CN107862675A - A kind of real-time vision detection method for electric bucket tooth missing - Google Patents

A kind of real-time vision detection method for electric bucket tooth missing Download PDF

Info

Publication number
CN107862675A
CN107862675A CN201710880249.XA CN201710880249A CN107862675A CN 107862675 A CN107862675 A CN 107862675A CN 201710880249 A CN201710880249 A CN 201710880249A CN 107862675 A CN107862675 A CN 107862675A
Authority
CN
China
Prior art keywords
bucket tooth
point
electric bucket
image
profile
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
Application number
CN201710880249.XA
Other languages
Chinese (zh)
Inventor
王日俊
董磊
段能全
王俊元
党长营
曾志强
杜文华
段宇秀
陈立
王俊凤
刘东曜
崔铮
薛亮
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
North University of China
Original Assignee
North University of China
Priority date (The priority date 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 date listed.)
Filing date
Publication date
Application filed by North University of China filed Critical North University of China
Priority to CN201710880249.XA priority Critical patent/CN107862675A/en
Publication of CN107862675A publication Critical patent/CN107862675A/en
Pending legal-status Critical Current

Links

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T7/00—Image analysis
    • G06T7/0002—Inspection of images, e.g. flaw detection
    • G06T7/0004—Industrial image inspection
    • G06T7/0008—Industrial image inspection checking presence/absence
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00—Pattern recognition
    • G06F18/20—Analysing
    • G06F18/22—Matching criteria, e.g. proximity measures
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00—Pattern recognition
    • G06F18/20—Analysing
    • G06F18/24—Classification techniques
    • G06F18/241—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches
    • G06F18/2411—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches based on the proximity to a decision surface, e.g. support vector machines
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
    • G06V10/00—Arrangements for image or video recognition or understanding
    • G06V10/40—Extraction of image or video features
    • G06V10/44—Local feature extraction by analysis of parts of the pattern, e.g. by detecting edges, contours, loops, corners, strokes or intersections; Connectivity analysis, e.g. of connected components
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • 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
    • G06V10/50—Extraction of image or video features by performing operations within image blocks; by using histograms, e.g. histogram of oriented gradients [HoG]; by summing image-intensity values; Projection analysis
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T2207/00—Indexing scheme for image analysis or image enhancement
    • G06T2207/30—Subject of image; Context of image processing
    • G06T2207/30108—Industrial image inspection
    • G06T2207/30164—Workpiece; Machine component

Abstract

The invention belongs to mining technique field, specifically a kind of real-time vision detection method for electric bucket tooth missing.Comprise the following steps：Ith, the positive and negative samples of one group of electric bucket tooth are shot in advance；IIth, positive and negative samples HOG features are extracted；IIIth, HOG features are input in SVM grader and trained, obtain decision function；IVth, the picture of live captured in real-time is inputted, HOG feature extractions are carried out by detection window, then by svm classifier, complete the Preliminary detection to electric bucket tooth；Vth, accurate detection of the SC constraint completions to electric bucket tooth is carried out to the electric bucket tooth testing result of acquisition；Comprise the following steps that, a. image preprocessings and contours extract；B. SC features are calculated；C. shape similarity calculates；VIth, the accurate testing result of electric bucket tooth is compared with number of the electric bucket tooth set in advance without bucket tooth when lacking, judges whether electric bucket tooth lacks.Instant invention overcomes Conventional visual detection method because accuracy rate is low caused by false retrieval, missing inspection the problem of, reduces rate of false alarm.

