CN109815975A - A kind of objective classification method and relevant apparatus based on robot - Google Patents
A kind of objective classification method and relevant apparatus based on robot Download PDFInfo
- Publication number
- CN109815975A CN109815975A CN201811535587.0A CN201811535587A CN109815975A CN 109815975 A CN109815975 A CN 109815975A CN 201811535587 A CN201811535587 A CN 201811535587A CN 109815975 A CN109815975 A CN 109815975A
- Authority
- CN
- China
- Prior art keywords
- image
- sorted
- target
- profile
- principal component
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
- 238000000034 method Methods 0.000 title claims abstract description 70
- 238000003860 storage Methods 0.000 claims abstract description 10
- 239000011159 matrix material Substances 0.000 claims description 15
- 238000012545 processing Methods 0.000 claims description 10
- 238000012549 training Methods 0.000 claims description 8
- 239000000284 extract Substances 0.000 claims description 7
- 238000005260 corrosion Methods 0.000 claims description 4
- 230000007797 corrosion Effects 0.000 claims description 4
- 230000005484 gravity Effects 0.000 claims description 4
- 238000005457 optimization Methods 0.000 claims description 3
- 238000012216 screening Methods 0.000 claims description 2
- 238000010586 diagram Methods 0.000 description 8
- 238000004364 calculation method Methods 0.000 description 5
- 230000006870 function Effects 0.000 description 2
- 238000009776 industrial production Methods 0.000 description 2
- 238000011161 development Methods 0.000 description 1
- 238000009792 diffusion process Methods 0.000 description 1
- 238000005516 engineering process Methods 0.000 description 1
- 230000007717 exclusion Effects 0.000 description 1
- 238000001914 filtration Methods 0.000 description 1
- 230000014759 maintenance of location Effects 0.000 description 1
- 239000000203 mixture Substances 0.000 description 1
- 238000007781 pre-processing Methods 0.000 description 1
- 238000002360 preparation method Methods 0.000 description 1
Landscapes
- Image Analysis (AREA)
Abstract
This application discloses a kind of objective classification methods based on robot, this method comprises: converting to the first image comprising target to be sorted, to obtain the second image of preset format；The profile of target to be sorted is extracted from the second image；Profile based on the target to be sorted pre-processes the first image, to obtain third image；Calculate the principal component of third image；The principal component of third image is compared with the principal component of preset multiple sample images for having carried out type mark, with the affiliated type of determination target to be sorted.It may be implemented to classify to the target to be sorted of multiple types and unlimited shape through the above way.Present invention also provides a kind of robot control system and storage mediums.
Description
Technical field
This application involves robot fields, more particularly to a kind of objective classification method based on robot and related dress
It sets.
Background technique
There is much being sorted or being classified for task in industrial production line, the prior art is for such as sorting or dividing
This kind of cumbersome boring task of class mostly be to be completed by worker's manual operations, the classification effectiveness that will cause in this way is lower, and need compared with
More human costs.And in recent years with the fast development of robot technology, using robot in industrial production line enforcement division
The classification task divided, but special mechanical structure is required to construct sorter, and constructed sorter is mostly only
Suitable for specific several workpiece, and it is unable to satisfy the classification to multiple types and the similar workpiece of shape, therefore needs one kind can be with
Solve the technical solution of above-mentioned technical problem.
Summary of the invention
The application classifies to the target to be sorted of multiple types and unlimited shape mainly solving the technical problems that realizing.
In order to solve the above technical problems, the technical solution that the application uses is: providing a kind of mesh based on robot
Mark classification method, which comprises
The first image comprising target to be sorted is converted, to obtain the second image of preset format；
The profile of the target to be sorted is extracted from second image；
Profile based on the target to be sorted pre-processes the first image, to obtain third image；
Calculate the principal component of the third image；
By the principal component of the principal component of the third image and preset multiple sample images for having carried out type mark into
Row compares, with the affiliated type of the determination target to be sorted.
In order to solve the above technical problems, another technical solution that the application uses is to provide a kind of robot control system
System, the system comprises processor, memory and telecommunication circuit, the processor and the memory and the telecommunication circuit phase
It connects；
Wherein, the telecommunication circuit is used to for the instruction of the processor being sent to the mechanical structure of robot, so that
Target to be sorted is moved to setting regions by the mechanical structure；
The memory is for storing program data；
The processor is for running described program data, to execute method as described above.
