CN110046623A - A kind of characteristics of image point extracting method and camera - Google Patents
A kind of characteristics of image point extracting method and camera Download PDFInfo
- Publication number
- CN110046623A CN110046623A CN201910161155.6A CN201910161155A CN110046623A CN 110046623 A CN110046623 A CN 110046623A CN 201910161155 A CN201910161155 A CN 201910161155A CN 110046623 A CN110046623 A CN 110046623A
- Authority
- CN
- China
- Prior art keywords
- point
- block
- image
- pixels
- feature point
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Granted
Links
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/22—Matching criteria, e.g. proximity measures
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T7/00—Image analysis
- G06T7/10—Segmentation; Edge detection
- G06T7/136—Segmentation; Edge detection involving thresholding
-
- 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
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/20—Special algorithmic details
- G06T2207/20016—Hierarchical, coarse-to-fine, multiscale or multiresolution image processing; Pyramid transform
Abstract
The invention discloses a kind of characteristics of image point extracting method and camera, method includes: the original image for obtaining camera acquisition, forms image pyramid according to original image；The bottom layer image of image pyramid is divided into multiple block of pixels, determines fisrt feature point threshold value corresponding with block of pixels, characteristic point is carried out to block of pixels using fisrt feature point threshold value and detects fisrt feature point；Second feature point threshold value is obtained using bottom layer image information update fisrt feature point threshold value, characteristic point detection is carried out using respective pixel block of the second feature point threshold value to the second tomographic image, obtains second feature point.Primitive character point corresponding with fisrt feature point and second feature point respectively is determined on original image according to image pyramid, obtains extracting result according to the location of pixels screening primitive character point of primitive character point.The embodiment of the present invention meets the discrete distribution of characteristic point in image characteristic point extraction process, scale invariability, requirement of real-time simultaneously.
Description
Technical field
The present invention relates to technical field of image processing, and in particular to a kind of characteristics of image point extracting method and camera.
Background technique
Image characteristics extraction and matching be realize image mosaic, image calibration, target recognition and tracking committed step it
One, it has been widely used in the fields such as three-dimensional reconstruction, vision guided navigation, SLAM.Domestic and foreign scholars have carried out greatly in feature extraction
Quantity research, but focus mostly in the improvement of the robustness such as scale invariability, rotational invariance, and for extracting changing for threshold value setting
Into fairly simple, for example, often using setting fixed threshold or using the multiple feature of multi-threshold for FAST feature point extraction
The mode of extraction realizes discrete distributed effect.It is easy to lead in the violent situation of illumination variation by the way of setting fixed threshold
Cause the characteristic point quantity extracted unstable, and although the mode that characteristic point is repeatedly extracted using multi-threshold is able to solve this and asked
Topic, but extraction time-consuming can be dramatically increased with extraction time.
It follows that the image characteristic point extraction of the prior art cannot meet the discrete distribution of image characteristic point, scale simultaneously
Invariance, requirement of real-time, this technical problem are urgently to be resolved.
Summary of the invention
The present invention provides a kind of characteristics of image point extracting method and cameras, while meeting image characteristic point extraction process
In the discrete distribution of image characteristic point, scale invariability, requirement of real-time.
According to the one aspect of the application, a kind of characteristics of image point extracting method is provided, comprising:
The original image for obtaining camera acquisition forms image pyramid according to the original image；
The pyramidal bottom layer image of described image is divided into multiple block of pixels, determines corresponding with the block of pixels first
Characteristic point threshold value carries out characteristic point detection to the block of pixels according to the fisrt feature point threshold value, obtains fisrt feature point,
In, the bottom layer image is the original image；
According to the information of the next tomographic image adjacent with current tomographic image, each block of pixels with the current tomographic image is obtained
Corresponding second feature point threshold value carries out special according to respective pixel block of the second feature point threshold value to the current tomographic image
Sign point detection, obtains second feature point, wherein the current layer is any of the pyramidal second layer of described image into top layer
Layer；
Screen the primitive character point according to the location of pixels of primitive character point, obtain image characteristic point extract as a result, its
In, the primitive character point includes the fisrt feature point and projects to the spy that original image obtains by the second feature point
Sign point.