Description

A kind of real-time vision detection method for electric bucket tooth missing
Technical field
The invention belongs to mining technique field, specifically a kind of real-time vision detection method for electric bucket tooth missing.
Background technology
When power shovel is in digging operation, because of the interphase interaction of electric bucket tooth and material, electric bucket tooth is sheared and squeezed Active force is pressed, and bears certain shock loading, the electric bucket tooth being fixed on originally on toothholder on scraper bowl is easily broken off or loosened Come off.
After electric bucket tooth comes off, frictional resistance during power shovel operation is not only increased, causes scraper bowl to be cut when excavating material Difficulty, accelerate the abrasion of scraper bowl, influence the service life of scraper bowl；What is more important power shovel made of potassium steel or steel alloy Bucket tooth, hardness are far above ore, if hard electric bucket tooth together enters in stone crusher with ore materials, will cause rubble Machine damages, and destroys whole digging --- crushing production line, and the entire production line maintenance difficulty is big, time-consuming, costly, once it is raw Producing line is stopped work, and will cause huge economic loss；What is more important, the electric bucket tooth to come off are likely to draw into stone crusher Send out stone crusher mechanical accident, the serious personal safety for even influencing whether operating personnel.And mine excavator is bulky, work Make environment complexity, be not easy to the state that excavator operating personnel monitor electric bucket tooth by way of manually observing.
Being come off context of detection in electric bucket tooth, domestic and foreign scholars and relevant enterprise have done substantial amounts of research-and-development activity, And form the product of correlation.2016, the electric bucket tooth based on sample that Ser Nam Lim etc. have studied complete set came off Vision detection system, it extracts the image pattern of power shovel scraper bowl in motion process first, is first positioned using these sample informations The approximate location of electric bucket tooth, then result is modified using frame difference method and optical flow method, comprehensively utilizes template again afterwards Match with the method for tooth trace fitting to be accurately positioned electric bucket tooth target.Finally, it is special with reference to the related gray scale of electric bucket tooth image Sign just may determine that whether electric bucket tooth comes off.But from the point of view of the service condition of various regions, there is the ratio of rate of false alarm still Compare high.
The content of the invention
In view of the above-mentioned deficiencies in the prior art, the technical problem to be solved by the present invention is that providing one kind is used for power shovel The real-time vision detection method of bucket tooth missing.
The present invention takes following technical scheme：A kind of real-time vision detection method for electric bucket tooth missing, the side Method can detect the missing of bucket tooth in real time in power shovel operation process, and bucket tooth target, realization pair are identified by using SVM combinations SC The detection of electric bucket tooth missing, comprises the following steps：
Ith, the positive and negative samples of one group of electric bucket tooth are shot in advance；
IIth, positive and negative samples HOG features are extracted；
IIIth, HOG features are input in SVM grader and trained, obtain decision function；
IVth, the picture of live captured in real-time is inputted, HOG feature extractions are carried out by detection window, then by svm classifier, Complete the Preliminary detection to electric bucket tooth；
Vth, accurate detection of the SC constraint completions to electric bucket tooth is carried out to the electric bucket tooth testing result of acquisition；Specific step Rapid as follows, a. image preprocessings and contours extract；B. SC features are calculated；C. shape similarity calculates；
VIth, by the accurate testing result of electric bucket tooth compared with electric bucket tooth set in advance is without the number of bucket tooth when lacking Compared with judging whether electric bucket tooth lacks.
The positive and negative samples of electric bucket tooth are shot in the step I in advance, cover electric bucket tooth in all works of operative scenario Make situation, and preset number of the electric bucket tooth without bucket tooth when coming off, the positive and negative samples of monodentate should be obtained during shooting.
The step II includes：
Step 1, the positive and negative samples to one group of electric bucket tooth of collection carry out image preprocessing, and image preprocessing includes ash Degreeization and Gamma are corrected；
Step 2, the image gradient for calculating electric bucket tooth positive and negative samples；
Step 3, gradient direction statistic histogram is calculated, obtain HOG features.
The gray scale for the image that the main purpose of Gamma corrections is correction in the case where image is excessively bright or excessively dark, to increase Strong contrast, reduces the influence to image of illumination variation and shade, and in step II, Gamma correction calculation formula (1) are as follows
Y (x, y)=I (x, y)γFormula (1)
Wherein I (x, y) represents gray scale of the input picture in point (x, y) place pixel, and Y (x, y) represents the pixel after correction The gray value of point, it is contemplated that the noise of camera sensor is directly proportional to the square root of illumination, therefore takes γ=0.5, with maximum limit The reduction noise of degree, and improve the contrast of image.
The calculation formula of the image gradient size and Orientation for the electric bucket tooth positive and negative samples extracted is as follows：
Gx(x, y)=H (x+1, y)-H (x-1, y) formula (2)
Gy(x, y)=H (x, y+1)-H (x, y-1) formula (3)
Wherein GX(x,y)、GY(x, y) represents the Grad of x directions and y directions epigraph respectively, and H (x, y) is represented in image The gray value of certain point, G (x, y) represent the size of image point gradient vector, and α (x, y) represents the direction of this gradient vector.
Gradient direction statistic histogram is calculated, obtains HOG features, specific implementation method comprises the following steps：
A) cell factory cell size is selected, because detection window pixel size is 64 × 128, herein by detection window point For 128 cell factories, the size of each cell factory is 8 × 8 pixels.
B) each cell factory inside gradient direction histogram is calculated, computational methods are as follows：
The range set of gradient direction is arrived into 180 degree for 0, and gradient direction is divided into 9 sections, then counts the cell factory The number that different gradient directions occur in interior each pixel, gradient direction occurs on each each pixel of cell factory is counted , it is necessary to be multiplied by a weight coefficient to each pixel during number, the size that the weight coefficient is each pixel gradient, example are taken Such as, the gradient direction of some pixel is 10 degree, and its gradient magnitude is 1.5, then the number that this o'clock occurs in 0 to 20 degree sections is 1 × 1.5 time.The calculation formula (6) of each pixel occurrence number in each gradient direction section is as follows：
