CN104063711B - A kind of corridor end point fast algorithm of detecting based on K means methods - Google Patents
A kind of corridor end point fast algorithm of detecting based on K means methods Download PDFInfo
- Publication number
- CN104063711B CN104063711B CN201410283061.3A CN201410283061A CN104063711B CN 104063711 B CN104063711 B CN 104063711B CN 201410283061 A CN201410283061 A CN 201410283061A CN 104063711 B CN104063711 B CN 104063711B
- Authority
- CN
- China
- Prior art keywords
- image
- end point
- lines
- point
- cluster
- 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.)
- Active
Links
Landscapes
- Image Analysis (AREA)
Abstract
The present invention proposes a kind of corridor end point fast algorithm of detecting based on K means methods, and after the view data that robot returns in real time is obtained, processing early stage, including down-sampled, gray processing processing, histogram equalization and Canny rim detections are first carried out to it；The extraction that imagery exploitation probability Hough transformation algorithm carries out lines is obtained to previous step again；Finally the straight line detected is gathered for four classes according to slope using K means algorithms, calculate the average at every cluster straight line midpoint, then the slope and midpoint average obtained using cluster builds the straight line that four straight lines replace previous step to detect, four straight randoms are divided into two groups, its intersection point is calculated respectively, takes the midpoint of intersection point as end point.This method can be quickly detected the position of end point, and then the direction of advance of robot is modified using end point, realizes the navigation to robot.Compared with existing vanishing Point Detection Method method, method proposed by the present invention has the characteristics of simple efficient, real-time is good, stability is strong.
Description
Technical field
The present invention relates to robot autonomous navigation field, specially a kind of corridor end point inspection based on K-means methods
Method of determining and calculating, the position coordinates of end point is detected in the image that accurately can be obtained in real time from robot, and then be robot
Motion provide navigation information.It is can also be applied to autonomous flight of the Small and micro-satellite in building interior edge corridor is appointed
Business.
Background technology
One group of space parallel lines, one and only one intersection point of imaging, referred to as end point on the image plane, it is wrapped
The directional information of parallel lines on image is contained, therefore can be used to provide navigation information for robot, this technology is wide
General application.When robot moves in corridor, using the real-time image information obtained from camera, end point in corridor is calculated
Coordinate, the continuous direction of advance of amendment robot accordingly, the autokinetic movement in final achievable Robot corridor.
In order that being navigated with motion of the end point to robot, have to Vanishing Point Detecting Algorithm two it is important will
Ask：(1) amount of calculation is small, and real-time is high；(2) the end point position detected is accurate.
Several Vanishing Point Detecting Algorithms popular at present select different mostly using Hough transformation as basic thought
Parameter space, main parameter space have Gaussian sphere, hough space and image space.Barnard,Stephen T."
Interpreting perspective images."Artificial intelligence21.4(1983):In 435-462.
It is proposed that the Gaussian sphere using centered on camera optics center is used as accumulation space.Straight line in image is mapped in Gaussian sphere
For a circle, end point correspond to a point in Gaussian sphere.Because the unlimited plane of delineation is mapped to a limited height
This spherical space, therefore the end point of unlimited distance is equally mapped to a point.It is determined that the cumulative voting stage of end point, high
This ball is divided into multiple accumulation junior units, and each circle is voted the cumulative unit that it passes through, the cumulative unit more than votes
Corresponding to end point.This method can effective detection to limit remote and the end point of unlimited distance.Tuytelaars,
Tinne,et al."The cascaded Hough transform as an aid in aerial image
interpretation."Computer Vision,1998.Sixth International Conference on.IEEE,
1998. use the thought similar with Barnard, are converted the straight line parameter in plane to hough space using Hough transformation,
Hough space carries out cumulative voting, determines end point, reduces amount of calculation to a certain extent.Rother,Carsten."A
new approach to vanishing point detection in architectural environments."
Image and Vision Computing20.9(2002):647-655. will detect straight line two-by-two in all planes of delineation
Candidate target of the intersection point as end point, and some constraintss are given to determine end point, this method can obtain more
For accurate result.