According to further aspect of the application, a kind of camera, including camera module and picture processing chip are provided,
The original image is sent to described image processing chip for acquiring original image by the camera module；
Described image handles chip, for obtaining the original image of camera acquisition, forms image according to the original image
Pyramid；The pyramidal bottom layer image of described image is divided into multiple block of pixels, determines corresponding with the block of pixels first
Characteristic point threshold value carries out characteristic point detection to the block of pixels according to the fisrt feature point threshold value, obtains fisrt feature point,
In, the bottom layer image is the original image；Using the information of the next tomographic image adjacent with current tomographic image, acquisition and institute
The corresponding second feature point threshold value of each block of pixels for stating current tomographic image, according to the second feature point threshold value to the current layer
The respective pixel block of image carries out characteristic point detection, obtains second feature point, wherein the current layer is described image pyramid
Any layer of the second layer into top layer；The primitive character point is screened according to the location of pixels of the primitive character point, is obtained
Image characteristic point extracts result, wherein the primitive character point includes the fisrt feature point and by the second feature point
Project to the characteristic point that original image obtains.
The utility model has the advantages that meeting graphical rule not by constructing image pyramid using the technical solution of the embodiment of the present invention
Denaturation requires, by calculating the feature point extraction threshold value that threshold value determines each piece, other tomographic images of pyramid to original image piecemeal
Then according to upper tomographic image information update threshold value, the problem unstable using characteristic point quantity caused by fixed threshold is avoided,
Screening finally is carried out to all characteristic points and guarantees discrete distribution.The method of the present embodiment only extracts a characteristic point with FAST,
Real-time is guaranteed, and generalization is good, especially in the uneven scene of grain distribution (such as a large amount of white walls, snowfield) and illumination
In the case of acute variation, biggish advantage is embodied.
Detailed description of the invention
Fig. 1 is the flow chart of the characteristics of image point extracting method of one embodiment of the invention；
Fig. 2 is the flow diagram of the characteristics of image point extracting method of another embodiment of the present invention；
Fig. 3 is the characteristic point quaternary tree building flow chart of one embodiment of the invention；
Fig. 4 is the block diagram of the camera of one embodiment of the invention.
Specific embodiment
In order to make the foregoing objectives, features and advantages of the present invention clearer and more comprehensible, with reference to the accompanying drawing and specific real
Applying mode, the present invention is described in further detail.Obviously, described embodiments are some of the embodiments of the present invention, without
It is whole embodiments.Based on the embodiments of the present invention, those of ordinary skill in the art are not before making creative work
Every other embodiment obtained is put, shall fall within the protection scope of the present invention.
Design concept of the invention is: the embodiment of the present invention is by image block threshold value, extraction characteristic point, then makes
With tree structure storage characteristic point, low-response value tag point is rejected, guarantees that characteristic point quantity is stable and discrete, be uniformly distributed, prevent
Only characteristic point aggregation degenerates to the appearance of the coplanar situation of characteristic point, is conducive to the successful match for improving characteristic point between different images
The precision of rate and relative transform matrix.Primitive image features point is extracted by design threshold adaptive algorithm, is inserted using bilinearity
The characteristic point on high-rise image is extracted after the threshold value of the determining image pyramid high level of value again, is uniformly divided in guarantee image characteristic point
Under conditions of cloth, characteristic point meet scale invariability, the time-consuming of feature point extraction is greatly reduced, there is very high application value.
