class method is the loosening using device or built-in algorithms automatic detection bolt.Add the bolt of detector
It is to read the signal that sensor is sent by detector to loosen detection method, will be rotated when nut loosens with active
It is dynamic, when the metal coating region on driven gear surface goes to the top of sensor mounting groove, will block sensor to
The signal that detector is sent, loosens so that fastener just can be detected.
The more commonly used built-in algorithms are the loosenings that bumper bolt is considered using bolt rotation angle.Common practices is
One is drawn at bolt connection position continuously vertically clearly to mark, and the angle between bolt is necessarily caused when bolt loosens
Degree changes, and now extracts photo and detects that the angle between mark can detect that whether bolt loosens, but the method
It is not good with regard to failture evacuation effect in bolt just rotation integer numbers circle.
The content of the invention
It is contemplated that at least solving technical problem present in prior art, one kind is especially innovatively proposed
In order to realize the above-mentioned purpose of the present invention, the invention provides the image recognition inspection that a kind of elevator fastening nut loosens
Survey method, including：
S1, allows threshold value by training enough bumper bolt positive sample images tightened to obtain optimum distance between each nut
Candidate shape is matched with best alignment；
S2, goes out some candidate frames, and filter out unnecessary candidate frame according to color value by BING algorithms selections；
S3, carries out rim detection to the candidate frame of reservation, obtains the edge closure shape of bolt color mark part；
Whether S4, loosen according to edge geometry and Distance Judgment bolt.
The image recognizing and detecting method that described elevator fastening nut loosens, it is preferred that the S1 includes：
S1-1, the small spacing between each nut, nut school are obtained by enough bumper bolt positive sample images tightened
The average allowable error of quasi- shape horizontal alignment and vertical alignment, and calibration pairing candidate shape geometry, colouring information, are trained
Threshold value and best alignment is allowed to match candidate shape to optimum distance.
The image recognizing and detecting method that described elevator fastening nut loosens, it is preferred that the S2 is selected by BING algorithms
Selecting out some candidate frames includes：
The sample image for selecting label target position in S2-1, BING algorithm first generates the positive negative sample of different scale,
And by each sample size scaling to fixing 8 × 8 sizes；Under the size, the minutia such as bolt texture, local shape will lose
Lose, the boundary gradient contour feature of an object of reservation；Contour of object Grad is higher, forms strong right with its surrounding background area
Than；
S2-2, secondly trains positive and negative sample set using linear SVM, obtains target like physical property detection template；One
The destination object is more likely expressed with the Pixel Dimensions candidate frame of size with destination object, therefore the template equally ensures to be 64
8 × 8 sizes of dimension；Then scoring system mechanism is utilized, the filtering score under each yardstick is calculated and sorts, utilize non-maximum
Restrainable algorithms (Non-maximum suppression, NMS) eliminate local redundancy；
S2-3, finally finds the marking corresponding candidate frame size of point and preserves in original image.
The image recognizing and detecting method that described elevator fastening nut loosens, it is preferred that the S2-3 includes：
a：BING calculates Gradient Features to each size sample, is matched according to Gradient Features and obtains one group with filtering score
Candidate frame, filtering score slIt is defined as follows：
s(i,x,y)=＜ w, G(i,x,y)＞,
Wherein, the model parameter that w obtains for training, G(i,x,y)It is the frames images scaling for the position coordinates (x, y) that yardstick is i
NG (Normed Gradients) feature under to small size, symbol "<,>" represent inner product operation；W, G are matrix；
Calculate, obtained using Prewitt operator modes for the gradient magnitude of each pixel：
g(i,x,y)=min | gx|+|gy|, 255 }, wherein g (i, x, y) represent yardstick i position coordinates (x, y) each
Pixel gradient amplitude, gxFor the pixel gradient amplitude of X-coordinate, gyFor the pixel gradient amplitude of Y-coordinate；
Because filtering fraction can have deviation because of the difference of size, therefore needed when being reordered to final candidate score
Calibrate and obtain candidate's window group score that final different sizes are carried, define objectivity score olIt is as follows：
Wherein, viIt is the independent study coefficient under size i, tiRepresent size i offset；
The candidate frame of acquisition should not only have the top score in global scope, should also have neighborhood highest to obtain
Point, to avoid that because highest window score excessively concentrates a certain region the phenomenon of face missing inspection may be caused；Therefore using non-
Maximum bounding algorithm selects marginal point, so that while final candidate window has global higher score, being also equipped with neighbour
Domain highest scoring.
The image recognizing and detecting method that described elevator fastening nut loosens, it is preferred that the S2 includes：
To improve calculating speed, accelerate feature extraction and test process, the tool of the binaryzation approximation of template and feature
Body step is as follows：
Feature binaryzation：Binary system is carried out due to gradient magnitude span [0,255], therefore using equation below mode
Bit stream is replaced；
Wherein, bk∈ { 0,1 } represents kth position binary value, NgThe binary digit number chosen is represented, due to rear four place value
To gradients affect and little；
8 × 8 every corresponding kth positions are combined and are expressed as binary stream bk∈{0,1}64；Therefore 8 × 8 matrix areas of correspondence
Domain has：
Template binaryzation：The model parameter obtained for w for training, then regard the combination of multiple base vectors, approximate representation as
ForWherein, βjRepresent the coefficient of j-th of base vector, ajRepresent j-th of base vector, aj∈{-1,1}64,
NwBase vector number is represented, by ajIt is transformed into [0,1] scope：
Therefore w is expressed as：
Formula s(i,x,y)=＜ w, G(i,x,y)＞ binaryzation formula is expressed as follows：
WhereinRefer to
By ajNegative value in vector is set to zero, on the occasion of then constant.
The image recognizing and detecting method that described elevator fastening nut loosens, it is preferred that the process of the template w includes：
Input：The base vector ε of initialization,
Output：Template w,
The image recognizing and detecting method that described elevator fastening nut loosens, it is preferred that also include：
For eliminate it is unnecessary select frame, be implemented as follows：
(1) framed score descending is arranged, chooses best result and its corresponding frame,
(2) remaining frame is traveled through, if being more than certain threshold value with the overlapping area (IOU) of current best result frame, we are just
Frame is deleted,
(3) continue to select highest scoring from untreated frame, repeat said process.