In order to solve the above technical problems, another technical solution that the application uses is to provide a kind of storage medium, it is described
Storage medium is stored with program data, and described program data are performed a kind of realization target based on robot as described above
Classification method.
Above scheme, by that will include after the first image of target to be sorted is converted, to obtain the second of preset format
Then image goes out the profile of target to be sorted from the second image zooming-out, the profile based on target to be sorted carries out the first image
Pretreatment, to obtain third image and seek the principal component of third image, then by the principal component of resulting third image and in advance
If the principal components of multiple sample images for having carried out type mark be compared, with the affiliated type of determination target to be sorted.
In the process, realize under the conditions ofs being not limited to the shapes textures etc. of target to be sorted, realize to a variety of targets to be sorted into
Row classification processing, and when the type of target to be sorted changes or outer dimension changes is having no need to change point
It can be realized in the case where the logic of class and classify to multiple batches of target to be sorted, without special more exchange device either equipment
Corresponding structure etc., the range of preferably widened classification application.
Detailed description of the invention
Fig. 1 is a kind of flow diagram of one embodiment of objective classification method based on robot of the application；
Fig. 2 is a kind of flow diagram of another embodiment of objective classification method based on robot of the application；
Fig. 3 is a kind of flow diagram of the another embodiment of objective classification method based on robot of the application；
Fig. 4 is the called algorithm of step S303 in a kind of another embodiment of objective classification method based on robot of the application
Operation chart；
Fig. 5 is the structural schematic diagram of one embodiment of the application robot control system；
Fig. 6 is a kind of structural schematic diagram of one embodiment of storage medium of the application.
Specific embodiment
Below in conjunction with the attached drawing in the embodiment of the present application, technical solutions in the embodiments of the present application carries out clear, complete
Site preparation description.It is understood that specific embodiment described herein is only used for explaining the application, rather than to the limit of the application
It is fixed.Based on the embodiment in the application, obtained by those of ordinary skill in the art without making creative efforts
Every other embodiment, shall fall in the protection scope of this application.
In the description of the present application, the meaning of " plurality " is at least two, such as two, three etc., unless otherwise clearly having
Limit to body.In addition, term " includes " and " having " and their any deformations, it is intended that cover and non-exclusive include.Example
The process, method, system, product or equipment for such as containing a series of steps or units are not limited to listed step or list
Member, but optionally further comprising the step of not listing or unit, or optionally further comprising for these process, methods, product
Or other step or units that equipment is intrinsic.
Referenced herein " embodiment " is it is meant that a particular feature, structure, or characteristic described can wrap in conjunction with the embodiments
It is contained at least one embodiment of the application.Each position in the description occur the phrase might not each mean it is identical
Embodiment, nor the independent or alternative embodiment with other embodiments mutual exclusion.Those skilled in the art explicitly and
Implicitly understand, embodiment described herein can be combined with other embodiments.
A kind of executing subject of objective classification method based on robot provided herein can be robot, may be used also
To be robot control system.Wherein, robot control system can be used for controlling multiple robots while execute and treats
The classification task of class object.
Referring to Figure 1, Fig. 1 is the process signal in a kind of one embodiment of objective classification method based on robot of the application
Figure.In the present example, method provided herein includes:
S110: converting the first image comprising target to be sorted, to obtain the second image of preset format.
Obtain include target to be sorted the first image, specifically can be directly by being directly connected to current robot
Capture apparatus carries out shooting acquisition, can also be that the equipment being indirectly connected with therewith is obtained, is then forwarded to holding for this method
At row main body.Wherein, target to be sorted refers to the object that current robot is classified, specific to include at least in industry
Workpiece, the type for target to be sorted and appearance etc. do not do any restriction herein.Target to be sorted is in the first image
Completely, i.e., it can see the integrity profile of target to be sorted in the first image, and in the present example, to the first image
Included in the quantity of target to be sorted do not do any restriction, i.e., it is to be sorted included in the first image in present example
The quantity of target can be multiple, be also possible to one.
After getting the first image of target to be sorted, the first image can be converted, to obtain preset format
The second image.In the present example, the first image is RGB image, and the second image of preset format is binary image, two
The pixel gray value of background and target to be sorted is respectively set to 0 and 255 in value image.It should be understood that in other realities
It applies in example, the first image and the second image can also be the image of extended formatting, specific numerous to list herein.