Wherein, α (x, y) be cell factory in certain point gradient direction, binkFor k-th section, Vk(x, y) represents picture The number that vegetarian refreshments occurs in k-th gradient direction section.
By above-mentioned calculating process, in each cell factory, by the gradient magnitude of each pixel according to different gradients Direction is counted, and the HOG characteristic vectors of the cell factory are precisely formed.
C) several adjacent cell factories are formed into a description block block, to cell factory all in description block Statistic histogram combination of eigenvectors form the characteristic vector of description block together, by that analogy, calculate all description blocks HOG characteristic vectors.
D) histogram in description block is normalized, calculation formula (7) is as follows：
Wherein v represents characteristic vector, and ε represents the number of a very little, can be taken as 0.000001, it is therefore an objective to prevent divisor For 0 situation, the characteristic vector of all description blocks, which joins end to end, after normalization can form HOG features.
The calculation procedure of the step III is as follows：
1) sample set { x of given training is seti,yi, i=1,2...n, x ∈ Rn, y ∈ { ± 1 }, x represents sample in formula Characteristic vector, y represent category label, if required optimal hyperlane is：H:wTX+b=0；
2) Lagrange multiplier α value is calculated, calculation formula is as follows：
Bring α values into formula (9), formula (10) can obtain hyperplane ω, b optimal solution；
WhereinTherefore optimal classification plane ω is obtained*·x+b*=0, and optimal decision function is：
The step IV includes：
Step 1) setting camera captures image temporal interval, the working region real-time grasp shoot figure using camera to electric bucket tooth Picture, and by the image transmitting captured to computer, to read image that camera captures in real time and to be pre-processed；
Step 2) is detected to the every piece image captured, and the detection window used is being schemed according to the step-length of 10 pixels Slided as in until covering entire image, often slides the HOG features for once just extracting image in detection window, then, by positive and negative The decision function that template is obtained by SVM classifier training is judged characteristic vector, to determine the image in window Whether it is electric bucket tooth target, and will determine that result is identified in the picture for positive window, the like, wait slip to detect The window for the expression same target that overlaps is combined again after end, completes the Preliminary detection to electric bucket tooth.
The step V includes：
A. the image in detection window is pre-processed and extracts profile, comprised the following steps：
1) bianry image is converted the image into, calculation formula (12) is as follows：
Wherein, T1、T2For threshold value set in advance, scope is 0~255.
2) boundary tracking algorithm is used, extracts objective contour, realizes that step is combined shown in figure (8), grey parts table in figure Show the pixel that gray value is 1, white portion represents the pixel that gray value is 0, during calculating, from upper left side point b0Start, from Left-to-right, the point b that gray value is 1 is found from top to bottom0And its left dorsal sight spot c0, from c0Start along b08 neighborhood directions Clockwise search is carried out, until running into the point b that next gray value is 11, this seasonal c1Equal to b1That 8 neighborhood before Middle background dot, by that analogy, until correcting action, it accurately can obtain the profile of bianry image.
B. SC features are calculated, are comprised the following steps：
1) by extracting profile and the point on profile being sampled, a position for containing profile up-sampling point is obtained Data set P={ the p of information1,p2...pn, some point p that first selected point is concentrated1(x1,y1) polar limit is used as, calculate Remaining n-1 point pn(xn,yn) and p1(x1,y1) vectorial size and Orientation is formed, similarly, then with other point works in set P Angle and distance information is calculated successively for polar origin, the matrix R and A of accurate available two n × (n-1) size, its The distance and angle information of all sampled points on the profile are stored, calculation formula is as follows：
2) all range information r are normalized using experience densimetry, and to the r after normalizationnorCarry out Logarithmic transformation, each point (x, y) in rectangular coordinate system is accurately mapped to log-polar (logr, θ) in, normalization is public Formula (15) is as follows：
3) polar coordinate system as shown in Figure 10, is divided into m × n subinterval, wherein m is distance parameter, and n joins for angle Number, the number for falling into profile point in each subinterval (dash area in such as Figure 10) is counted afterwards, so as to obtain SC statistics Nogatas Figure；Similarly, respectively using difference as origin, the SC features of whole profile can be calculated.
Calculation formula (16) is as follows：
hi(k)={ q ≠ pi:(q-pi) ∈ bin (k) formula (16)
Wherein k=1,2 ... K, K=m × n, represent k-th of component in histogram, PiRepresent that point concentration is used as pole The point of coordinate limit, q represent other points, (q-pi) ∈ bin (k) represent relative to pi, point q belongs to k-th of subinterval；
C. shape similarity calculates, and formula (17) is as follows：
Wherein piRepresent the SC histograms using point i as origin, q on a profile pjRepresent on another profile q using point j as The SC histograms of origin；Cost=C (pi,qj) represent the distance between the two statistic histograms, that is, of two profiles With cost；hiAnd h (k)j(k) two shape histogram p are represented respectivelyiAnd qjThe value of middle kth level, it is known that Ci,jSpan exist Between 0 to 1, the similarity of its value and two profiles is inversely proportional, i.e. Ci,jSmaller, two profiles are more similar, Ci,jIt is bigger, two wheels It is wide more dissimilar.
According to formula (17), a point P on profile p is first calculated1With on sample profile q Matching power flow a little, i.e., All points on the upper point of profile p and profile q are matched, the matrix of 1 × n size can be obtained, retain this square The minimum value of element is as point P in battle array1Matching power flow, have found P equivalent on profile q1The point most matched, with identical Method calculates all-pair on profile p and, in q Matching power flow, accurately can equally obtain the matrix of 1 × n size, calculate The sum of whole elements in this matrix, a single non-vector index is accurately can obtain, in order to describe the two profiles Between similitude.
In the step VI, judge that the method whether electric bucket tooth lacks is as follows, for power shovel, bucket tooth on its scraper bowl Number be changeless, therefore can judge whether bucket tooth takes off as an index according to this invariant of bucket tooth number Fall, if a known scraper bowl has N number of bucket tooth, if bucket tooth does not come off, then the number M for the bucket tooth being detected in the picture N should be equal to, only need to compare M and N difference it may determine that whether bucket tooth comes off.