The image recognizing and detecting method that described elevator fastening nut loosens, it is preferred that the S3 also includes：
S3-1, carries out rim detection to candidate region using Canny, first carries out convolution to eliminate with 2D gaussian filterings template
Picture noise, calculates each pixel in filtered image the small big and direction of its gradient, this is found by gradient direction
The adjacent pixels in pixel gradient direction, bolt edge closure is obtained finally by non-maximum suppression and thresholding and edge link
Pattern curve, calculates the area of edge closure shape；
S3-2, algorithmic procedure is as follows：
The Gaussian kernel for the size=5 for using 2D gaussian filterings template use in convolution noise reduction is as follows：
Calculate pixel amplitude and direction, here according to Sobel filter the step of be introduced：
A. with a pair of convolution arrays, x, y directions, array are respectively acting on
B. Grad and direction are calculated using following equation：
One of approximate to four possible angles of gradient direction；
Along the maximum point of argument angle detecting modulus value, i.e. marginal point, pixel is traveled through, each pixel local derviation value and phase
The modulus value of adjacent pixel compares, and it is marginal point to take its maximum, and it is 0 to put grey scale pixel value；
S3-3, detects and connects edge with double-purpose dual threashold value-based algorithm：
A. two threshold value th are acted on to non-maxima suppression image1And th2, both sides relation th1=0.4th2；We are ladder
Angle value is less than th1The gray value of pixel be set to 0, obtain image 1；Then Grad is less than th2The gray value of pixel be set to
0, obtain image 2；Because the threshold value of image 2 is higher, most of noise is removed, but also have lost useful marginal information simultaneously；
And the threshold value of image 1 is relatively low, more information is remained, we can be linked based on image 2 with image 1 for supplement
The edge of image；
B. comprising the following steps that for edge is linked：
Image 2 is scanned, when running into pixel p (x, y) of non-zero gray scale, tracking is with p (x, y) for starting point
Contour line, until the terminal q (x, y) of contour line；Put in image under consideration 1 with q (x, y) in image 2 the corresponding point s in position (x,
Y) 8 adjacent domains；If thering is non-zero pixels s (x, y) to exist in 8 adjacent domains that s (x, y) is put, then it is included to figure
In 2, r (x, y) points are used as；Since r (x, y), the first step is repeated, until we can not continue in image 1 and image 2
Untill；
After the link to the contour line comprising p (x, y) is completed, this contour line is labeled as to have accessed；Return to
The first step, finds next contour line；The first step, second step, the 3rd step being repeated, being until can not find new contour line in image 2
Only；
So far, the rim detection of canny operators is completed, two edge closure shapes of bolt red-label part are obtained.
The image recognizing and detecting method that described elevator fastening nut loosens, it is preferred that the S4 includes：
S4-1, calculates the edge closure shape of two red-label parts of bolt, and carry out with close-shaped in positive sample
Compare, when close-shaped area with positive sample differs more, or only one of which is close-shaped, then think that pine occurs for bolt
It is dynamic；
S4-2, on the premise of geometry is met, then to judge whether two close-shaped intermediate distances are being allowed
Threshold range in, if in threshold range, then it is assumed that bolt does not loosen, if be unsatisfactory for, assert bolt hair
Loosening is given birth to.
In summary, by adopting the above-described technical solution, the beneficial effects of the invention are as follows：
Come together to judge whether bolt loosens in terms of angle and geometry two, further increase detection
Accuracy rate.
The additional aspect and advantage of the present invention will be set forth in part in the description, and will partly become from the following description
Obtain substantially, or recognized by the practice of the present invention.
Brief description of the drawings
The above-mentioned and/or additional aspect and advantage of the present invention will become from description of the accompanying drawings below to embodiment is combined
Substantially and be readily appreciated that, wherein：
Fig. 1 is that bolt looseness detection method flow chart is shouldered in insurance during elevator of the present invention is run；
Fig. 2 is that the schematic diagram that bolt looseness detection method bolt is tightened and loosened is shouldered in insurance during elevator of the present invention is run；
The step of Fig. 3 is and completes rim detection using canny algorithms is schemed；
Fig. 4 A are edge direction schematic diagram, and Fig. 4 B are 8 neighborhood argument direction schematic diagrams；
Embodiment
Embodiments of the invention are described below in detail, the example of the embodiment is shown in the drawings, wherein from beginning to end
Same or similar label represents same or similar element or the element with same or like function.Below with reference to attached
The embodiment of figure description is exemplary, is only used for explaining the present invention, and is not considered as limiting the invention.
In the description of the invention, it is to be understood that term " longitudinal direction ", " transverse direction ", " on ", " under ", "front", "rear",
The orientation or position relationship of the instruction such as "left", "right", " vertical ", " level ", " top ", " bottom " " interior ", " outer " is based on accompanying drawing institutes
The orientation or position relationship shown, is for only for ease of the description present invention and simplifies description, rather than indicate or imply signified dress
Put or element there must be specific orientation, with specific azimuth configuration and operation, therefore it is not intended that to the limit of the present invention
System.
In the description of the invention, unless otherwise prescribed with limit, it is necessary to explanation, term " installation ", " connected ",
" connection " should be interpreted broadly, for example, it may be mechanically connect or electrical connection or the connection of two element internals, can
To be to be joined directly together, it can also be indirectly connected to by intermediary, for the ordinary skill in the art, can basis
Concrete condition understands the concrete meaning of above-mentioned term.
The detection method of bolt looseness is shouldered the invention provides insurance in a kind of operation of elevator, innovatively from distance and several
The angle of what shape recognizes inaccurate problem to solve bolt looseness.
In order to realize the above-mentioned purpose of the present invention, the invention provides the inspection that bolt looseness is shouldered in insurance in a kind of operation of elevator
Survey method, including：
Step 1：Allowed by training enough bumper bolt positive sample images tightened to obtain optimum distance between each nut
Threshold value and best alignment pairing candidate shape；
Step 2：Some candidate frames are gone out by BING algorithms selections, and unnecessary candidate frame is filtered out according to color value.
Step 3：Rim detection is carried out to the candidate frame of reservation, the edge closure shape of bolt color mark part is obtained；
Step 4：Whether loosened according to edge geometry and Distance Judgment bolt.