Wherein, when the first image is RGB image, and the second image is binary image, the first image is converted to second
When image, the first image is converted into hsv color space from RGB color first, then selection is suitable based on experience value
Tone H and lightness V obtains the binary image comprising objective contour to be sorted so that the filtering background in the first image falls.Tool
It is as follows from RGB color to be converted the publicity to hsv color space by body for first image.
M=max (R, G, B), m=min (R, G, B), C=M-m
Wherein, HtempThe tone that is represented in 0~6, H of value.R, G, B refer respectively to the picture of each color in RGB image
Element value, it is meaningless that H=undefined, which represents current tone value,.
S120: the profile of target to be sorted is extracted from the second image.
From the second image obtained in above-mentioned steps S110, the profile of target to be sorted is extracted, and extracts profile and is
It connects continuous and with same grayscale point to obtain curve, to obtain the wheel of target to be sorted by extracted curve
It is wide.Wherein, in the present example, Eight directions Contour searching algorithm is called to extract profile.Wherein, about calling Eight directions
The detailed process that Contour searching algorithm extracts profile refers to the elaboration of the part Fig. 4.
S130: the profile based on target to be sorted pre-processes the first image, to obtain third image.
In step S120 after the profile for extracting target to be sorted in the second image, it can be based on being mentioned from the second image
The profile of the target to be sorted taken pre-processes position corresponding to the profile in the first image, to obtain third image.
Wherein, pretreatment includes at least the direction for identifying target to be sorted and obtains third image based on the first image zooming-out.
Further, in another embodiment, Fig. 2 is referred to, Fig. 2 is a kind of classification method based on robot of the application
The flow diagram of another embodiment, in present example, step S130 includes:
S201: according in the first image with the calculated for pixel values target to be sorted in the profile corresponding region of target to be sorted
Image reform.
In conjunction with the rotational invariance of the center of gravity of image, the center of gravity of the corresponding image of target to be sorted can be calculated first, made
For one of reference point in the direction of mark target to be sorted.The formula for specifically calculating center of gravity is as follows:
Wherein, mpqIt is image moment, p, q are the orders of square, and function I (x, y) is the profile of target to be sorted in the first image
The pixel value of middle corresponding region, x, y are pixel coordinates.Wherein, the calculation formula of I (x, y) is as follows:
I (x, y)=0.299*R (x, y)+0.587*G (x, y)+0.114*B (x, y).
S202: the geometric center of the minimum circumscribed rectangle of the profile of target to be sorted is determined.
In the present example, using the external square of minimum area for obtaining the profile of target to be sorted by calculating
Shape is then based on minimum area boundary rectangle and determines corresponding geometric center.
S203: the mark direction of target to be sorted is determined based on image reform and geometric center.
According to resulting image reform and geometric center, the direction that image reform is directed toward geometric center is determined as wait divide
Classification target identifies direction.It, can be in corresponding image, according to target to be sorted behind the mark direction for obtaining target to be sorted
Place profile, in the picture by resulting mark direction signs.
Image-region corresponding to the profile of target to be sorted is extracted in S204: the first image.
When including multiple targets to be sorted in the first image, and after extracting the profile of multiple targets to be sorted, can be based on
Resulting profile extracts target image to be sorted in the first image.It specifically in the present example, is to extract mesh to be sorted
The corresponding part of minimum area boundary rectangle corresponding to target profile, as in the first image comprising 7 targets to be sorted and
When extracting the profile of 7 targets to be sorted, then it can be extracted in the first image according to the objective contour to be sorted extracted
The corresponding image of minimum area boundary rectangle corresponding to 7 profiles extracts 7 small images comprising target to be sorted.
It should be understood that in other embodiments, target to be sorted can be extracted based on other reference standards, do not limited herein specifically
It is fixed.
S205: image-region is rotated, so that mark direction be arranged along preset target direction, and then acquisition the
Three images.
It, can be by the profile of target to be sorted extracted in step S204 after acquiring the mark direction of target to be sorted
Corresponding image-region is rotated, so that the target identification direction to be sorted identified is set along preset target direction
It sets, it is specifically parallel with preset target direction.When there are multiple images region (in first image there are it is multiple to
When class object), then can the geometric center for not changing target minimum area boundary rectangle to be sorted in the first image substantially
Position, i.e., the relative position where not changing each target to be sorted between small image, which is rotated, so that mark
Direction it is parallel with preset direction, and then obtain third image.When in the first image including multiple targets to be sorted, then pass through
It is identical for crossing the mark direction of the target to be sorted after image preprocessing in resulting third image.