But there is some following deficiency in above-mentioned several method：(1) in Gauss spherical space and hough space method, warp
Gaussian sphere is converted with after Hough transformation, reduces line segment and the spatial positional information of end point, meanwhile, both approaches detect
End point precision had a great influence by cumulative unit size, when cumulative unit is larger, it is larger to obtain the error of end point, when
When cumulative unit is too small, amount of calculation will be multiplied, it is impossible to meet requirement of real-time；(2) end point is carried out in the plane of delineation
Detection when, all possible straight-line intersection is calculated, the precision of this method is higher, but in practical application, due to
Detect that straight line quantity is larger, algorithm complex is high, computationally intensive, therefore real-time is poor, can not meet to be actually needed.
The content of the invention
For overcome the deficiencies in the prior art, it is fast that the present invention proposes a kind of corridor end point based on K-means methods
Fast detection algorithm, vanishing Point Detection Method is carried out using K-means methods, it is only necessary to calculate two intersection points of straight line, you can quick accurate
The true position for calculating end point in image.
The technical scheme is that：
A kind of corridor end point fast algorithm of detecting based on K-means methods, it is characterised in that：Using following step
Suddenly：
Step 1：The image of collection is pre-processed, gets rid of the redundancy in image, reduces the influence of noise；In advance
Processing procedure is followed successively by：Down-sampled, gray processing processing, histogram equalization are carried out to the image of collection, is entered using Canny operators
Row rim detection；
Step 2：Image after being handled using probability Hough transformation algorithm detecting step 1, extract the lines in image；
Step 3：The lines for using K-means methods to be extracted step 2 according to lines slope are clustered as four classes, are obtained respectively
The average at four class lines midpoints；Four straight lines are determined according to every class lines slope and corresponding lines midpoint average altogether, four straight
Line is randomly divided into two groups, and every group of two straight lines obtain an intersection point, take the midpoint of two intersection points as end point.
Further preferred scheme is：
A kind of corridor end point fast algorithm of detecting based on K-means methods, it is characterised in that：Adopted in step 3
The lines cluster for being extracted step 2 according to lines slope with K-means methods is as follows for the process of four classes：
Step 3.1：The slope for the n bar lines that calculation procedure 2 detects, and using n obtained slope value as to be clustered
Object；Any selection four is used as initial cluster center from n object；Initial cluster center is to be polymerized to four classifications
Initial center；
Step 3.2：Each object to be clustered is calculated to draw to the Euclidean distance of all cluster centres, and by each clustering object
The cluster centre nearest apart from it is assigned to, a member of classification is corresponded to as the cluster centre；
Step 3.3：Recalculate the average of each member in each classification, the cluster centre new as the category；
Step 3.4：When each object to be clustered to its generic cluster centre can not reduce again apart from sum, gather
Class terminates, otherwise return to step 3.2.
Beneficial effect
The present invention realizes a kind of corridor Vanishing Point Detecting Algorithm of precise and high efficiency, what this method can obtain in robot
In image, the position of end point is quickly detected, then the direction of advance of robot is modified using end point, is realized
Navigation to robot.Compared with existing vanishing Point Detection Method method, method proposed by the present invention has simple efficient, real-time
Well, the characteristics of stability is strong.
The present invention why there is its reason of above-mentioned beneficial effect to be：Method proposed by the present invention obtains to robot first
The corridor image obtained is pre-processed, and gets rid of and does not have helpful view data to calculating end point, is only retained and is determined that a corridor disappears
The image data portions of point are lost, considerably reduce the information content of required processing, reduce the complexity of calculating.And due to general
Rate Hough transformation is a relatively time-consuming process, and after step (1) removes a large amount of extraneous datas, the lines of step (2) carry
Process is taken to also save calculating cost, and determination of the lines detected to end point is useful.It is determined that end point
During position, not using the method found intersection two-by-two to a large amount of straight lines, but the architectural feature in corridor is taken into full account, using K-
Means methods gather the straight line detected in image for four classes, are replaced using four obtained straight lines previously detected a large amount of
Straight line, this four straight lines can determine the position of end point, further simplify calculating in this way, finally only need
In the case of calculating two intersection points of straight line, so that it may obtain the end point in corridor.
Brief description of the drawings
Fig. 1：Flow chart of the method for the present invention；
Fig. 2：Processing image in embodiment.
Embodiment
The present invention is further described with reference to the accompanying drawings and examples.
Accompanying drawing 1 illustrates the overall flow that the present invention realizes vanishing Point Detection Method in corridor, and the flow contains determination and disappeared
Each key step needed for point.The purpose of the present invention is the image information for utilizing the robot moved in corridor to return, quickly
The position of end point in corridor is calculated, and the motion of Robot corridor is navigated using the end point detected.Accompanying drawing 2
Illustrate each process for detecting end point in the corridor image of a width random shooting using method proposed by the present invention.