Above-mentioned technical proposal has the beneficial effect that：Bolt detection method proposed by the present invention, from geometry and distance two
Individual standard is more comprehensively made that detection to bolt looseness, and this method has higher accuracy rate, and can carry out automatic
Detection, it is time saving and energy saving, improve the specific aim of bolt looseness detection, it is ensured that the safe and reliable operation of equipment；The present invention need not
Add any equipment, it is only necessary to which BR bolt picture, which reaches processing end, to detect whether bolt loosens in real time, and
Energy and alarm.
Bolt looseness detection method is shouldered in insurance in described elevator operation, it is preferred that the step 1 includes：
Step 1-1 obtains the small spacing between each nut, nut by enough bumper bolt positive sample images tightened
Calibrate shape horizontal alignment and the average allowable error of vertical alignment, and calibration pairing candidate shape geometry, colouring information, training
Obtaining optimum distance allows threshold value and best alignment to match candidate shape；
Above-mentioned technical method has the beneficial effect that：Threshold value and best alignment is most preferably allowed to match by training positive sample to obtain
Candidate shape, provides to bolt looseness and accurately judges to mark, improve the precision ratio and recall ratio of bolt looseness.
Bolt looseness detection method is shouldered in insurance in described elevator operation, it is preferred that the step 2 includes：
Step 2-1, bolt candidate frame region is detected using BING algorithms,
(1) sample image for selecting label target position in BING algorithms first generates the positive negative sample of different scale, and
By each sample size scaling to fixing 8 × 8 sizes.Under the size, the minutia such as bolt texture, local shape will lose,
The boundary gradient contour feature of object of reservation.Contour of object Grad is higher, and sharp contrast is formed with its surrounding background area.
(2) positive and negative sample set secondly is trained using linear SVM, obtains target like physical property detection template.One with
Destination object more likely expresses the destination object with the Pixel Dimensions candidate frame of size, therefore the template equally ensures as 64 dimensions
8 × 8 sizes.Then scoring system mechanism is utilized, the filtering score under each yardstick is calculated and sorts, pressed down using non-maximum
Algorithm (Non-maximum suppression, NMS) processed eliminates local redundancy.
(3) it is last that the marking corresponding candidate frame size of point is found in original image and is preserved.
a：BING calculates Gradient Features to each size sample, is matched according to Gradient Features and obtains one group with filtering score
Candidate frame, filtering score slIt is defined as follows：
s(i,x,y)=＜ w, G(i,x,y)＞
Wherein, w obtains for training model parameter (hereinafter template w), G(i,x,y)It is that i, position coordinates are for yardstick
The NG features that the frames images of (x, y) are zoomed under small size, here using 8 × 8 sizes.And symbol "<,>" represent inner product operation.
Calculate, obtained here using the Prewitt operator modes of second chapter for the gradient magnitude of each pixel：
g(i,x,y)=min | gx|+|gy|,255}
Because filtering fraction can have deviation because of the difference of size, for example, there is face in 10 × 80 window
Possibility is far smaller than in the window that size is 30 × 30.Therefore calibration is needed to obtain most when being reordered to final candidate score
Candidate's window group score that difference size is carried eventually, defines objectivity score olIt is as follows：
Wherein, viIt is the independent study coefficient and t under size iiRepresent size i offset.
The candidate frame of acquisition should not only have the top score in global scope, should also have neighborhood highest to obtain
Point, to avoid that because highest window score excessively concentrates a certain region the phenomenon of face missing inspection may be caused.Therefore using non-
Maximum bounding algorithm selects marginal point, so that while final candidate window has global higher score, being also equipped with neighbour
Domain highest scoring.
To improve calculating speed, accelerate feature extraction and test process, the tool of the binaryzation approximation of template and feature
Body step is as follows：
1st, feature binaryzation：Two are carried out due to gradient magnitude span [0,255], therefore using equation below mode
System bit stream is replaced.
Wherein, bk∈ { 0,1 } represents kth position binary value, NgThe binary digit number chosen is represented, due to rear four place value
To gradients affect and little, to further speed up computing, N hereg=4, only retain high four figures value.By taking numerical value 122 as an example, two
It is 01111010 after value, it is 0111 to retain high four figures value.
For the region for extending to 8 × 8,64 gradient magnitudes are co-existed in, the gradient magnitude of every can all use binary system ratio
Spy's stream.Now 8 × 8 every corresponding kth positions are combined and are expressed as binary stream bk∈{0,1}64.Therefore 8 × 8 matrix areas of correspondence
Have：
2nd, template binaryzation：For template w, then it can regard the combination of multiple base vectors as, approximate representation is
Wherein, βjRepresent the coefficient of j-th of base vector, ajRepresent j-th of base vector, aj∈{-1,1}64, NwBase vector number is represented, this
In take 2.It is convenient for processing, by ajIt is transformed into [0,1] scope：
Therefore w is represented by：
Template w specific calculation process is as follows：
To sum up, formula s(i,x,y)=＜ w, G(i,x,y)＞ binaryzation formula is expressed as follows：
b：Using non-maxima suppression using quite varied, be herein for see eliminate it is unnecessary select frame, implement as
Under：
(1) framed score descending is arranged, chooses best result and its corresponding frame,
(2) remaining frame is traveled through, if being more than certain threshold value with the overlapping area (IOU) of current best result frame, we are just
Frame is deleted,
(3) continue to select highest scoring from untreated frame, repeat said process.
Above-mentioned technical method has the beneficial effect that：Can be by all near regions that connect with positive sample Grad using BING algorithms
Field mark comes out, and combines the region that a NMS elimination parts are unsatisfactory for requiring.
Step 2-2：Color value scope according to being marked in bolt filters out a part of candidate frame again, obtains comprising bolt most
Good candidate frame；
, it is necessary to link the clear and vertical red bar that position draws one wide about 1 centimetre in bolt in actual scene
Shape frame, and can keep colour-fast for a long time, marking is simple to operate, and application cost is low, can be to automatic knowledge below
Very big booster action is not played.In this experiment, the color averages of RGB tri- in red-label region are respectively 119.37,43.45
With 53.73, therefore we can be limited when object RGB meets three color ranges in candidate frame, just think that this candidate frame is included
Bolt region.