Specifically, third image is the corresponding external square of profile minimum area of target to be sorted extracted from the first image
The image-region of shape, and along the image of preset target direction setting after rotation.Wherein, in the present example, may be used
To set the direction parallel with y-axis for preset target direction, it is possible to understand that in other embodiments, can also will preset
Target direction be set as other directions, the actual setting of concrete foundation is adjusted.
Therefore the angle rotated in the present example to image-region is the image district identified after direction and rotation
The angle in direction is identified in domain, it is corresponding, when the inceptive direction of mechanical structure is identical as preset target direction, then in machinery knot
Structure when obtaining target to be sorted, can correspond to by mechanical structure with rotating machinery structure in the opposite direction in image rotation direction
Identical angle is sent the target to corresponding specification area with clamping the target to be sorted.
It further, in other embodiments, further include S206 after step S205.
S206: zooming in and out third image, so that the resolution ratio of third image meets preset resolution requirement.
Of different sizes due to target to be sorted, the corresponding size in the first image is not also identical, therefore is getting
It, can also be to third image memory scaling processing, so that the resolution ratio of third image can satisfy preset want after third image
It asks.In the present example, preset resolution requirement is 100 × 100.
S140: the principal component of third image is calculated.
After acquiring third image, the principal component for calculating third image is needed further exist for.Wherein in present example, figure
The principal component of picture includes at least the PCA calculated result of image.
Further, in another embodiment, step S140 includes: that the pixel value x of third image is substituted into preset formula
Calculate the principal component y for acquiring third image.
Wherein, preset formula is To be counted respectively by all types of sample images
The transposed matrix of the orthogonal matrix obtained is calculated, μ is the pixel mean value that acquisition is calculated separately by all types of sample images.
In another embodiment,The orthogonal matrix of acquisition can also be calculated jointly for the sample image of multiple types
Transposed matrix, μ is that the sample image of the multiple type calculates the image mean value of acquisition jointly.
S150: by the principal component of the principal component of third image and preset multiple sample images for having carried out type mark into
Row compares, with the affiliated type of determination target to be sorted.
The principal component of third image obtained in step S140 will be passed through, with preset multiple samples for having carried out type mark
This image principal component is compared, with the affiliated type of determination target to be sorted.Further, in one embodiment, step
S150 further include: seek the principal component of third image and the principal component of multiple sample images for having carried out type mark it is European away from
From, by resulting multiple Euclidean distances be mutually compared size calculate, to obtain the smallest Euclidean distance.Getting minimum
After Euclidean distance, type of the corresponding type of output minimum euclidean distance sample image as target to be sorted.
Wherein, preset multiple sample images for having carried out type mark include at least current all targets to be sorted.
Preset multiple corresponding principal components of sample image for having carried out type mark are based on to current all types target to be sorted
Image be trained after, acquire and save into system in case treat calling when class object is classified.
It, can be pre- advanced before targets to be sorted multiple to certain are classified in technical solution provided herein
Row training is to obtain the corresponding type of principal component, principal component and other relevant parameters of sample image, and by resulting sample
The corresponding type of principal component, principal component of image and other relevant parameters save, so as to it needs to be determined that belonging to target to be sorted
It is called when type.
Wherein, the process of the training carried out in advance also will include above-mentioned each step, with treat class object into
Unlike row identification assorting process, training process can be trained of a sort Target Acquisition plurality of pictures to be sorted, with
Acquire the principal component and other relevant parameters of current type target to be sorted, and by the every class target to be sorted acquired it is main at
Divide and other relevant parameters are saved to system.
In embodiment corresponding to Fig. 1, by that will include to be obtained in advance after the first image of target to be sorted is converted
If the second image of format, then goes out the profile of target to be sorted from the second image zooming-out, the profile pair based on target to be sorted
First image is pre-processed, to obtain third image and seek the principal component of third image, then by resulting third image
Principal component be compared with the principal component of preset multiple sample images for having carried out type mark, with determination target to be sorted
Affiliated type.In the process, it realizes and is not limited under the conditions of shapes textures of target to be sorted etc., realize to a variety of wait divide
Class target carries out classification processing, and when the type of target to be sorted changes or outer dimension changes, is being not required to
It can be realized in the case where changing classification logic and classify to multiple batches of target to be sorted, set without specially replacing
Standby either corresponding structure of equipment etc., the range of preferably widened classification application.