Fig. 1 is the flow chart of the characteristics of image point extracting method of one embodiment of the invention, referring to Fig. 1, the present embodiment
Characteristics of image point extracting method includes:
Step S101 obtains the original image of camera acquisition, forms image pyramid according to the original image；
The pyramidal bottom layer image of described image is divided into multiple block of pixels, the determining and block of pixels by step S102
Corresponding fisrt feature point threshold value carries out characteristic point detection to the block of pixels according to the fisrt feature point threshold value, obtains the
One characteristic point, wherein the bottom layer image is the original image；
Step S103 is obtained and the current tomographic image according to the information of next tomographic image adjacent with current tomographic image
The corresponding second feature point threshold value of each block of pixels, according to the second feature point threshold value to the corresponding picture of the current tomographic image
Plain block carries out characteristic point detection, obtains second feature point, wherein the current layer is that the pyramidal second layer of described image extremely pushes up
Any layer in layer；
Step S104 screens the primitive character point according to the location of pixels of primitive character point, obtains image characteristic point and mention
Take result, wherein the primitive character point includes the fisrt feature point and projects to original graph by the second feature point
As obtained characteristic point.
As shown in Figure 1 it is found that the characteristics of image point extracting method of the present embodiment, obtains original image, according to original image
Image pyramid is formed, meets the requirement of graphical rule invariance by constructing image pyramid.By the bottom figure of image pyramid
As being divided into multiple block of pixels, feature point extraction is carried out after determining fisrt feature point threshold value corresponding with block of pixels, by original
Beginning image block calculates the feature point extraction threshold value that threshold value determines each piece, other tomographic images of pyramid are then believed according to a upper tomographic image
Breath updates threshold value, avoids the problem unstable using characteristic point quantity caused by fixed threshold.According to the picture of primitive character point
Primitive character point is screened in plain position, i.e., carries out screening to all characteristic points and guarantee discrete distribution.The present embodiment is only mentioned with FAST
A characteristic point is taken, real-time is guaranteed, and generalization is good, especially (such as a large amount of white in the uneven scene of grain distribution
Wall, snowfield) and illumination acute variation in the case of.
In order to guarantee that the discrete distribution of image characteristic point, scale and direction invariance, real-time etc. require, the embodiment of the present invention
Provide the discrete distribution characteristics point quick extraction method based on adaptive threshold FAST feature extraction and quaternary tree screening, the party
Method is to original image piecemeal calculating local mean value and variance so that it is determined that the FAST feature point extraction threshold value of each block image, pyramid
Other tomographic images then update threshold value according to upper one layer of local mean value and covariance information, finally utilize building to all characteristic points
Quaternary tree carries out screening and completes to extract.
The step of the present embodiment, specifically includes that
Step 2-1 constructs image pyramid.
In order to meet the scale invariability of characteristic point, the present embodiment obtains the original image of camera acquisition, according to original graph
As building image pyramid, that is, next using being put into after Gaussian convolution and linear scale using original image as pyramid bottom
Layer recycles Level times altogether, the image pyramid of Level layers of building.
Step 2-2, bottom layer image adaptive threshold calculate.
It has been observed that the bottom layer image of image pyramid is the original image without any processing.Bottom layer image it is adaptive
The bottom layer image of image pyramid is specifically divided into multiple block of pixels by the calculating of threshold value, determines corresponding with block of pixels first
Characteristic point threshold value carries out characteristic point detection to the block of pixels according to the fisrt feature point threshold value, obtains fisrt feature point.?
That is first dividing the image into the block image of same size, local mean value and mean square deviation are calculated to each block image, wherein
Mean square deviation can be used aggregate value method of weighting component mode and be accelerated.FAST feature extraction threshold value then is obtained according to mean value and variance,
And the bound of threshold range is set, it adapts to FAST-9 and extracts characteristic point.
Note: calculating mean square deviation using aggregate value method of weighting component accelerometer is specifically that following formula is utilized to calculate mean square deviation:
Step 2-3 uses interpolation processing image pyramid.