The color ranges of RGB tri- in red-label region are as follows on bolt：
The candidate frame for not including this colouring information can be filtered out according to the color range values of RGB tri-, optimal candidate frame is obtained.
Above-mentioned technical method has the beneficial effect that：Simply have similar to positive sample by the candidate frame obtained by step 2
The region of Grad, and and include the region of bolt part, therefore the method for sample color mark can further filter out face
The incongruent region of color information.
Bolt looseness detection method is shouldered in insurance in described elevator operation, it is preferred that the step 3 includes：
Step 3-1, using Canny to candidate region carry out rim detection, first carried out with 2D gaussian filterings template convolution with
Picture noise is eliminated, the small big and direction of its gradient is calculated each pixel in filtered image, is looked for by gradient direction
To the adjacent pixels in the pixel gradient direction, bolt edge is obtained finally by non-maximum suppression and thresholding and edge link
Close-shaped curve, calculates the area of edge closure shape.Algorithmic procedure is as follows：
(1) Gaussian kernel using the size=5 used in 2D gaussian filterings template progress convolution noise reduction is as follows：
(2) calculate pixel amplitude and direction, here according to Sobel filter the step of be introduced：
A. with a pair of convolution arrays (being respectively acting on x, y directions)
B. Grad and direction are calculated using following equation：
One of approximate to four possible angles of gradient direction (general 0,45,90,135 degree)
(3) gradient for only obtaining the overall situation e insufficient to determine edge, therefore to determine edge, it is necessary to retain partial gradient
Maximum point.The essence of non-maxima suppression algorithm (Non-maximum suppression, NMS) is search local maximum,
Suppress non-maximum element.
Fig. 4 A are edge direction schematic diagram, and Fig. 4 B are 8 neighborhood argument direction schematic diagrams；
As shown in figure 4, along the maximum point of argument angle detecting modulus value, i.e. marginal point, 8 direction pixels are traveled through,
Each pixel local derviation value is compared with the modulus value of adjacent pixel, and it is marginal point to take its maximum, and it is 0 to put grey scale pixel value.
(4) with the detection of double-purpose dual threashold value-based algorithm and connection edge：
A. two threshold value th are acted on to non-maxima suppression image1And th2, both sides relation th1=0.4th2.We are ladder
Angle value is less than th1The gray value of pixel be set to 0, obtain image 1.Then Grad is less than th2The gray value of pixel be set to
0, obtain image 2.Because the threshold value of image 2 is higher, most of noise is removed, but also have lost useful marginal information simultaneously.
And the threshold value of image 1 is relatively low, more information is remained, we can be linked based on image 2 with image 1 for supplement
The edge of image.
B. comprising the following steps that for edge is linked：
Image 2 is scanned, when running into pixel p (x, y) of non-zero gray scale, tracking is with p (x, y) for starting point
Contour line, until the terminal q (x, y) of contour line.Put in image under consideration 1 with q (x, y) in image 2 the corresponding point s in position (x,
Y) 8 adjacent domains.If thering is non-zero pixels s (x, y) to exist in 8 adjacent domains that s (x, y) is put, then it is included to figure
In 2, r (x, y) points are used as.Since r (x, y), the first step is repeated, until we can not continue in image 1 and image 2
Untill.
After the link to the contour line comprising p (x, y) is completed, this contour line is labeled as to have accessed.Return to
The first step, finds next contour line.The first step, second step, the 3rd step being repeated, being until can not find new contour line in image 2
Only.
So far, the rim detection of canny operators is completed, two edge closure shapes of bolt red-label part are obtained.
Above-mentioned technical method has the beneficial effect that：When the colouring information that there is colouring information in picture with marked on bolt
When close, bottom cannot be determined well, and which is the sample for including bolt region.Therefore jointing edge is wanted to examine
The edge geometry for calculating object contained by candidate frame is surveyed compared with the geometry that positive sample is included, with further
The candidate frame for the condition of being unsatisfactory for is filtered out, bolt region is eventually found.
Bolt looseness detection method is shouldered in insurance in described elevator operation, it is preferred that the step 1 includes：
Step 4-1, calculates the edge closure shape of bolt two red-label parts, and with positive sample it is close-shaped enter
Row compares, when close-shaped area with positive sample differs more, or only one of which is close-shaped, then think bolt
Loosen.
Step 4-2, on the premise of geometry is met, then to judge whether two close-shaped intermediate distances are being permitted
Perhaps in threshold range, if in threshold range, then it is assumed that bolt does not loosen, if be unsatisfactory for, we then think
Bolt is loosened.
Above-mentioned technical method has the beneficial effect that：Combine whether judge bolt in terms of geometric angle and distance two
Loosened, flase drop and the probability of missing inspection can be reduced to a certain extent, so as to lift the engineer applied valency of whole algorithm
Value.
In summary, by adopting the above-described technical solution, the beneficial effects of the invention are as follows：
(1) detection program built in, being capable of automatic detection bolt looseness situation, it is to avoid the poor efficiency manually detected, and
Detect when bolt loosens can very first time alarm, it is to avoid the potential safety hazard brought by bolt looseness；
(2) sets of plan need not buy other auxiliary equipments, only program need to be built into single-chip microcomputer, save trouble and labor,
And it is a kind of relatively inexpensive scheme.
(3) the set bolt looseness detection algorithm carries out loosening detection, energy in terms of geometry and distance two to bolt
Loosening situation is gone out with larger Probability Detection, with relatively low loss and flase drop.
The present invention discloses bumper bolt looseness detection method in a kind of elevator operation, including：
Step 1：Allowed by training enough bumper bolt positive sample images tightened to obtain optimum distance between each nut
Threshold value and best alignment pairing candidate shape；
Step 2：Some candidate frames are gone out by BING algorithms selections, and unnecessary candidate frame is filtered out according to color value.
Step 3：Rim detection is carried out to the candidate frame of reservation, the edge closure shape of bolt color mark part is obtained；
Step 4：Whether loosened according to edge geometry and Distance Judgment bolt.
As shown in figure 1, when sample exposal model when, the bolt looseness based on geometry and centre distance add survey method by
Following steps are constituted：
S1, the geometry of the bolt region of calculating positive sample picture and two mark positions centre distance, are determined several
What shape and distance threshold.