Here is concrete implementation step, and whole process is illustrated by taking the input picture in accompanying drawing 2 as an example.
First, pretreatment early stage of image
After robot returns to a pictures, relevant treatment first is carried out to image using the method for computer vision, to subtract
The interference of noise in few surrounding environment, get rid of a pair determination end point and do not have effective view data, and to the edge in image
Detected, be ready for extraction lines in next step.
(1) the down-sampled processing of image
Enter to be about to down-sampled processing to the image collected first, reduce the size that need to handle image, in practical operation,
We are by the length of image and wide are changed into original 1/2.The step, which reduces, needs information content to be processed, reduces calculating process
In time cost.
(2) gray level image
Because the method for computer vision is highly susceptible to the influence of illumination condition in local environment, in order to by this influence
Minimize, down-sampled obtained RGB image is converted into gray level image by we using formula (1), is reduced illumination variation and is offseted
Lose the influence of point detection.Obtain image shown in accompanying drawing 2-b.
Gray=0.229R+0.587G+0.114B (1)
(3) histogram equalization
Histogram equalization is the method being adjusted in image processing field using image histogram to picture contrast.
The gray level image obtained for step (2), allows niRepresent the number that gray scale i occurs, the pixel that gray scale is i in this sampled images going out
Now rate is
In formula, L is greys all in image, and n is the number of pixels of gray level image altogether, and p is the histogram of image,
0 is normalized to ..., 1.The cumulative probability function that c is p is defined, is defined as：
The transforming function transformation function that a form is y=T (x) is defined, for each pixel x in original image, with regard to one can be produced
Individual y, such y cumulative probability function can are linearized in the range of all values, are defined conversion formula and are：
yi=T (xi)=c (i) (4)
Different grades is mapped to 0 by T ..., 1 domain, in order to which these values are mapped back into their initial domains, it is necessary in result
Using following conversion：
yi'=yi(max-min)+min (5)
Wherein max is the maximum in gray level image set of pixels obtained by step (2), and min is the minimum value of wherein pixel.It is logical
Histogram equalization is crossed, improves the contrast of image, is advantageous to rim detection in the picture.Obtain accompanying drawing 2-c institutes diagram
Picture.
(4) rim detection is carried out using Canny operators
The imagery exploitation Canny operators obtained to step (3) carry out rim detection.Canny operators carry out rim detection can
It is summarized as four steps：
1st, image is smoothed using Gaussian smoothing function, to eliminate the influence of noise；
The present invention is handled image using following Gaussian smoothing function：
It is smoothed, can be put down with gaussian kernel function H (x, y, σ) the image f (x, y) obtained to step (3)
Image g (x, y) after cunning, specific formula are：
G (x, y)=H (x, y, σ) * f (x, y) (7)
In formula, * represents convolution algorithm.
2nd, the amplitude of gradient and direction are calculated with the finite difference of single order local derviation, to reach the purpose of edge enhancing；
The first difference of image pixel can be used to divide to carry out approximation for the gradient of image intensity value, must can thus scheme
As two matrixes of partial derivative in the x and y direction.
We carry out convolution algorithm to image using the convolution operator shown in following table in the present invention, and what is embodied on the left of following table is
The difference of image in the x direction, what right side was embodied is the difference of image in y-direction.
Obtaining first-order partial derivative matrix P and Q of the image in x directions, y directions is respectively：
P (x, y)=[f (x+1, y)-f (x, y)+f (x+1, y+1)-f (x, y+1)]/2 (8)
Q (x, y)=[f (x, y+1)-f (x, y)+f (x+1, y+1)-f (x+1, y)]/2 (9)
Obtain the amplitude and deflection of gradient simultaneously, gradient magnitude M and gradient direction θ mathematic(al) representation are：
θ (x, y)=arctan (P (x, y)/Q (x, y)) (11)
M (x, y) reflects the intensity at edge, and θ (x, y) reflects the direction at edge.So that M (x, y) obtains local maxima
The deflection θ (x, y) of value, it is just the direction at the edge.
By the step, the image that the feature at marginal position is reinforced is obtained.