In one embodiment, according to the information of the next tomographic image adjacent with current tomographic image, obtain and the current layer
The corresponding second feature point threshold value of each block of pixels of image, according to the second feature point threshold value to the phase of the current tomographic image
It answers block of pixels to carry out characteristic point detection, obtains second feature point, wherein the current layer is the pyramidal second layer of described image
Any layer into top layer.That is, after the feature extraction threshold value for calculating bottom layer image, according to the information of bottom layer image
Calculate the feature extraction threshold value of the 2nd tomographic image, and so on until image pyramid top layer.For example, using bilinear interpolation
The local mean value and mean square deviation of algorithm and specific other layers of more new strategy sequential update, and determine with this all layers of each block diagram
The FAST-9 threshold value of picture simultaneously completes feature point extraction work.
Bilinear interpolation principle is as follows: setting liLayer is based on n*m block image FAST feature extraction threshold obtained by local mean square deviation
Value Tfast[n*m], then li+1Layer has (n/s) * (m/s) piecemeal.For li+1(x, y) block image in layer, by the block block image
Central projection is to liLayer, coordinate are (x0,y0), it is new according to being calculated to adjacent 4 pieces of picture centre distances using bilinear interpolation
Mean μ and variance var.
Interpolation calculation process is as follows:
Step 231, bilinear interpolation calculates threshold value.
According to the block of pixels of the scaling and the bottom layer image of the bottom layer image and second tomographic image
Number, multiple neighborhood blocks of block of pixels where determining the second central point of second tomographic image, wherein second central point by
The first nodal point of the bottom layer image projects to obtain；According to the mean value of each neighborhood block, and using bilinear interpolation to institute
Block of pixels carries out interpolation where stating the second central point, the second mean value of block of pixels where obtaining second central point, and by institute
The mean value for stating the second mean value and each neighborhood block obtains second variance；More based on second mean value and the second variance
The new fisrt feature point threshold value, obtains second feature point threshold value.
For example, coordinate is (x0,y0) the adjacent four pieces of image mean values of point be μ11=(x1,y1),μ12=(x1,y2),μ21=
(x2,y1),μ22=(x2,y2),
Linear interpolation first is carried out in the direction x, is obtained:
Then interpolation is carried out to the direction y, obtained:
Final updating mean square deviation
It should be noted that the bilinear interpolation in abovementioned steps 2-3 is it is possible that Tfast=-1, image mean-squared deviation
The special circumstances of Var=0, at this moment, local mean value update are still averaged.For example, setting the block of Var=0 as block a3, other blocks
For a0,a1,a2, mean value update is identical as the mean value update step in aforementioned interpolation arithmetic, and the difference of mean square deviation is with 4 points
The mean value of block is as a3Local mean square deviation calculation basis:
Step 2-4, FAST feature extraction and screening.
In one embodiment, the second of each layer in the fisrt feature point and upper layer images being extracted on bottom layer image
After characteristic point, the primitive character point on the corresponding bottom layer image of second feature point is determined, since first on bottom layer image is special
Sign point does not need to do the high-rise characteristic point to bottom and maps, thus directly using the fisrt feature of the extraction on bottom layer image point as
Primitive character point.That is, according to image pyramid, determine on the original image respectively with the fisrt feature point and described second
The corresponding primitive character point of characteristic point screens primitive character point according to the location of pixels of primitive character point, obtains image characteristic point
Extract result.For example construction feature point quaternary tree storage organization carries out node split, ties after splitting into the feature point extraction upper limit
Beam building.It to the characteristic point stored in each leaf node of quad-tree structure, sorts according to Harris response, each node only takes
Maximum value is as output result.In this way, ensure that the characteristic point finally extracted meets the requirement of discrete distribution on the image, it is convenient
Camera relative motion is resolved with Feature Points Matching, improves accuracy.