S2, fixed video camera is set up in hoistway wall or pipe well, picture is persistently shot to bolt region.
The position of bolt in usual elevator is changeless, and the picture captured by camera has the more fixed back of the body
Scape, this has booster action to automatic detection later.The present invention considers floor floor height, elevator translational speed, shooting
The factors such as machine visual angle range, video camera frame per second, terminal storage capacity, choose reasonable time interval and are shot.
S3, using BING algorithms relevant treatment is carried out to captured picture, extract the candidate frame for meeting related request.
In order to preferably be detected to bolt looseness, the positive sample picture that bolt does not loosen is inputted first, this area is obtained
The relevant information in domain, is secondly delivered to processor by picture to be detected, is picked out using BING algorithms similar to positive sample information
Candidate frame.After the completion of, filter out the relatively low candidate frame of confidence level in conjunction with non-maxima suppression algorithm, for guarantee detect below it is suitable
Profit is carried out.
S4：The candidate frame that previous step is obtained is generally more, therefore recycles the colouring information that bolt is marked to enter
One step filters out a part of candidate frame.
First choice obtains the RGB average values of bolt red-label part in positive sample, and sets different TGB to take according to this value
It is worth scope；Then the RGB color information that each candidate frame includes object in picture to be detected is calculated again, if in the candidate frame
The rgb value of included object retains this candidate frame in set scope, just, otherwise just rejects.
Even if S5, the edge shape for retaining marked region in candidate frame, and calculate its geological information.
The candidate frame remained by upper two step is all consistent in terms of gradient and colouring information with positive sample, but works as
Occur in background picture the object similar to mark part color when, follow-up identification work will be produced and disturbed.Now
Equally the picture to be detected for having marked some candidate frames is detected, the object in candidate frame is entered using canny algorithms
Row rim detection, obtains its edge closure shape, and calculate its geological information.
S6：Judge whether the geological information calculated is consistent with the geological information of positive sample, illustrate if not being consistent
Bolt is loosened, and further judges whether the intermediate distance of two enclosed regions is less than the related threshold set if be consistent
Value, is loosened if greater than bolt is then also considered as, then is alarmed in the very first time.If two above condition is all met, I
Be considered as bolt and be up.
In the description of this specification, reference term " one embodiment ", " some embodiments ", " example ", " specifically show
The description of example " or " some examples " etc. means to combine specific features, structure, material or the spy that the embodiment or example are described
Point is contained at least one embodiment of the present invention or example.In this manual, to the schematic representation of above-mentioned term not
Necessarily refer to identical embodiment or example.Moreover, specific features, structure, material or the feature of description can be any
One or more embodiments or example in combine in an appropriate manner.
Although an embodiment of the present invention has been shown and described, it will be understood by those skilled in the art that：Not
In the case of departing from the principle and objective of the present invention a variety of change, modification, replacement and modification can be carried out to these embodiments, this
The scope of invention is limited by claim and its equivalent.
Claims (9)
1. the image recognizing and detecting method that a kind of elevator fastening nut loosens, it is characterised in that including：
S1, allows threshold value and most by training enough bumper bolt positive sample images tightened to obtain optimum distance between each nut
Good calibration pairing candidate shape；
S2, goes out some candidate frames, and filter out unnecessary candidate frame according to color value by BING algorithms selections；
S3, carries out rim detection to the candidate frame of reservation, obtains the edge closure shape of bolt color mark part；
Whether S4, loosen according to edge geometry and Distance Judgment bolt.
2. the image recognizing and detecting method that elevator fastening nut according to claim 1 loosens, it is characterised in that the S1
Including：
S1-1, the small spacing between each nut, nut calibration shape are obtained by enough bumper bolt positive sample images tightened
Shape horizontal alignment and the average allowable error of vertical alignment, and calibration pairing candidate shape geometry, colouring information, training are obtained most
Good distance allows threshold value and best alignment to match candidate shape.
3. the image recognizing and detecting method that elevator fastening nut according to claim 1 loosens, it is characterised in that the S2
Going out some candidate frames by BING algorithms selections includes：
The sample image for selecting label target position in S2-1, BING algorithm first generates the positive negative sample of different scale, and will
Each sample size scaling is to fixing 8 × 8 sizes；Under the size, the minutia such as bolt texture, local shape will lose, only
The boundary gradient contour feature of object of reservation；Contour of object Grad is higher, and sharp contrast is formed with its surrounding background area；
S2-2, secondly trains positive and negative sample set using linear SVM, obtains target like physical property detection template；One and mesh
Mark object more likely expresses the destination object with the Pixel Dimensions candidate frame of size, therefore the template equally ensures as 64 dimensions
8 × 8 sizes；Then scoring system mechanism is utilized, the filtering score under each yardstick is calculated and sorts, utilize non-maxima suppression
Algorithm (Non-maximum suppression, NMS) eliminates local redundancy；
S2-3, finally finds the marking corresponding candidate frame size of point and preserves in original image.
4. the image recognizing and detecting method that elevator fastening nut according to claim 3 loosens, it is characterised in that described
S2-3 includes：
a：BING calculates Gradient Features to each size sample, is matched according to Gradient Features and obtains one group of candidate with filtering score
Frame, filtering score slIt is defined as follows：
s(i,x,y)=＜ w, G(i,x,y)＞,
Wherein, the model parameter that w obtains for training, G(i,x,y)Be yardstick be i position coordinates (x, y) frames images zoom to it is small
NG (Normed Gradients) feature under size, symbol "<,>" represent inner product operation；W, G are matrix；
Calculate, obtained using Prewitt operator modes for the gradient magnitude of each pixel：
g(i,x,y)=min | gx|+|gy|, 255 }, wherein g (i, x, y) represents each pixel of yardstick i position coordinates (x, y)
Gradient magnitude, gxFor the pixel gradient amplitude of X-coordinate, gyFor the pixel gradient amplitude of Y-coordinate；
Because filtering fraction can have deviation because of the difference of size, therefore school is needed when being reordered to final candidate score
Candidate's window group score that final different sizes are carried will definitely be arrived, objectivity score o is definedlIt is as follows：
o(i,x,y)=vis(i,x,y)+ti,
Wherein, viIt is the independent study coefficient under size i, tiRepresent size i offset；
The candidate frame of acquisition should not only have the top score in global scope, should also have neighborhood top score, with
Avoid that because highest window score excessively concentrates a certain region the phenomenon of face missing inspection may be caused；Therefore non-maximum is utilized
Bounding algorithm selects marginal point, so that while final candidate window has global higher score, being also equipped with neighborhood score
Highest.