3rd, the point of partial gradient maximum in image is calculated, the point of most non-edge is weeded out with this；
For each pixel in image, its picture value element M (x, y) and two pixel value phases along gradient line are made
Than if the gray value of central pixel point is big unlike the gray value of two adjacent image points along gradient line, making central pixel point
Gray value is 0, otherwise keeps constant.The operation remains the maximum point of partial gradient value, and the pixel value of other points is set
It is set to 0.
4th, bilinearity threshold value is carried out, obtains the marginal point of image；
To the image obtained in the previous step for remaining partial gradient value maximum, two threshold alphas are selected1And α2, wherein α1For
Low threshold, α2For high threshold.Grad is less than α by us1The gray value of pixel be arranged to 0, obtain image 1, then by Grad
Less than α2The gray value of pixel be set to 0, obtain image 2.Because the threshold value of image 2 is higher, most of noise is eliminated, but together
When also have lost useful marginal information.And the threshold value of image 1 is relatively low, more information is remained, we are base with image 2
Plinth, link the marginal point of image so that image 1 is supplement.Specific edge joining step is as follows：Image 2 is swept first
Retouch, when running into non-zero pixel p (x, y) of a gray value, track with the contour line that p (x, y) is starting point, until contour line
Terminal q (x, y).Then, the point s (x, y) of position correspondence 8 adjacent domains are put in image under consideration 1 with q (x, y) in image 2.
If s (x, y) point 8 adjacent domains in there is non-zero pixels s (x, y) to exist, then it is included in image 2, as r (x,
Y) point.Since r (x, y), continuation is searched for forward, untill we can not continue in image 1 and image 2.Work as completion
After link to the contour line comprising p (x, y), this contour line is labeled as having accessed.Continuation finds it in the picture
Its edge, until can not be untill new edge is found.
By the high-low threshold value α of Canny algorithms in the present invention1And α2Ratio be set to 5:2.Obtain such as accompanying drawing 2-d institutes diagram
Picture.
The edge that the step detects is the basis that lines are extracted from image, and reduces next step probability Hough transformation
Hunting zone, it is scanned for the place that marginal point be present, reduce time cost.
2nd, in image lines extraction
The picture that robot passes back in real time, handled through above step to eliminate correlated noise, while reduce and to be located
The information content of reason, have detected for lines can be carried out below.The present invention carries out lines in image using probability Hough transformation algorithm
Detection, the algorithm is highly suitable for response time certain real-time system.
Straight line on the plane of delineation can be expressed as under polar coordinate system：ρ=xcos θ+ysin θ, in formula, ρ is original
Point is to the distance of straight line, and angles of the θ between ρ and x-axis, on a known straight line, ρ and θ are constant, and (x, y) is on straight line
The coordinate at any point.A point (x, y) in image space, ρ and θ parameter space are mapped to, reforms into a curve.Figure
Straight line in image space, is mapped to ρ, θ parameter space, is just changed into many bar curves to intersect at a point, therefore, is scheming
In image space find straight line operation, translate into ρ, θ parameter space search curve intersection point the problem of.
Probability Hough transformation algorithm comprises the following steps that：
1st, multiple impartial minizones are turned to by ρ, θ parameter space under polar coordinates are discrete, corresponding one of each section is tired
Add device A (ρ, θ), its initial value is zero；Then all marginal points obtained by previous step Canny rim detections are put into pending side
Edge point set；
2nd, detect whether pending edge point set is empty, if it is algorithm terminates；Otherwise at random from pending marginal point
Concentration takes a pixel, is projected to ρ, θ parameter space and obtains a curve, all cells passed through to this curve
Between, the θ values that ρ values are corresponding in each minizone are calculated, and accumulator A (ρ, θ) corresponding to each minizone is added 1；
3rd, concentrated from pending marginal point and delete taken point；
4th, judge whether accumulator A (ρ, θ) value after renewal is more than threshold value threshold set in advance, if less than
Then return to second step；Otherwise go in next step；
If the 5, accumulator A (ρ, θ) value is more than threshold value threshold set in advance in 4, then explanation has a plurality of curve to pass through
The minizone, when the section that 1 is divided is smaller, then it is believed that this plurality of curve intersection is in a bit, this point is that section is enough
Point corresponding to hour.Then can equivalently obtain, these by the minizone ρ, θ parameter space curve, image (x,
Y) what is formed in plane is a line segment, and the number of pixel is the curved line number by the minizone on this line segment, two
Person is at least threshold value threshold.We record record of the beginning and end for detecting line segment as a line segment.