Fig. 3 is referred to, Fig. 3 is a kind of process of objective classification method based on robot of the application in another embodiment
Schematic diagram.In the present example, method provided herein includes:
S301: converting the first image comprising target to be sorted, to obtain the second image of preset format.
S302: optimizing processing to the second image, to eliminate the noise spot in the second image.
The first image comprising target to be sorted is converted in step S301, to obtain the second image of preset format
Later, method provided herein in present example can also optimize processing to the second image, to eliminate the second image
In noise spot.Specifically, noise spot includes: noise spot, boundary point, lesser nonsensical point.
Wherein, optimization processing includes corrosion treatment and/or expansion process.
Specifically, corrosion treatment is carried out to the second image, specifically by boundary to the process of central reduction, to eliminate boundary
Point, lesser point and nonsensical point.Carrying out expansion process to image is to merge the point being connected with target, and boundary is outside
The process of diffusion, the region that can be filled in the blanks.
Specifically, it when optimization processing includes corrosion treatment and expansion process, then needs first to corrode the second image
Processing, then carry out expansion process.
In the present example, the step of profile of target to be sorted is extracted from the second image further comprises step
Content shown in S303.
S303: extracting and connects that the gray value in the second image is identical and continuous pixel, to obtain target to be sorted
Profile.
It in the present example, is the profile for calling Eight directions Contour searching algorithm to obtain target to be sorted.Specifically,
Fig. 4 is referred to, Fig. 4 calls calculation by step S303 in a kind of another embodiment of objective classification method based on robot of the application
The operation chart of method.Specifically, the key step of Eight directions Contour searching algorithm includes:
1. can be upper left from a direction of the second image and start to search for image until finding pixel in new region
Point.If pixel P0(figure does not identify P0) be all column of row in new region minimum image vegetarian refreshments.
②P0It is the starting pixels point of zone boundary.A variable dir is defined come boundary element before storing to working as front
The direction of motion of bound component.
3. according to Eight directions searching algorithm, it is assumed that dir=7, it is determined that 8 directions of search, i.e. the 0 of (a) in Fig. 4~
7 corresponding directions (in such as Fig. 4 shown in (a)).
4. starting 3 × 3 neighborhood search in a clockwise direction according to current pixel point.
5. (b) is shown in Fig. 4, wherein dir=(dir+7) mod8 if dir is even number.
If dir is odd number, (c) is shown in Fig. 4, wherein dir=(dir+6) mod8.
Finding first point equal with current pixel point gray value is new outline elements Pn, update dir.
6. if PN=P1And PN=P0, wherein P1It is second outline elements, Pn-1Outline elements before being.Otherwise, it repeats
4. and 5. step.
7. profile is by P0..., Pn-2A series of pixel compositions (in such as Fig. 4 shown in (d)).
S304: the area of the profile extracted from the second image is compared with preset area threshold, screens face
Product is greater than profile of the profile of area threshold as target to be sorted.
Due to that may have noise spot in image, it will appear the useless profile in part in this way.It is default to be arranged with preset
Area threshold, for the corresponding meaningless profile of noise to be screened out, Retention area be greater than area threshold profile, as to
The profile of class object.In method provided herein, by the screening to profile, preferably avoid bringing due to noise spot
Interference.
S305: the profile based on target to be sorted pre-processes the first image, to obtain third image.
S306: the principal component of third image is calculated.
S307: by the principal component of the principal component of third image and preset multiple sample images for having carried out type mark into
Row compares, with the affiliated type of determination target to be sorted.
In the present example, step S301, the correlation step phase in any one of S305 to S307 and above-mentioned Fig. 1
Together, the specifically elaboration of part referring to Figure 1, details are not described herein.
Further, referring again to Fig. 3, in another embodiment, after determining the affiliated type of target to be sorted,
Method provided herein further include:
S308: control instruction is generated, so that target to be sorted is moved to setting regions by mechanical structure.