5. the image recognizing and detecting method that elevator fastening nut according to claim 1 loosens, it is characterised in that the S2
Including：
To improve calculating speed, accelerate feature extraction and test process, the specific step of the binaryzation approximation of template and feature
It is rapid as follows：
Feature binaryzation：Binary bits are carried out due to gradient magnitude span [0,255], therefore using equation below mode
Stream is replaced；
Wherein, bk∈ { 0,1 } represents kth position binary value, NgThe binary digit number chosen is represented, because rear four place value is to ladder
Degree influence is simultaneously little；
8 × 8 every corresponding kth positions are combined and are expressed as binary stream bk∈{0,1}64；Therefore 8 × 8 matrix areas of correspondence have：
Template binaryzation：The model parameter obtained for w for training, then regard the combination of multiple base vectors as, approximate representation isWherein, βjRepresent the coefficient of j-th of base vector, ajRepresent j-th of base vector, aj∈{-1,1}64, Nw
Base vector number is represented, by ajIt is transformed into [0,1] scope：
Therefore w is expressed as：
Formula s(i,x,y)=＜ w, G(i,x,y)＞ binaryzation formula is expressed as follows：
WhereinRefer to aj
Negative value in vector is set to zero, on the occasion of then constant.
6. the image recognizing and detecting method that elevator fastening nut according to claim 5 loosens, it is characterised in that the mould
Plate w process includes：
Input：The base vector ε of initialization,
Output：Template w,
1:For j=1 to Nw do
2:aj=sign (ε)
3:
4:ε=ε-βjaj
5:W=w+ βjaj。
7. the image recognizing and detecting method that elevator fastening nut according to claim 5 loosens, it is characterised in that also wrap
Include：
For eliminate it is unnecessary select frame, be implemented as follows：
(1) framed score descending is arranged, chooses best result and its corresponding frame,
(2) remaining frame is traveled through, if being more than certain threshold value with the overlapping area (IOU) of current best result frame, we are just by frame
Delete,
(3) continue to select highest scoring from untreated frame, repeat said process.
8. the image recognizing and detecting method that elevator fastening nut according to claim 1 loosens, it is characterised in that the S3
Also include：
S3-1, carries out rim detection to candidate region using Canny, first carries out convolution to eliminate image with 2D gaussian filterings template
Noise, calculates each pixel in filtered image the small big and direction of its gradient, the pixel is found by gradient direction
The adjacent pixels of gradient direction, bolt edge closure shape is obtained finally by non-maximum suppression and thresholding and edge link
Curve, calculates the area of edge closure shape；
S3-2, algorithmic procedure is as follows：
The Gaussian kernel for the size=5 for using 2D gaussian filterings template use in convolution noise reduction is as follows：
Calculate pixel amplitude and direction, here according to Sobel filter the step of be introduced：
A. with a pair of convolution arrays, x, y directions, array are respectively acting on
B. Grad and direction are calculated using following equation：
One of approximate to four possible angles of gradient direction；
Along the maximum point of argument angle detecting modulus value, i.e. marginal point, pixel is traveled through, each pixel local derviation value and adjacent picture
The modulus value of element compares, and it is marginal point to take its maximum, and it is 0 to put grey scale pixel value；
S3-3, detects and connects edge with double-purpose dual threashold value-based algorithm：
A. two threshold value th are acted on to non-maxima suppression image1And th2, both sides relation th1=0.4th2；We are small Grad
In th1The gray value of pixel be set to 0, obtain image 1；Then Grad is less than th2The gray value of pixel be set to 0, obtain
Image 2；Because the threshold value of image 2 is higher, most of noise is removed, but also have lost useful marginal information simultaneously；And image 1
Threshold value it is relatively low, remain more information, we can link the side of image with image 1 based on image 2 for supplement
Edge；
B. comprising the following steps that for edge is linked：
Image 2 is scanned, when running into pixel p (x, y) of non-zero gray scale, tracked with wheel that p (x, y) is starting point
Profile, until the terminal q (x, y) of contour line；The 8 of the corresponding point s (x, y) in position is put in image under consideration 1 with q (x, y) in image 2
Adjacent domain；If thering is non-zero pixels s (x, y) to exist in 8 adjacent domains that s (x, y) is put, then it is included in image 2,
It is used as r (x, y) points；Since r (x, y), the first step is repeated, untill we can not continue in image 1 and image 2；
After the link to the contour line comprising p (x, y) is completed, this contour line is labeled as to have accessed；Return to first
Step, finds next contour line；The first step, second step, the 3rd step are repeated, untill it can not find new contour line in image 2；
So far, the rim detection of canny operators is completed, two edge closure shapes of bolt red-label part are obtained.