Then the point that pending marginal point is centrally located on the line segment is deleted, this accumulator is reset；
6th, the 2nd step is returned.
The image after Canny rim detections is handled with probability Hough transformation algorithm, you can detect to determine
The straight line of end point in image, every straight line determine by thereon two points of record.Obtain image shown in accompanying drawing 2-e.
3rd, the end point in image is determined
The straight line detected according to probability Hough transformation algorithm, the position of end point can be determined in two steps：
(1) end point in corridor is determined by the straight line along corridor direction, and is easiest to be detected in corridor
The straight line straight line that to be vertical wall be crossed to form with ground and ceiling, by entering to projection of these straight lines in the plane of delineation
Row analysis, it is found that these straight lines are substantially all along plane of delineation diagonal and be distributed, can substantially be divided into four clusters, its slope
Four classes can be divided into.In the straight line detected by step 2, these straight lines account for the overwhelming majority, therefore, will using K-means algorithms
The straight line that step 2 detects gathers for four classes.
K-means algorithms are a kind of widely used clustering algorithms, n object are divided into the class number specified, so that class
It is interior that there is higher similarity, and the similarity between class is as small as possible.The calculating of similarity is averaged according to object in each class
Value is carried out.It is as follows using the algorithm steps in the present invention：
1. the slope for the n bar straight lines that calculation procedure two detects, and using n obtained slope value as pair to be clustered
As then any selection four is used as initial cluster center from this n object, that is, to be polymerized to the initial of four classifications
Center；
2. each object to be clustered is calculated to the Euclidean distance of all cluster centres；And by each clustering object be divided into away from
The cluster centre nearest from it, a member as the category；
3. the average of each classification is recalculated, the cluster centre new as the category；
4. when each object to be clustered to its generic cluster centre can not reduce again apart from sum, i.e., algorithm has been
During convergence, terminate；Otherwise step is returned to 2..
(2) using the result of cluster, calculate the average per cluster straight line midpoint respectively, using the obtained straight slope of cluster with
The midpoint average obtained builds four straight lines according to point slope form, and previously detected straight line is replaced using this four straight lines.By four
Bar straight random is divided into two groups, and every group of two straight lines can all obtain an intersection point, and the midpoint for taking two intersection points is end point.
The end point detected in the input image is as shown in accompanying drawing 2-f.
It is summarized as follows with reference to Part I inventive algorithm：
(1) after the view data that robot returns in real time is obtained, processing early stage is first carried out to it, including it is down-sampled, grey
Degreeization processing, histogram equalization and Canny rim detections；
(2) extraction that imagery exploitation probability Hough transformation algorithm carries out lines is obtained to previous step；
(3) straight line that detects is gathered for four classes according to slope using K-means algorithms, calculates every cluster straight line midpoint
Average, the slope then obtained using cluster and midpoint average build the straight line that four straight lines replace previous step to detect, by four
Bar straight random is divided into two groups, calculates its intersection point respectively, takes the midpoint of intersection point as end point.
Evaluation index
In order to evaluate the result of the vanishing Point Detection Method performance of the present invention and comparative experiments, illustrated using following index：Disappear
Lose deviation and the operation time of point coordinates.The deviation definition of wherein disappearance point coordinates is：Its
In, (x1,y1) for actual disappearance point coordinates, (x2,y2) trial and error procedure to be measured (using algorithm proposed by the present invention) detects
Disappearance point coordinates.
Effect of the present invention can be further illustrated by following emulation experiment：
In order to further illustrate the validity of proposition method of the present invention, devise and verified below：
Experimental program：The picture that several sizes of random shooting are about 350x450 in different corridors, proposed using the present invention
Algorithm to every width picture automatic detection end point, and time used in recording, then with manual measurement to end point (can be considered
Actual end point) it is compared, end point grid deviation is calculated, then takes its average value to be compared.Disappeared from general
Point detection algorithm carries out vanishing Point Detection Method：It is then directly right first using the straight line in the Hough transformation algorithm detection image of standard
The method of intersection between lines point obtains end point.Record time used in this method and the disappearance point coordinates detected, and with this hair
The result of bright proposition method is compared.
Experimental result：By the threshold of Hough transformation in the present invention (when the intersection point number of every piece of zonule curve is more than
During threshold, then it is assumed that parameter corresponding to the region shows as a line segment on the image plane) it is set to 30,45 and
When 60, obtained experimental result is as shown in Table 1.