It should be noted that in abovementioned steps 2-1 when constructing image pyramid, only using realization image scale transform
One linear kernel function Gaussian convolution, for the scale space of two dimensional image is defined as:
L (x, y, σ)=G (x, y, σ) * I (x, y)
Here gaussian kernel function σ=2 are taken.
Setting original image is pyramid bottom, makees to be put into after Gaussian convolution and linear scale to each tomographic image next
Layer recycles Level times altogether, constructs Level=4 layers, the image pyramid of down-sampling scaling s=1.2.
In addition, it is identical that bottom layer image is divided into multiple sizes when calculating threshold value to image block in abovementioned steps 2-2
Block of pixels judges whether the block of pixels is abnormal mass according to the gray value of pixel in each block of pixels；If the picture
Plain block is abnormal mass, then the variance of the block of pixels is set as first constant, and the fisrt feature point threshold value is set as second often
Number, and the abnormal mass is detected without characteristic point；If the block of pixels is not abnormal mass, according in the block of pixels
The gray value of pixel calculates mean value and variance, and according to the mean value and the variance, obtains the fisrt feature point threshold value.
For example, firstly, according to 30 × 30 picture sizes to image block, since subsequent extracted FAST characteristic point needs to be traversed for
Around to each pixel radius be 3 circle, therefore after piecemeal as feature extraction image it is practical be 36*36 (expand boundary
Border=3).
Secondly, using I to the gray level image of every piece of 36*36 sizeijIndicate the pixel value at the position (i, j), then the image block
Interior local mean values and local mean square deviation, are shown below:
Here part is the part for monolith image.
Again, FAST feature extraction reference threshold is calculated by local mean values and mean square deviation, for example passes through formulaCalculate reference threshold.
In the present embodiment, the determination fisrt feature point threshold value corresponding with the block of pixels the step of after, the party
Method further include: the fisrt feature point threshold value is compared with the upper limit value of preset threshold range and lower limit value respectively and is compared
Relatively result；If the comparison result is the fisrt feature point threshold value in the threshold range, not to the fisrt feature
Point threshold value is modified；If the comparison result is that the fisrt feature point threshold value is greater than the upper limit value, by the upper limit
Value is assigned to the fisrt feature point threshold value, if the comparison result is that the fisrt feature point threshold value is less than the lower limit value,
The lower limit value is then assigned to the fisrt feature point threshold value.For example, bound T ∈ is arranged to gained reference threshold T is calculated
[10,30] final threshold value T is obtainedfast。
Here the gray value according to pixel in each block of pixels judges whether the block of pixels is that abnormal mass includes:
To the gray value of pixel in the block of pixels according to sorting from large to small, the mean value of preset number gray value is obtained before calculating
First mean value, and the mean value of preset number gray value obtains the second mean value after calculating；Judge first mean value and described
Whether the difference of two mean values is less than predetermined threshold value；It is then, to determine that the block of pixels is abnormal mass, otherwise, it determines the pixel
Block is not abnormal mass.
3 gray value maximum values and 3 minimum gray values are extracted when for example, calculating block image mean value, if maximum value is equal
Value and minimum value mean value difference then remember threshold value T less than 5fast=-1, image mean-squared deviation Var=0, the image block is without spy
Sign point extracts.If maximum value mean value and minimum value mean value difference show that this block image may be that grain distribution is uneven less than 5
Region, do not have characteristic point.
In addition, it is necessary to be according to the primitive character in explanation, abovementioned steps 2-4, FAST feature extraction and screening
The location of pixels screening primitive character point of point when specific implementation, constructs the Storage Structure of Tree of the original image, will be described
Original image is divided into multiple block of pixels, and the primitive character point in each block of pixels is correspondingly placed into the tree-like storage
In the node of structure；According to the number loop fission of the primitive character point in the node, newly-increased node is obtained, it will be described newly-increased
Node, which is added in the Storage Structure of Tree, obtains first node list；Obtain institute in each node of the first node list
The response for stating primitive character point sorts to the primitive character point according to the response size, filters out and make number one
The primitive character point.