9. the image recognizing and detecting method that elevator fastening nut according to claim 1 loosens, it is characterised in that the S4
Including：
S4-1, calculates the edge closure shape of two red-label parts of bolt, and is compared with close-shaped in positive sample,
When close-shaped area with positive sample differs more, or only one of which is close-shaped, then think that bolt loosens；
S4-2, on the premise of geometry is met, then to judge two close-shaped intermediate distances whether in the threshold allowed
In the range of value, if in threshold range, then it is assumed that bolt does not loosen, if be unsatisfactory for, assert that bolt there occurs
Loosen.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201710301044.1A CN107145905B (en) | 2017-05-02 | 2017-05-02 | Image recognition detection method for looseness of elevator fastening nut |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201710301044.1A CN107145905B (en) | 2017-05-02 | 2017-05-02 | Image recognition detection method for looseness of elevator fastening nut |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN107145905A true CN107145905A (en) | 2017-09-08 |
| CN107145905B CN107145905B (en) | 2020-04-21 |
Family
ID=59775084
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201710301044.1A Active CN107145905B (en) | 2017-05-02 | 2017-05-02 | Image recognition detection method for looseness of elevator fastening nut |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN107145905B (en) |
Cited By (20)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN107657275A (en) * | 2017-09-20 | 2018-02-02 | 电子科技大学 | Based on the car plate pre-detection method for improving BING algorithms |
| CN107966270A (en) * | 2017-11-21 | 2018-04-27 | 韶关学院 | A kind of bolt based on sparse dictionary tightens detection method |
| CN108416287A (en) * | 2018-03-04 | 2018-08-17 | 南京理工大学 | A kind of pedestrian detection method excavated based on omission negative sample |
| CN108416336A (en) * | 2018-04-18 | 2018-08-17 | 特斯联（北京）科技有限公司 | A kind of method and system of intelligence community recognition of face |
| CN109029382A (en) * | 2018-08-20 | 2018-12-18 | 上海矩尺土木科技有限公司 | A kind of screw retention early warning and monitoring device |
| CN109389566A (en) * | 2018-10-19 | 2019-02-26 | 辽宁奇辉电子系统工程有限公司 | Subway height adjusting valve fastening nut defective mode detection method based on boundary characteristic |
| CN110097536A (en) * | 2019-04-10 | 2019-08-06 | 东南大学 | Hexagon bolt looseness detection method based on deep learning and Hough transformation |
| WO2019183475A1 (en) * | 2018-03-23 | 2019-09-26 | University Of Kansas | Vision-based fastener loosening detection |
| CN110390668A (en) * | 2019-06-26 | 2019-10-29 | 石家庄铁道大学 | Bolt looseness detection method, terminal device and storage medium |
| CN110570472A (en) * | 2018-06-05 | 2019-12-13 | 成都精工华耀科技有限公司 | Fastener loosening detection method based on artificial marking image |
| CN110930366A (en) * | 2019-10-30 | 2020-03-27 | 同济大学 | Nut loosening detection method for wind power tower cylinder |
| CN111122135A (en) * | 2019-12-05 | 2020-05-08 | 西安交通大学 | Method for evaluating looseness degree of flange bolt connection structure |
| CN111666943A (en) * | 2020-04-20 | 2020-09-15 | 国网浙江省电力有限公司 | Matching detection method and system for bolt and nut of power transmission line based on image recognition |
| CN111814709A (en) * | 2020-07-14 | 2020-10-23 | 南京航空航天大学苏州研究院 | Airplane surface bolt tightening detection method |
| CN113222935A (en) * | 2021-05-13 | 2021-08-06 | 江苏科技大学 | Method for detecting looseness and pretightening force loss of steel bridge bolt |
| CN113469966A (en) * | 2021-06-25 | 2021-10-01 | 西南交通大学 | Train bolt looseness detection method based on anti-loosening line identification |
| US11714024B2 (en) | 2017-11-30 | 2023-08-01 | University Of Kansas | Vision-based fatigue crack detection using feature tracking |
| CN116580005A (en) * | 2023-05-15 | 2023-08-11 | 青岛环球重工科技有限公司 | Guiding method of duct piece mold opening system based on image processing and deep learning |
| CN117105038A (en) * | 2023-10-17 | 2023-11-24 | 山西戴德测控技术股份有限公司 | Elevator operation monitoring method, device, equipment and storage medium |
| US11954844B2 (en) | 2018-08-21 | 2024-04-09 | University Of Kansas | Fatigue crack detection in civil infrastructure |
Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN105787966A (en) * | 2016-03-21 | 2016-07-20 | 复旦大学 | An aesthetic evaluation method for computer pictures |
| WO2016123538A1 (en) * | 2015-01-29 | 2016-08-04 | Clear Research Corporation | Mobile visual commerce system |
| CN106446890A (en) * | 2016-10-28 | 2017-02-22 | 中国人民解放军信息工程大学 | Candidate area extraction method based on window scoring and superpixel segmentation |
-
2017
- 2017-05-02 CN CN201710301044.1A patent/CN107145905B/en active Active
Patent Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| WO2016123538A1 (en) * | 2015-01-29 | 2016-08-04 | Clear Research Corporation | Mobile visual commerce system |
| CN105787966A (en) * | 2016-03-21 | 2016-07-20 | 复旦大学 | An aesthetic evaluation method for computer pictures |
| CN106446890A (en) * | 2016-10-28 | 2017-02-22 | 中国人民解放军信息工程大学 | Candidate area extraction method based on window scoring and superpixel segmentation |
Non-Patent Citations (2)
| Title |
|---|
| MING-MING ZHANG 等: "《BING: Binarized Normed Gradients for Objectness Estimation at 300 fps》", 《2014 IEEE CONFERENCE ON COMPUTER VISION AND PATTERN RECOGNITION》 * |
| ZHIWEN FANG 等: "《Refine BING using effective cascade ranking》", 《APPLIED SOFT COMPUTING》 * |
Cited By (30)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN107657275A (en) * | 2017-09-20 | 2018-02-02 | 电子科技大学 | Based on the car plate pre-detection method for improving BING algorithms |
| CN107657275B (en) * | 2017-09-20 | 2020-07-21 | 电子科技大学 | License plate pre-detection method based on improved BING algorithm |
| CN107966270A (en) * | 2017-11-21 | 2018-04-27 | 韶关学院 | A kind of bolt based on sparse dictionary tightens detection method |
| US11714024B2 (en) | 2017-11-30 | 2023-08-01 | University Of Kansas | Vision-based fatigue crack detection using feature tracking |
| CN108416287A (en) * | 2018-03-04 | 2018-08-17 | 南京理工大学 | A kind of pedestrian detection method excavated based on omission negative sample |
| US11354814B2 (en) | 2018-03-23 | 2022-06-07 | University Of Kansas | Vision-based fastener loosening detection |
| WO2019183475A1 (en) * | 2018-03-23 | 2019-09-26 | University Of Kansas | Vision-based fastener loosening detection |