After determining the type of target to be sorted, the mark side based on target to be sorted obtained in above-mentioned steps
To and the angle that is rotated of rotation third image, it is identical along rotation third image opposite direction rotation to generate control mechanical structure
The instruction and clamping of angle simultaneously move the instruction of target to be sorted to corresponding region, and are sent at mechanical structure and either send
To mechanical structure control circuit, the mobile target to be sorted of control mechanical structure is realized to complete to treat the classification of class object.
In technical solution provided herein, before executing to the classification of certain targets to be sorted, it can obtain respectively
Set quantity only includes the image of each target to be sorted, with training acquire the image principal component of the corresponding target to be sorted with
And corresponding partial parameters.Wherein, parameter includes at least:And μ, training is using PCA training, it is possible to understand that,
In other embodiments, the other methods that can also be used are trained.
1. calculating μ first with following formula, calculation formula is as follows:Wherein, current public
In formula, m represents the training sample image quantity of target to be sorted.
2. it is further, after acquiring μ, covariance matrix is calculated, calculation formula is shown below:
Wherein, what x was indicated is the pixel value in picture.
After 3. calculating acquires covariance matrix, based on covariance matrix calculating matrix Σ feature vector: [U S V]=
svd(Σ)。
4. calculating PCA result.Calculation formula is as follows:
U=[μ1,μ2,...,μn]∈Rn×n,Ureduce=U (:, 1:k), wherein 6. the worth determination of k refers to step.
The principal component of image is calculated, following formula are referred to:
Wherein X=[x1-μ,x2-μ,...,xn-μ],Ureduce∈Rn×kIt is k feature vector before U, Y is the first image
PCA's, as a result, y is the PCA calculated result of third image, x is the corresponding pixel value of each third image.
5. calculating orthogonal matrixTransposed matrix.Because of UreduceIt is orthogonal matrix, Ureduce'=Ureduce -1,
Therefore:
xapprox=(Ureduce -1) * Y=(Ureduce -1)-1*Y
=Ureduce*Y
6. selecting k principal component, mean error:
Overall error:
K minimum value is chosen to meet:
The different physical attributes based on target to be sorted are corresponding compared with the prior art designs different and mechanical device, this
A kind of objective classification method based on robot provided by applying, can preferably be applicable in frequent changes working environment and replacement
The situation of classification task, the provided method scope of application is wider, has good robustness, and recognition accuracy is higher.
Fig. 5 is referred to, Fig. 5 is a kind of structural schematic diagram of one embodiment of robot control system of the application.Current real
It applies in example, robot control system 500 includes processor 501, memory 502 and telecommunication circuit 503.Processor 501 and storage
Device 502 and telecommunication circuit 503 are connected with each other.
Wherein, telecommunication circuit 503 is used to for the instruction of processor 501 being sent to the mechanical structure (not shown) of robot,
So that target to be sorted is moved to setting regions by mechanical structure.
For storing program data, which can be executed memory 502 by processor 501.
Processor 501 is used for 502 program data of run memory, to execute method described in Fig. 1 to Fig. 4 as above.
Further, robot control system 500 provided herein can be the control integrated with robot
System, i.e. a robot control system 500 are exclusively used in one robot of control.
In another embodiment, robot control system 500 can be also used for controlling multiple robots.I.e. to be sorted
During target is classified, robot is controlled by robot control system 500, is then executed and is treated class object
Classification task.Specifically, robot control system 500 generates corresponding control after the type for completing to identify target to be sorted
System instruction is sent at robot, so that the mechanical structure in robot rotates corresponding angle, completes to clamp mesh to be sorted
Class object is transferred to the region of setting by mark, completes the classification for treating class object.
Referring to Fig. 6, the application also provides a kind of storage medium 600.The storage medium 600 is stored with program data 601, should
Program data 601 is performed institute in a kind of realization objective classification method based on robot as described above and each embodiment
The method of description.Specifically, the above-mentioned storage medium 600 with store function can be memory, personal computer, service
Device, the network equipment or USB flash disk etc. are one such.
The foregoing is merely presently filed embodiments, are not intended to limit the scope of the patents of the application, all to utilize this
Equivalent structure or equivalent flow shift made by application specification and accompanying drawing content, it is relevant to be applied directly or indirectly in other
Technical field similarly includes in the scope of patent protection of the application.