Compared with prior art, the present invention is completed to the Preliminary detection of electric bucket tooth simultaneously by classification of the SVM to HOG features Using constraints of the SC to Preliminary detection result, realize the accurate detection to electric bucket tooth, overcome Conventional visual detection method because The problem of accuracy rate caused by false retrieval, missing inspection is low, reduces rate of false alarm.
Brief description of the drawings
Fig. 1 is electric bucket tooth target detection flow chart；
Fig. 2 is HOG feature calculation flow charts；
Fig. 3 is form fit algorithm flow chart；
Fig. 4 is bucket tooth positive and negative samples collection schematic diagram
Fig. 5 is the HOG features and angular divisions schematic diagram of individual cells unit；
Fig. 6 is cell factory and block division signal；
Fig. 7 is the visualization HOG features of bucket tooth；
Fig. 8 is bianry image contour-tracking algorithm；
Fig. 9 is the binaryzation result and Contour tracing result of bucket tooth；
Figure 10 is distance and angle calculation.
Embodiment
The embodiment of the present invention is described in further detail below in conjunction with the accompanying drawings.
As shown in figure 1, Fig. 1 is the system basic framework figure of the electric bucket tooth detection method based on SVM and SC；Based on SVM Trained, obtained by extracting the HOG features of positive and negative samples and inputting in SVM grader with SC electric bucket tooth detection method Decision function, HOG feature extractions are carried out to the picture of live captured in real-time, and HOG is classified using decision function, to reach pair The Preliminary detection of electric bucket tooth, and row constraint is entered to the Preliminary detection result of electric bucket tooth with reference to SC, complete to electric bucket tooth Accurate detection, the number of detected electric bucket tooth is obtained, pass through electric bucket tooth number more set in advance and accurate inspection The number of the electric bucket tooth measured, you can judge whether electric bucket tooth lacks.Illustrate：
A kind of real-time vision detection method for electric bucket tooth missing, this method are special by the HOG for extracting positive and negative samples Levy and be input in SVM grader and train, obtain decision function and to being carried out to the picture to live captured in real-time to power shovel The Preliminary detection of bucket tooth, and row constraint is entered to the Preliminary detection result of electric bucket tooth with reference to SC, reach to the accurate of electric bucket tooth Detection.Comprise the following steps：
Step I, the positive and negative samples of electric bucket tooth are shot, wherein, for positive sample, it is desirable to which positive sample should be as far as possible Only comprising bucket tooth in itself；And for negative sample, then required without too many, it is any to choose an equal amount of region in background image , while require that the quantity of negative sample will be more than the quantity of positive sample.And electric bucket tooth is preset without bucket tooth when coming off Number.
Step II, the HOG features of electric bucket tooth positive and negative samples are extracted, are comprised the following steps：
Step 1, in view of shooting photo when, the picture quality caused by the influence of various factors is low, causes detection accurate The problem of really rate is low, the picture that need to be shot to camera pre-process, including gray processing and Gamma corrections.
The gray scale for the image that the main purpose of Gamma corrections is correction in the case where image is excessively bright or excessively dark, to increase Strong contrast, reduces the influence of illumination variation and shade to image, and calculation formula (1) is as follows
Y (x, y)=I (x, y)γFormula (1)
Wherein I (x, y) represents gray scale of the input picture in point (x, y) place pixel, and Y (x, y) represents the pixel after correction The gray value of point.It is directly proportional in view of the noise of camera sensor and the square root of illumination, therefore γ=0.5 is taken, with maximum limit The reduction noise of degree, and improve the contrast of image.
Step 2, the gradient magnitude for calculating extracted electric bucket tooth positive and negative samples and direction, calculation formula formula are as follows：
Gx(x, y)=H (x+1, y)-H (x-1, y) formula (2)
Gy(x, y)=H (x, y+1)-H (x, y-1) formula (3)
Wherein GX(x,y)、GY(x, y) represents the Grad of x directions and y directions epigraph respectively, and H (x, y) is represented in image The gray value of certain point, G (x, y) represent the size of image point gradient vector, and α (x, y) represents the direction of this gradient vector.
Step 3, gradient direction statistic histogram is calculated, obtain HOG features, specific implementation method comprises the following steps：
A) size of cell factory (cell) is selected, because detection window pixel size is 64 × 128, herein by detection window It is divided into 128 cell factories, the size of each cell factory is 8 × 8 pixels；
B) each cell factory inside gradient direction histogram is calculated, computational methods are as follows：
The range set of gradient direction is arrived into 180 degree for 0, and gradient direction is divided into 9 sections, then counts the cell factory The number that different gradient directions occur in interior each pixel, gradient direction occurs on each each pixel of cell factory is counted , it is necessary to be multiplied by a weight coefficient to each pixel during number, the size that the weight coefficient is each pixel gradient, example are taken Such as, the gradient direction of some pixel is 10 degree, and its gradient magnitude is 1.5, then the number that this o'clock occurs in 0 to 20 degree sections is 1 × 1.5 time.The calculation formula (6) of each pixel occurrence number in each gradient direction section is as follows：
Wherein, α (x, y) be cell factory in certain point gradient direction, binkFor k-th section, Vk(x, y) represents picture The number that vegetarian refreshments occurs in k-th gradient direction section.
By above-mentioned calculating process, in each cell factory, by the gradient magnitude of each pixel according to different gradients Direction is counted, and the HOG characteristic vectors of the cell factory are precisely formed.
C) several adjacent cell factories are formed into a description soon (block), to cell list all in description block The statistic histogram combination of eigenvectors of member forms the characteristic vector of description block together, by that analogy, calculates all description blocks HOG characteristic vectors；
4) histogram in description block is normalized, calculation formula (7) is as follows：
Wherein v represents characteristic vector, and ε represents the number of a very little, it is therefore an objective to prevents the situation that divisor is 0.Normalizing The characteristic vector of all description blocks, which joins end to end, after change can form HOG features.
Step III, HOG features are input in SVM grader and trained, obtain the decision function of this feature, calculate step It is rapid as follows：