| CN108416336A (en) * | 2018-04-18 | 2018-08-17 | 特斯联（北京）科技有限公司 | A kind of method and system of intelligence community recognition of face |
| CN108416336B (en) * | 2018-04-18 | 2019-01-18 | 特斯联（北京）科技有限公司 | A kind of method and system of intelligence community recognition of face |
| CN110570472A (en) * | 2018-06-05 | 2019-12-13 | 成都精工华耀科技有限公司 | Fastener loosening detection method based on artificial marking image |
| CN109029382B (en) * | 2018-08-20 | 2023-08-08 | 上海矩尺土木科技有限公司 | Bolt looseness-prevention early warning monitoring device |
| CN109029382A (en) * | 2018-08-20 | 2018-12-18 | 上海矩尺土木科技有限公司 | A kind of screw retention early warning and monitoring device |
| US11954844B2 (en) | 2018-08-21 | 2024-04-09 | University Of Kansas | Fatigue crack detection in civil infrastructure |
| CN109389566A (en) * | 2018-10-19 | 2019-02-26 | 辽宁奇辉电子系统工程有限公司 | Subway height adjusting valve fastening nut defective mode detection method based on boundary characteristic |
| CN109389566B (en) * | 2018-10-19 | 2022-01-11 | 辽宁奇辉电子系统工程有限公司 | Method for detecting bad state of fastening nut of subway height adjusting valve based on boundary characteristics |
| CN110097536A (en) * | 2019-04-10 | 2019-08-06 | 东南大学 | Hexagon bolt looseness detection method based on deep learning and Hough transformation |
| CN110390668A (en) * | 2019-06-26 | 2019-10-29 | 石家庄铁道大学 | Bolt looseness detection method, terminal device and storage medium |
| CN110390668B (en) * | 2019-06-26 | 2022-02-01 | 石家庄铁道大学 | Bolt looseness detection method, terminal device and storage medium |
| CN110930366A (en) * | 2019-10-30 | 2020-03-27 | 同济大学 | Nut loosening detection method for wind power tower cylinder |
| CN111122135A (en) * | 2019-12-05 | 2020-05-08 | 西安交通大学 | Method for evaluating looseness degree of flange bolt connection structure |
| CN111666943A (en) * | 2020-04-20 | 2020-09-15 | 国网浙江省电力有限公司 | Matching detection method and system for bolt and nut of power transmission line based on image recognition |
| CN111814709A (en) * | 2020-07-14 | 2020-10-23 | 南京航空航天大学苏州研究院 | Airplane surface bolt tightening detection method |
| CN113222935A (en) * | 2021-05-13 | 2021-08-06 | 江苏科技大学 | Method for detecting looseness and pretightening force loss of steel bridge bolt |
| CN113222935B (en) * | 2021-05-13 | 2024-04-23 | 江苏科技大学 | Method for detecting looseness and pretightening force loss of steel bridge bolt |
| CN113469966B (en) * | 2021-06-25 | 2023-04-18 | 西南交通大学 | Train bolt looseness detection method based on anti-loosening line identification |
| CN113469966A (en) * | 2021-06-25 | 2021-10-01 | 西南交通大学 | Train bolt looseness detection method based on anti-loosening line identification |
| CN116580005A (en) * | 2023-05-15 | 2023-08-11 | 青岛环球重工科技有限公司 | Guiding method of duct piece mold opening system based on image processing and deep learning |
| CN116580005B (en) * | 2023-05-15 | 2024-03-08 | 青岛环球重工科技有限公司 | Guiding method of duct piece mold opening system based on image processing and deep learning |
| CN117105038A (en) * | 2023-10-17 | 2023-11-24 | 山西戴德测控技术股份有限公司 | Elevator operation monitoring method, device, equipment and storage medium |
| CN117105038B (en) * | 2023-10-17 | 2024-01-05 | 山西戴德测控技术股份有限公司 | Elevator operation monitoring method, device, equipment and storage medium |
Also Published As
| Publication number | Publication date |
|---|---|
| CN107145905B (en) | 2020-04-21 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN107145905A (en) | The image recognizing and detecting method that elevator fastening nut loosens | |
| CN104916163B (en) | Parking space detection method | |
| CN104282020B (en) | A kind of vehicle speed detection method based on target trajectory | |
| CN107330376B (en) | Lane line identification method and system | |
| Kong et al. | General road detection from a single image | |
| CN112819094B (en) | Target detection and identification method based on structural similarity measurement | |
| CN112418103B (en) | Bridge crane hoisting safety anti-collision system and method based on dynamic binocular vision | |
| CN107463890B (en) | A kind of Foregut fermenters and tracking based on monocular forward sight camera | |
| CN106156723B (en) | A kind of crossing fine positioning method of view-based access control model | |
| EP3196853A1 (en) | Machine vision-based method and system for aircraft docking guidance and aircraft type identification | |
| CN104298996B (en) | A kind of underwater active visual tracking method applied to bionic machine fish | |
| CN110414355A (en) | The right bit sky parking stall of view-based access control model and parking stall line detecting method during parking | |
| CN110647850A (en) | Automatic lane deviation measuring method based on inverse perspective principle | |
| Li et al. | Nighttime lane markings recognition based on Canny detection and Hough transform | |
| CN107392885A (en) | A kind of method for detecting infrared puniness target of view-based access control model contrast mechanism | |
| CN107169953A (en) | Bridge concrete surface crack detection method based on HOG features | |
| CN112465706A (en) | Automatic gate container residual inspection method | |
| CN105404874B (en) | A kind of vehicle window identifying system based on projection and hough straight-line detections | |
| CN106407924A (en) | Binocular road identifying and detecting method based on pavement characteristics | |
| CN106778551A (en) | A kind of fastlink and urban road Lane detection method | |
| CN113624225B (en) | Pose resolving method for mounting engine positioning pins | |
| CN107816943A (en) | A kind of box for material circulation volume weight measuring system and its implementation | |
| CN114758322B (en) | Road quality detection system based on machine identification | |
| CN101383005A (en) | Method for separating passenger target image and background by auxiliary regular veins | |
| CN111242888A (en) | Image processing method and system based on machine vision |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant | ||
| CP03 | Change of name, title or address | ||
| CP03 | Change of name, title or address |
Address after: 400030 No. 174 Sha Jie street, Shapingba District, Chongqing Patentee after: Chongqing University Patentee after: Lingwang Elevator Co.,Ltd. Address before: 400044 No. 174 Sha Jie street, Shapingba District, Chongqing Patentee before: Chongqing University Patentee before: WINONE ELEVATOR Co.,Ltd. |