Claims (12)
1. a kind of objective classification method based on robot, which is characterized in that the described method includes:
The first image comprising target to be sorted is converted, to obtain the second image of preset format；
The profile of the target to be sorted is extracted from second image；
Profile based on the target to be sorted pre-processes the first image, to obtain third image；
Calculate the principal component of the third image；
The principal component of the third image and the principal component of preset multiple sample images for having carried out type mark are compared
It is right, with the affiliated type of the determination target to be sorted.
2. second image is the method according to claim 1, wherein the first image is RGB image
Binary image.
3. the method according to claim 1, wherein
It is described treat class object where the first image converted, the step of the second image to obtain preset format it
Afterwards, further includes:
Processing is optimized to second image, to eliminate the noise spot in second image, wherein the optimization processing
Including corrosion treatment and/or expansion process.
4. the method according to claim 1, wherein described extract the mesh to be sorted from second image
The step of target profile includes:
It extracts and connects that the gray value in second image is identical and continuous pixel, to obtain the target to be sorted
Profile.
5. according to the method described in claim 4, it is characterized in that, described extract the mesh to be sorted from second image
After the step of target profile, further includes:
The area of the profile extracted from second image is compared with preset area threshold, screening area is greater than
Profile of the profile of the area threshold as the target to be sorted.
6. the method according to claim 1, wherein the profile based on the target to be sorted is to described
One image carries out pretreated step
According in the first image with it is to be sorted described in the calculated for pixel values in the profile corresponding region of the target to be sorted
The image reform of target；
Determine the geometric center of the minimum circumscribed rectangle of the profile of the target to be sorted；
The mark direction of the target to be sorted is determined based on described image center of gravity and the geometric center；
Image-region corresponding to the profile of the target to be sorted is extracted in the first image；
Described image region is rotated, so that the mark direction is arranged along preset target direction, and then obtains institute
State third image.
7. according to the method described in claim 6, it is characterized in that, the profile based on the target to be sorted is to described
One image carries out pretreated step
The third image is zoomed in and out, so that the resolution ratio of the third image meets preset resolution requirement.
8. according to the method described in claim 6, it is characterized in that,
The step of principal component for calculating the third image, further comprises:
The pixel value x of the third image is substituted into the principal component y that preset formula calculating acquires the third image；
The preset formula is
It is describedFor calculated separately by all types of sample images acquisition orthogonal matrix transposed matrix, the μ
For the pixel mean value for calculating separately acquisition by all types of sample images；Or, describedFor the multiple sample graph
As the common transposed matrix for calculating the orthogonal matrix obtained, the μ is the image that the multiple sample image calculates acquisition jointly
Mean value.
9. according to the method described in claim 6, it is characterized in that, the affiliated type of the determination target to be sorted it
Afterwards, the method also includes:
Control instruction is generated, so that the target to be sorted is moved to setting regions by mechanical structure.
10. the method according to claim 1, wherein described pair of the first image comprising target to be sorted carries out
It converts, before the second image to obtain preset format further include:
The every type for obtaining setting quantity includes first image of target to be sorted, acquires the target institute to be sorted with training
The corresponding default class object image principal component.
11. a kind of robot control system, which is characterized in that described the system comprises processor, memory and telecommunication circuit
Processor and the memory and the telecommunication circuit are connected with each other；
Wherein, the telecommunication circuit is used to for the instruction of the processor being sent to the mechanical structure of robot, so that described
Target to be sorted is moved to setting regions by mechanical structure；
The memory is for storing program data；
The processor is for running described program data, to execute such as the described in any item methods of claim 1~10.