1) sample set { x of given training is seti,yi, i=1,2...n, x ∈ Rn, y ∈ { ± 1 }, x represents sample in formula Characteristic vector, y represent category label, if required optimal hyperlane is：H:wTX+b=0
2) Lagrange multiplier α value is calculated, calculation formula (8) is as follows：
Bring α values into formula (9), formula (10) can obtain hyperplane ω, b optimal solution
WhereinTherefore optimal classification plane ω is obtained*·x+b*=0, and optimal decision function is：
Step IV, the picture of live captured in real-time is inputted, HOG feature extractions are carried out by detection window, then pass through SVM points Class, the Preliminary detection to electric bucket tooth is completed, is comprised the following steps：
Step 1) setting camera captures image temporal interval, and the working region of electric bucket tooth is grabbed in real time using infrared camera Image is clapped, and by the image transmitting captured into computer specified folder, to read the image that camera is captured in real time And pre-processed.
After step 2) pre-processes to image, detection window is slided until covering in the picture according to certain step-length Lid entire image, the HOG characteristic vectors for once just extracting image in window are often slided, then, by the decision-making letter obtained by training Count to judge characteristic vector, to determine whether the image in window is electric bucket tooth target, if bucket tooth target, then sentence Disconnected result will determine that result is identified in the picture for positive window to be otherwise negative just, the like, wait slip to examine Survey after terminating and be again combined the window for the expression same target that overlaps, complete the Preliminary detection to electric bucket tooth.
Step V, using the method for form fit, compare similar between the image in detection window and bucket tooth template samples Property degree, is added to SVM testing results, to the electric bucket tooth of acquisition using this similarity evaluation index as a kind of constraints Testing result carries out SC constraints, completes the accurate detection to electric bucket tooth；
A. the image in detection window is pre-processed and extracts profile, comprised the following steps：
1) bianry image is converted the image into, calculation formula (12) is as follows：
Wherein, T1、T2For threshold value set in advance, scope is 0~255.
2) boundary tracking algorithm is used, extracts objective contour, realizes that step is combined shown in figure (8), grey parts table in figure Show the pixel that gray value is 1, white portion represents the pixel that gray value is 0, during calculating, from upper left side point b0Start, from Left-to-right, the point b that gray value is 1 is found from top to bottom0And its left dorsal sight spot c0, from c0Start along b08 neighborhood directions Carry out clockwise search, it is known that run into the point b that next gray value is 11, this seasonal c1Equal to b1That 8 neighborhood before Middle background dot, by that analogy, until correcting action, it accurately can obtain the profile of bianry image.
B. SC features are calculated, are comprised the following steps：
1) by extracting profile and the point on profile being sampled, one can be obtained and contain profile up-sampling point Data set P={ the p of positional information1,p2...pn, some point p that first selected point is concentrated1(x1,y1) polar limit is used as, Calculate remaining n-1 point pn(xn,yn) and p1(x1,y1) vectorial size and Orientation is formed, similarly, then with other in set P Point calculates angle and distance information as polar origin successively, the matrix R of accurate available two n × (n-1) size and A, which stores the distance and angle information of all sampled points on the profile.Calculation formula is as follows：
2) to ensure that point as much as possible all falls in a coordinate system, all range information r are entered using experience densimetry Row normalized, and to the r after normalizationnorLogarithmic transformation is carried out, accurately by each point (x, y) in rectangular coordinate system It is mapped to log-polar (logr, θ) in, normalization formula (15) is as follows：
3) polar coordinate system as shown in Figure 10, is divided into m × n subinterval, wherein m is distance parameter, and n joins for angle Number, the number for falling into profile point in each subinterval (dash area in such as Figure 10) is counted afterwards, so as to obtain SC statistics Nogatas Figure.Similarly, respectively using difference as origin, the SC features of whole profile can be calculated.Calculation formula (16) is as follows：
hi(k)={ q ≠ pi:(q-pi) ∈ bin (k) formula (16)
Wherein k=1,2 ... K, K=m × n, represent k-th of component in histogram, PiRepresent that point concentration is used as pole The point of coordinate limit, q represent other points, (q-pi) ∈ bin (k) represent relative to pi, point q belongs to k-th of subinterval.
C. shape similarity calculates, under formula (17) enters：
Wherein piRepresent the SC histograms using point i as origin, q on a profile pjRepresent on another profile q using point j as The SC histograms of origin；Cost=C (pi,qj) represent the distance between the two statistic histograms, that is, of two profiles With cost；hiAnd h (k)j(k) two shape histogram p are represented respectivelyiAnd qjThe value of middle kth level.Understand Ci,jSpan exist Between 0 to 1, the similarity of its value and two profiles is inversely proportional, i.e. Ci,jSmaller, two profiles are more similar, Ci,jIt is bigger, two wheels It is wide more dissimilar.
According to formula (17), a point P on profile p is first calculated1With on sample profile q Matching power flow a little, i.e., All points on the upper point of profile p and profile q are matched, the matrix of 1 × n size can be obtained, retain this square The minimum value of element is as point P in battle array1Matching power flow, have found P equivalent on profile q1The point most matched.With identical Method calculates all-pair on profile p and, in q Matching power flow, accurately can equally obtain the matrix of 1 × n size, calculate The sum of whole elements in this matrix, a single non-vector index is accurately can obtain, in order to describe the two profiles Between similitude.
Step VI, for power shovel, the number of bucket tooth is changeless on its scraper bowl, therefore can be according to bucket tooth This invariant of number judges whether bucket tooth comes off as an index, if a known scraper bowl has N number of bucket tooth, if bucket tooth does not have Have and come off, then the number M for the bucket tooth being detected in the picture should be equal to N, and the difference can that need to only compare M and N is sentenced Whether disconnected bucket tooth comes off.
Finally it should be noted that：Above example is merely to illustrate the technical scheme of the application rather than to its protection domain Limitation, although the application is described in detail with reference to above-described embodiment, the those of ordinary skill in the field should Understand：Those skilled in the art can still carry out a variety of changes after reading the application to the embodiment of application, change or wait With replacement, but these changes, modification or equivalent substitution, applying within pending claims.