It can be seen that by table one, algorithm proposed by the present invention has very high end point computational accuracy, average end point deviation
Within 15 pixels, and there is very low calculating cost, robot can be allow to reserve more times and carry out it
He operates.Compared to utilizing Hough transformation detection of straight lines, method of the straight-line intersection as end point is then asked for, the present invention proposes
Selection of the method to Hough transformation threshold value threshold have very high robustness, when threshold value changes, during calculating
Between be held essentially constant, the end point grid deviation detected is maintained in the range of very little.Therefore method tool proposed by the present invention
There is good environmental suitability, there is very high practical value.
The vanishing Point Detection Method experimental result of table one
Claims (1)
- A kind of 1. corridor end point quick determination method based on K-means methods, it is characterised in that：Using following steps：Step 1：The image of collection is pre-processed, gets rid of the redundancy in image, reduces the influence of noise；Pretreatment Process is followed successively by：Down-sampled, gray processing processing, histogram equalization are carried out, using Canny operators progress side to the image of collection Edge detects；Step 2：Image after being handled using probability Hough transformation algorithm detecting step 1, extract the lines in image；Step 3：The rectilinear strip for using K-means methods to be extracted step 2 according to lines slope is clustered as four classes, is obtained respectively The average at four class lines midpoints；Four straight lines are determined according to every class lines slope and corresponding lines midpoint average altogether, four straight Line is randomly divided into two groups, and every group of two straight lines obtain an intersection point, take the midpoint of two intersection points as end point；The rectilinear strip cluster for wherein using K-means methods to be extracted step 2 according to lines slope is as follows for the process of four classes：Step 3.1：The slope for the n bar lines that calculation procedure 2 detects, and using n obtained slope value as pair to be clustered As；Any selection four is used as initial cluster center from n object；Initial cluster center for be polymerized to four classifications just Beginning center；Step 3.2：Each object to be clustered is calculated to the Euclidean distance of all cluster centres, and each clustering object is divided into The cluster centre nearest apart from it, a member of classification is corresponded to as the cluster centre；Step 3.3：Recalculate the average of each member in each classification, the cluster centre new as the category；Step 3.4：When each object to be clustered to its generic cluster centre can not reduce again apart from sum, cluster knot Beam, otherwise return to step 3.2.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201410283061.3A CN104063711B (en) | 2014-06-23 | 2014-06-23 | A kind of corridor end point fast algorithm of detecting based on K means methods |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201410283061.3A CN104063711B (en) | 2014-06-23 | 2014-06-23 | A kind of corridor end point fast algorithm of detecting based on K means methods |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN104063711A CN104063711A (en) | 2014-09-24 |
| CN104063711B true CN104063711B (en) | 2018-04-06 |
Family
ID=51551413
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201410283061.3A Active CN104063711B (en) | 2014-06-23 | 2014-06-23 | A kind of corridor end point fast algorithm of detecting based on K means methods |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN104063711B (en) |
Families Citing this family (10)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN106127751B (en) * | 2016-06-20 | 2020-04-14 | 北京小米移动软件有限公司 | Image detection method, device and system |
| CN109388131B (en) * | 2017-08-02 | 2022-04-26 | 上海阿科伯特机器人有限公司 | Robot attitude control method and system based on angular point feature recognition and robot |
| CN109388093B (en) * | 2017-08-02 | 2020-09-15 | 苏州珊口智能科技有限公司 | Robot attitude control method and system based on line feature recognition and robot |
| CN108986125B (en) * | 2017-11-30 | 2022-02-01 | 成都通甲优博科技有限责任公司 | Object edge extraction method and device and electronic equipment |
| CN107918775B (en) * | 2017-12-28 | 2020-04-17 | 聊城大学 | Zebra crossing detection method and system for assisting safe driving of vehicle |
| CN109238288A (en) * | 2018-09-10 | 2019-01-18 | 电子科技大学 | Autonomous navigation method in a kind of unmanned plane room |