The building and cyclic process of quaternary tree are: according to the number of the point of primitive character described in the node of start node list
Whether loop fission, the primitive character point number specifically included in the first node for judging the start node list are greater than
1, it is then, N number of second node to be generated, and block of pixels corresponding with the first node is divided into N number of subregion, by the N
The primitive character point in sub-regions is correspondingly placed into N number of second node, wherein N is even number；By the original spy
It levies the second node that point number is not 0 and is used as newly-increased node, be added in the start node list, obtain first node
List length；Otherwise, the first node is deleted from the start node list, obtains first node list length；Judgement
Whether the first node list length is greater than or equal to pre-set length threshold, is then, to stop division and obtain the first node
The step of list, otherwise return executes the number loop fission of the point of the primitive character according to the node of start node list.
From the foregoing, it will be observed that the Select to use of characteristic point quaternary tree realization in the present embodiment, before guaranteeing that characteristic point is spread
It puts and improves angle point stability and quality.When quaternary tree leaf node reaches NmaxWhen, tree structure stops bifurcated building, right
Each leaf node preservative feature point of quad-tree structure sorts according to response, and each node only takes maximum characteristic point.
In one embodiment, quaternary tree screening characteristic point specifically includes following sub-step:
Step 241, initialisation structures body.Each node Node includes rectangle frame information and image in the rectangle frame
All characteristic points, original image divide equally 4 pieces and are put into queue, and queue Q is as initialization entrance, initial queue length L=4.
Step 242, node loop fission.Queue Q interior joint Node is successively taken from the beginning to the endi, by rectangle in the structural body
4 pieces are divided into, constitutes new node with the characteristic point in rectangle after segmentation, if deleting the section without characteristic point in newly-generated node
Point, other are put into the head of queue Q, and queue length updates (division increases by 1~3 every time).For being newly put into the structure on head
Body NodeiJudge whether it continues to divide and (judge according to feature point number in node).
Step 243, determine whether to terminate, when all Q interior nodes Node can not divide or queue nodes number L >=N againmax,
Quaternary tree building terminates.
Step 244, characteristic point sorts in node, exports result.All Node are traversed, wherein characteristic point response maximum is taken
Characteristic point as the node export.
Fig. 2 is the flow diagram of the characteristics of image point extracting method of another embodiment of the present invention, referring to fig. 2 to this hair
The characteristics of image point extracting method of bright embodiment is described further,
The specific implementation steps are as follows: process starts,
Step 1 is executed, the image for being used for feature extraction is obtained.
Firstly, being input image to be processed, for example original image is obtained from camera, original image is changed into grayscale image,
Original image is, for example, the image of a 640*480 size.Here camera is for acquiring original image.
Step 2, fragmental image processing；
In one embodiment, 30*30 is set by the starting point step size of original image piecemeal, takes 36* from starting point every time
The piecemeal of 36 sizes.That is, first, in accordance with 30*30 picture size piecemeal, due to subsequent extracted FAST characteristic point need to be traversed for it is each
Around pixel radius be 3 circle, therefore after piecemeal as feature extraction image it is practical be 36*36 (be equivalent to expand boundary
Border=3).
Step 210 is executed, the difference △ of block image mean value, maxima and minima is calculated；
To the gray level image of every piece of 36 × 36 sizes, I is usedijIt indicates the pixel value at the position (i, j), traverses block image,
Seek three maximum value μ _ max μ _ max μ _ max, three minimum value mean μ _ min and local mean μ.
Step 211, judge △ whether less than 5；It is that then otherwise threshold value=- 1 variance=0, execution step 212 execute
Step 213；
Here it is the difference △ for calculating μ _ max and μ _ min, judges △ whether less than 5, if determining the equal of the piecemeal less than 5
T is arranged in variance Var=0, threshold valuefast=-1, without feature point extraction；If more than 5, then integrogram and variance are calculated.