Claims (8)

  1. A kind of 1. real-time vision detection method for electric bucket tooth missing, it is characterised in that：Comprise the following steps：
    Ith, the positive and negative samples of one group of electric bucket tooth are shot in advance；
    IIth, positive and negative samples HOG features are extracted；
    IIIth, HOG features are input in SVM grader and trained, obtain decision function；
    IVth, the picture of live captured in real-time is inputted, HOG feature extractions are carried out by detection window, then by svm classifier, complete To the Preliminary detection of electric bucket tooth；
    Vth, SC constraints are carried out to the electric bucket tooth testing result of acquisition, completes the accurate detection to electric bucket tooth；Specific steps are such as Under, a. image preprocessings and contours extract；B. SC features are calculated；C. shape similarity calculates；
    VIth, the accurate testing result of electric bucket tooth is compared with number of the electric bucket tooth set in advance without bucket tooth when lacking, Judge whether electric bucket tooth lacks.
  2. 2. the real-time vision detection method according to claim 1 for electric bucket tooth missing, it is characterised in that：The step Positive and negative samples of shooting electric bucket tooth in advance in rapid I, and preset number of the electric bucket tooth without bucket tooth when coming off, during shooting The positive and negative samples of monodentate should be obtained.
  3. 3. the real-time vision detection method according to claim 2 for electric bucket tooth missing, it is characterised in that：The step Rapid II includes：
    Step 1, the positive and negative samples to one group of electric bucket tooth of collection carry out image preprocessing, and image preprocessing includes gray processing Corrected with Gamma；
    Step 2, the image gradient for calculating electric bucket tooth positive and negative samples；
    Step 3, gradient direction statistic histogram is calculated, obtain HOG features.
  4. 4. the real-time vision detection method according to claim 3 for electric bucket tooth missing, it is characterised in that：The step In rapid II, Gamma correction calculation formula (1) are as follows
    Y (x, y)=I (x, y)γFormula (1)
    Wherein I (x, y) represents gray scale of the input picture in point (x, y) place pixel, and Y (x, y) represents the pixel after correcting Gray value, take γ=0.5；
    The calculation formula of the image gradient size and Orientation for the electric bucket tooth positive and negative samples extracted is as follows：
    Gx(x, y)=H (x+1, y)-H (x-1, y) formula (2)
    Gy(x, y)=H (x, y+1)-H (x, y-1) formula (3)
    Wherein GX(x,y)、GY(x, y) represents the Grad of x directions and y directions epigraph respectively, and H (x, y) represents certain point in image Gray value, G (x, y) represents the size of image point gradient vector, and α (x, y) represents the direction of this gradient vector；
    Gradient direction statistic histogram is calculated, obtains HOG features, specific implementation method comprises the following steps：
    A) cell factory cell size is selected, because detection window pixel size is 64 × 128, is herein divided into detection window 128 cell factories, the size of each cell factory is 8 × 8 pixels；
    B) each cell factory inside gradient direction histogram is calculated, computational methods are as follows：
    The range set of gradient direction is arrived into 180 degree for 0, and gradient direction is divided into 9 sections, then counts every in the cell factory The number that different gradient directions occur in individual pixel, the number that gradient direction occurs on each each pixel of cell factory is counted When, it is necessary to be multiplied by a weight coefficient to each pixel, take the size that the weight coefficient is each pixel gradient, each pixel Point calculation formula (6) of occurrence number in each gradient direction section is as follows：
    Wherein, α (x, y) be cell factory in certain point gradient direction, binkFor k-th section, Vk(x, y) represents pixel The number occurred in k-th gradient direction section；
    By above-mentioned calculating process, in each cell factory, by the gradient magnitude of each pixel according to different gradient directions Counted, the HOG characteristic vectors of the cell factory are precisely formed；
    C) several adjacent cell factories are formed into a description block block, to the system of cell factory all in description block Meter histogram feature vector is grouped together into the characteristic vector of description block, and by that analogy, the HOG for calculating all description blocks is special Sign vector；
    D) histogram in description block is normalized, calculation formula (7) is as follows：
    L2-norm：
    Wherein v represents characteristic vector, and ε represents the number of a very little, is taken as 0.000001, the feature of all description blocks after normalization Vector, which joins end to end, can form HOG features.
  5. 5. the real-time vision detection method according to claim 4 for electric bucket tooth missing, it is characterised in that：The step Rapid III calculation procedure is as follows：
    1) sample set { x of given training is seti,yi, i=1,2...n, x ∈ Rn, y ∈ { ± 1 }, x represents the spy of sample in formula Sign vector, y represents category label, if required optimal hyperlane is：H:wTX+b=0；
    2) Lagrange multiplier α value is calculated, calculation formula is as follows：
    Bring α values into formula (9), formula (10) can obtain hyperplane ω, b optimal solution；