| CN111358364B (en) * | 2018-12-26 | 2021-09-07 | 珠海市一微半导体有限公司 | Dead angle cleaning method and device based on visual robot, chip and robot |
| CN111368659A (en) * | 2020-02-24 | 2020-07-03 | 同济大学 | Intelligent port AGV unmanned vehicle driving track correction method |
| CN112101163A (en) * | 2020-09-04 | 2020-12-18 | 淮阴工学院 | Lane line detection method |
| CN112598665B (en) * | 2020-12-31 | 2022-05-06 | 北京深睿博联科技有限责任公司 | Method and device for detecting vanishing points and vanishing lines of Manhattan scene |
Citations (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN103366156A (en) * | 2012-04-09 | 2013-10-23 | 通用汽车环球科技运作有限责任公司 | Road structure detection and tracking |
Family Cites Families (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US8395824B2 (en) * | 2008-07-17 | 2013-03-12 | Samsung Electronics Co., Ltd. | Method for determining ground line |
-
2014
- 2014-06-23 CN CN201410283061.3A patent/CN104063711B/en active Active
Patent Citations (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN103366156A (en) * | 2012-04-09 | 2013-10-23 | 通用汽车环球科技运作有限责任公司 | Road structure detection and tracking |
Non-Patent Citations (3)
| Title |
|---|
| Vanishing point detection in corridors: using Hough transform and K-means clustering;R. Ebrahimpour等;《IET Computer Vision》;20120119;第6卷(第1期);第40-51页 * |
| 一种新的消隐点自动测量算法;舒远等;《红外与激光工程》;20031031;第32卷(第5期);第479-483页 * |
| 基于K-means算法的温室移动机器人导航路径识别;高国琴等;《农业工程学报》;20140430;第30卷(第7期);第25-33页 * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN104063711A (en) | 2014-09-24 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN104063711B (en) | A kind of corridor end point fast algorithm of detecting based on K means methods | |
| CN108171748B (en) | Visual identification and positioning method for intelligent robot grabbing application | |
| CN110340891B (en) | Mechanical arm positioning and grabbing system and method based on point cloud template matching technology | |
| CN107705322A (en) | Motion estimate tracking and system | |
| CN111145228B (en) | Heterologous image registration method based on fusion of local contour points and shape features | |
| CN110070567B (en) | Ground laser point cloud registration method | |
| Yu et al. | Robust robot pose estimation for challenging scenes with an RGB-D camera | |
| CN109767454B (en) | Unmanned aerial vehicle aerial video moving target detection method based on time-space-frequency significance | |
| Li et al. | Road lane detection with gabor filters | |
| CN104036524A (en) | Fast target tracking method with improved SIFT algorithm | |
| Wang et al. | Real-time plane segmentation and obstacle detection of 3D point clouds for indoor scenes | |
| CN102446356A (en) | Parallel and adaptive matching method for acquiring remote sensing images with homogeneously-distributed matched points | |
| CN102289822A (en) | Method for tracking moving target collaboratively by multiple cameras | |
| CN105405138A (en) | Water surface target tracking method based on saliency detection | |
| Li et al. | RGBD relocalisation using pairwise geometry and concise key point sets | |
| Yang et al. | Robust line detection using two-orthogonal direction image scanning | |
| Zhao et al. | Boundary regularized building footprint extraction from satellite images using deep neural network | |
| CN114689038A (en) | Fruit detection positioning and orchard map construction method based on machine vision | |
| CN104484647B (en) | A kind of high-resolution remote sensing image cloud height detection method | |
| CN109215122B (en) | Streetscape three-dimensional reconstruction system and method and intelligent trolley | |
| CN111160231A (en) | Automatic driving environment road extraction method based on Mask R-CNN | |
| EP3137895A1 (en) | Method and apparatus for processing block to be processed of urine sediment image | |
| CN104616302A (en) | Real-time object identification method | |
| Roychoudhury et al. | Plane segmentation using depth-dependent flood fill | |
| Liu et al. | Identification of Damaged Building Regions from High-Resolution Images Using Superpixel-Based Gradient and Autocorrelation Analysis |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| C06 | Publication | ||
| PB01 | Publication | ||
| C10 | Entry into substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant | ||
| TR01 | Transfer of patent right | ||
| TR01 | Transfer of patent right |
Effective date of registration: 20230106 Address after: 710077 Room 12004, Block A, Xi'an National Digital Publishing Base, No.996, Tiangu 7th Road, Yuhua Street Office, Hi tech Zone, Xi'an, Shaanxi Patentee after: XI'AN INNNO AVIATION TECHNOLOGY Co.,Ltd. Address before: 710072 No. 127 Youyi West Road, Shaanxi, Xi'an Patentee before: Northwestern Polytechnical University |