12. a kind of storage medium, which is characterized in that the storage medium is stored with program data, and described program data are performed
The described in any item methods of Shi Shixian such as claim 1~10.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201811535587.0A CN109815975A (en) | 2018-12-14 | 2018-12-14 | A kind of objective classification method and relevant apparatus based on robot |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201811535587.0A CN109815975A (en) | 2018-12-14 | 2018-12-14 | A kind of objective classification method and relevant apparatus based on robot |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN109815975A true CN109815975A (en) | 2019-05-28 |
Family
ID=66602082
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201811535587.0A Pending CN109815975A (en) | 2018-12-14 | 2018-12-14 | A kind of objective classification method and relevant apparatus based on robot |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN109815975A (en) |
Cited By (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN111178190A (en) * | 2019-12-17 | 2020-05-19 | 中国科学院深圳先进技术研究院 | Target detection method and device based on depth image and storage medium |
| CN112083434A (en) * | 2020-09-09 | 2020-12-15 | 上海有个机器人有限公司 | Laser-based robot detection method and device |
Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN106651831A (en) * | 2016-09-30 | 2017-05-10 | 广西师范大学 | Bamboo piece defect detection method and bamboo piece defect detection system |
| CN106920226A (en) * | 2015-12-25 | 2017-07-04 | 惠州学院 | A kind of lichee mechanical damage method of discrimination |
| CN107482683A (en) * | 2017-09-15 | 2017-12-15 | 东北电力大学 | A kind of wind farm group transient voltage clustering recognition method based on principal component analysis |
| CN108985339A (en) * | 2018-06-21 | 2018-12-11 | 浙江工业大学 | A kind of supermarket's articles from the storeroom method for identifying and classifying based on target identification Yu KNN algorithm |
-
2018
- 2018-12-14 CN CN201811535587.0A patent/CN109815975A/en active Pending
Patent Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN106920226A (en) * | 2015-12-25 | 2017-07-04 | 惠州学院 | A kind of lichee mechanical damage method of discrimination |
| CN106651831A (en) * | 2016-09-30 | 2017-05-10 | 广西师范大学 | Bamboo piece defect detection method and bamboo piece defect detection system |
| CN107482683A (en) * | 2017-09-15 | 2017-12-15 | 东北电力大学 | A kind of wind farm group transient voltage clustering recognition method based on principal component analysis |
| CN108985339A (en) * | 2018-06-21 | 2018-12-11 | 浙江工业大学 | A kind of supermarket's articles from the storeroom method for identifying and classifying based on target identification Yu KNN algorithm |
Cited By (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN111178190A (en) * | 2019-12-17 | 2020-05-19 | 中国科学院深圳先进技术研究院 | Target detection method and device based on depth image and storage medium |
| CN111178190B (en) * | 2019-12-17 | 2023-10-27 | 中国科学院深圳先进技术研究院 | Target detection method and device based on depth image and storage medium |
| CN112083434A (en) * | 2020-09-09 | 2020-12-15 | 上海有个机器人有限公司 | Laser-based robot detection method and device |
| CN112083434B (en) * | 2020-09-09 | 2023-12-29 | 上海有个机器人有限公司 | Laser-based robot detection method and device |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN108520247B (en) | Method, device, terminal and readable medium for identifying object node in image | |
| CN110738207B (en) | Character detection method for fusing character area edge information in character image | |
| CN107038448B (en) | Target detection model construction method | |
| US8265350B2 (en) | Method and apparatus for detecting and processing specific pattern from image | |
| CN111583097A (en) | Image processing method, image processing device, electronic equipment and computer readable storage medium | |
| CN111178190B (en) | Target detection method and device based on depth image and storage medium | |
| US8363983B2 (en) | Real-time face detection apparatus | |
| CN110598788B (en) | Target detection method, target detection device, electronic equipment and storage medium | |
| CN111160140B (en) | Image detection method and device | |
| US20120033873A1 (en) | Method and device for determining a shape match in three dimensions | |
| CN110781882A (en) | License plate positioning and identifying method based on YOLO model | |
| CN104517113A (en) | Image feature extraction method and device and image sorting method and device | |
| CN110008900A (en) | A kind of visible remote sensing image candidate target extracting method by region to target | |
| CN106056138A (en) | Picture processing device and method | |
| CN109815975A (en) | A kind of objective classification method and relevant apparatus based on robot | |
| JP2009211490A (en) | Image recognition method and image recognition device | |
| CN106033613B (en) | Method for tracking target and device | |
| US8971669B2 (en) | Method and apparatus for image processing | |
| CN111860287A (en) | Target detection method and device and storage medium | |
| CN116563306A (en) | Self-adaptive fire trace spectrum image segmentation method and system | |
| JP4550768B2 (en) | Image detection method and image detection apparatus | |
| WO2020174350A1 (en) | Method for verifying the identity of a user by identifying an object within an image that has a biometric characteristic of the user and separating a portion of the image comprising the biometric characteristic from other portions of the image | |
| CN113033593A (en) | Text detection training method and device based on deep learning | |
| CN114627456A (en) | Bill text information detection method, device and system | |
| AlMousawi et al. | Hybrid method for face description using LBP and HOG |
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: 20190528 |
|
| RJ01 | Rejection of invention patent application after publication |