    WhereinTherefore optimal classification plane ω is obtained*·x+b*=0, and optimal decision function is：
  6. 6. the real-time vision detection method according to claim 5 for electric bucket tooth missing, it is characterised in that：The step Rapid IV includes：
    Step 1) setting camera candid photograph image temporal interval, the working region real-time grasp shoot image using camera to electric bucket tooth, And by the image transmitting captured to computer；
    Step 2) detects to the every piece image captured, the detection window used according to 10 pixels step-length in the picture Slide until covering entire image, often slides the HOG features for once just extracting image in detection window, then, by positive and negative template The decision function obtained by SVM classifier training is judged characteristic vector, whether to determine the image in window For electric bucket tooth target, and it will determine that result is identified in the picture for positive window, the like, wait slip detection to terminate The window for the expression same target that overlaps is combined again afterwards, completes the Preliminary detection to electric bucket tooth.
  7. 7. the real-time vision detection method for being used for electric bucket tooth and lacking stated according to claim 6, it is characterised in that：The step V includes：
    A. the image in detection window is pre-processed and extracts profile, comprised the following steps：
    1) bianry image is converted the image into, calculation formula (12) is as follows：
    Wherein, T1、T2For threshold value set in advance, scope is 0~255；
    2) boundary tracking algorithm is used, extracts objective contour；
    B. SC features are calculated, are comprised the following steps：
    1) by extracting profile and the point on profile being sampled, a positional information for containing profile up-sampling point is obtained Data set P={ p1,p2...pn, some point p that first selected point is concentrated1(x1,y1) polar limit is used as, calculate remaining N-1 point pn(xn,yn) and p1(x1,y1) vectorial size and Orientation is formed, similarly, then other points are used as pole using in set P Origin calculates angle and distance information successively, and the matrix R and A of accurate available two n × (n-1) size, it is stored The distance and angle information of all sampled points, calculation formula are as follows on the profile：
    2) all range information r are normalized using experience densimetry, and to the r after normalizationnorCarry out logarithm Conversion, is accurately mapped to log-polar (log by each point (x, y) in rectangular coordinate systemr, θ) in, normalize formula (15) it is as follows：
    3) polar coordinate system is divided into m × n subinterval, wherein m is distance parameter, and n is angle parameter, and statistics falls into every afterwards The number of profile point in individual subinterval, so as to obtain SC statistic histograms；Similarly, respectively using difference as origin, can calculate The SC features of whole profile；
    Calculation formula (16) is as follows：
    hi(k)={ q ≠ pi:(q-pi) ∈ bin (k) formula (16)
    Wherein k=1,2 ... K, K=m × n, represent k-th of component in histogram, PiRepresent that point concentration is used as polar coordinates pole The point of point, q represent other points, (q-pi) ∈ bin (k) represent relative to pi, point q belongs to k-th of subinterval；
    C. shape similarity calculates, and formula (17) is as follows：
    Wherein piRepresent the SC histograms using point i as origin, q on a profile pjRepresent on another profile q using point j as origin SC histograms；Cost=C (pi, qj) represent the distance between the two statistic histograms, that is, the matching generation of two profiles Valency；hiAnd h (k)j(k) two shape histogram p are represented respectivelyiAnd qjThe value of middle kth level, it is known that Ci,jSpan 0 to 1 Between, the similarity of its value and two profiles is inversely proportional, i.e. Ci,jSmaller, two profiles are more similar, Ci,jBigger, two profiles are got over It is dissimilar；
    According to formula (17), a point P on profile p is first calculated1With on sample profile q Matching power flow a little, i.e., by profile All points on the upper point of p and profile q are matched, and can obtain the matrix of 1 × n size, are retained first in this matrix The minimum value of element is as point P1Matching power flow, have found P equivalent on profile q1The point most matched, is counted in the same way Calculating all-pair on profile p, in q Matching power flow, accurately can equally obtain the matrix of 1 × n size, calculate this square The sum of whole elements in battle array, a single non-vector index is accurately can obtain, in order to describe phase between the two profiles Like property.
  8. 8. the real-time vision detection method for being used for electric bucket tooth and lacking stated according to claim 7, it is characterised in that：The step In VI, judge that the method whether electric bucket tooth lacks is as follows, if a known scraper bowl has N number of bucket tooth, if bucket tooth does not come off, The number M for the bucket tooth being detected in the picture should be equal to N, only need to compare M and N difference it may determine that whether bucket tooth takes off Fall.
CN201710880249.XA 2017-09-25 2017-09-25 A kind of real-time vision detection method for electric bucket tooth missing Pending CN107862675A (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN201710880249.XA CN107862675A (en) 2017-09-25 2017-09-25 A kind of real-time vision detection method for electric bucket tooth missing

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN201710880249.XA CN107862675A (en) 2017-09-25 2017-09-25 A kind of real-time vision detection method for electric bucket tooth missing

Publications (1)

Publication Number Publication Date
CN107862675A true CN107862675A (en) 2018-03-30

Family

ID=61698717

Family Applications (1)

Application Number Title Priority Date Filing Date
CN201710880249.XA Pending CN107862675A (en) 2017-09-25 2017-09-25 A kind of real-time vision detection method for electric bucket tooth missing

Country Status (1)

Country Link
CN (1) CN107862675A (en)

Cited By (6)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN111429026A (en) * 2020-04-14 2020-07-17 西安热工研究院有限公司 Method for evaluating performance of electric shovel of strip mine
CN111764456A (en) * 2020-06-16 2020-10-13 武汉理工大学 Intelligent monitoring and alarming device and method for dropping of bucket teeth of forklift
US20210262204A1 (en) * 2018-06-01 2021-08-26 Motion Metrics International Corp. Method, apparatus and system for monitoring a condition associated with operating heavy equipment such as a mining shovel or excavator
US11669956B2 (en) 2021-06-01 2023-06-06 Caterpillar Inc. Ground engaging tool wear and loss detection system and method
US11821177B2 (en) 2021-02-09 2023-11-21 Caterpillar Inc. Ground engaging tool wear and loss detection system and method
US11869331B2 (en) 2021-08-11 2024-01-09 Caterpillar Inc. Ground engaging tool wear and loss detection system and method

Cited By (7)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20210262204A1 (en) * 2018-06-01 2021-08-26 Motion Metrics International Corp. Method, apparatus and system for monitoring a condition associated with operating heavy equipment such as a mining shovel or excavator
CN111429026A (en) * 2020-04-14 2020-07-17 西安热工研究院有限公司 Method for evaluating performance of electric shovel of strip mine
CN111429026B (en) * 2020-04-14 2023-02-07 西安热工研究院有限公司 Method for evaluating performance of electric shovel of strip mine
CN111764456A (en) * 2020-06-16 2020-10-13 武汉理工大学 Intelligent monitoring and alarming device and method for dropping of bucket teeth of forklift
US11821177B2 (en) 2021-02-09 2023-11-21 Caterpillar Inc. Ground engaging tool wear and loss detection system and method
US11669956B2 (en) 2021-06-01 2023-06-06 Caterpillar Inc. Ground engaging tool wear and loss detection system and method
US11869331B2 (en) 2021-08-11 2024-01-09 Caterpillar Inc. Ground engaging tool wear and loss detection system and method

Similar Documents

Publication Publication Date Title
CN107862675A (en) A kind of real-time vision detection method for electric bucket tooth missing
CN110390691B (en) Ore dimension measuring method based on deep learning and application system
CN111951237A (en) Visual appearance detection method
CN106446926A (en) Transformer station worker helmet wear detection method based on video analysis
CN108564085B (en) Method for automatically reading of pointer type instrument
CN104978567B (en) Vehicle checking method based on scene classification
CN111161292B (en) Ore scale measurement method and application system
CN104166841A (en) Rapid detection identification method for specified pedestrian or vehicle in video monitoring network
Wang et al. Investigation into recognition algorithm of helmet violation based on YOLOv5-CBAM-DCN
CN104700092A (en) Small-character number identification method based on template and feature matching
CN101751603A (en) Online bar section image automatic counting device and method
CN107145884A (en) Gangue near-infrared image identification technology
CN106548131A (en) A kind of workmen's safety helmet real-time detection method based on pedestrian detection
CN112001878A (en) Deep learning ore scale measuring method based on binarization neural network and application system
CN105447859A (en) Field wheat aphid counting method
CN115797798A (en) Ecological restoration effect evaluation method based on abandoned mine remote sensing image
CN106485273A (en) A kind of method for detecting human face based on HOG feature and DNN grader
CN108009556A (en) A kind of floater in river detection method based on fixed point graphical analysis
CN106951863A (en) A kind of substation equipment infrared image change detecting method based on random forest
CN105354547A (en) Pedestrian detection method in combination of texture and color features
CN111259763A (en) Target detection method and device, electronic equipment and readable storage medium
CN109815784A (en) A kind of intelligent method for classifying based on thermal infrared imager, system and storage medium
CN107145820B (en) Binocular positioning method based on HOG characteristics and FAST algorithm
Abraham et al. Vehicle detection and classification from high resolution satellite images
CN117330582A (en) Polymer PE film surface crystal point detecting system

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

Application publication date: 20180330

RJ01 Rejection of invention patent application after publication