Step 212, FAST-9 extracts special characteristic point；
According in previous step mean value and variance calculate FAST-9 threshold value (due to image pyramid bottom layer image be original
Beginning image, so the threshold value of original image, that is, fisrt feature point threshold value), calculation formula isIn the step for determining fisrt feature point threshold value corresponding with block of pixels
After rapid, this method further include by the fisrt feature point threshold value respectively with preset threshold range (such as the model that 10-30 is limited
Enclose) upper limit value and lower limit value be compared to obtain comparison result and determine final threshold value, for example, according to commissioning experience, k1∈[1,
2], k2K should be traditionally arranged to be according to the brightness regulation of usage scenario2=1.5, last threshold limit Tfast=T ∈ [10,
30].Characteristic point: FAST (InputImage, KeyPoints, Tfast, true) is extracted using the FAST function in OpenCV.
Step 213, accelerated to calculate mean square deviation with integrogram；
Calculate integrogramAnd mean square deviation is calculated using integrogram, formula is as follows:
Step 214, threshold value is calculated with mean value and mean square deviation, executes step 212 later.
Step 3, Gaussian Blur and down-sampling construct pyramid；
Recycle Level=4 and construct image pyramid, since bottom to image using OpenCV function resize and
GaussianBlur obtains a high tomographic image.resize(Image[level-1],Image[level],newSize,0,0,
INTER_LINEAR)GaussianBlur(Image[level-1],Image[level],Size(5,5),2,2,BORDER_
REFLECT_101)。
Step 4, bilinearity difference update threshold value FAST and extract characteristic point；
For pyramid layer block image center (x, y), coordinate in a tomographic image is projected to, according to adjacent thereto
The mean value of four blocks of images first carries out linear interpolation in the direction y after the direction x carries out linear interpolation, finally according to mean square deviation.Here
Interpolation sequence it is unlimited.Y directional interpolation can first be carried out.In addition, when there is adjacent piecemeal Var=0, if piecemeal is a3,
Other blocks are a0,a1,a2, mean value update with it is identical under normal circumstances, and mean square deviation calculating is by a3It rejects, is not involved in calculating.Other
Degenerate case is similar with the calculating of aforementioned mean square deviation.Use calculating gained Var (x0,y0) and μ (x0,y0) update threshold value T=k1*Then characteristic point is extracted using FAST function.
Step 5, construction feature point quaternary tree；
With reference to Fig. 3, first initialisation structures body.Each node Node includes all features of the image in the rectangle frame
Original image is divided equally 4 pieces in the present embodiment and is put into queue by point, and queue Q is as initialization entrance, initial queue length L=
4。
Node loop fission.Referring to Fig. 3, queue Q interior joint Node is successively taken from the beginning to the endi, by rectangle in the structural body
4 pieces are divided into, constitutes new node with the characteristic point in rectangle after segmentation, if deleting the section without characteristic point in newly-generated node
Point, other are put into the head of queue Q, and (proper splitting increases by 3 nodes, occurs then being less than 3 when knot removal for queue length update
It is a).For being newly put into the node Node on headiJudge whether it continues to divide and (judge according to feature point number in node).
End mark position can not all divide again for all Q interior nodes Node or queue nodes number L >=Nmax。
After circulation terminates, in node characteristic point sort, only take wherein the maximum characteristic point of characteristic point response as the section
Point output.
1, the invention proposes one kind be based on fast discrete Feature Points Extraction, can meet extract characteristic point from
Distribution, characteristic point robustness are dissipated, can be worked normally in visual odometry, the front end SLAM and Visual Tracking System.2, this hair
It is bright that threshold value is calculated using pretreatment, a characteristic point only is extracted with FAST, real-time is guaranteed, and generalization is preferable, special
It is not to be embodied biggish excellent in the uneven scene of grain distribution (such as a large amount of white walls, snowfield) and illumination acute variation
Gesture.The realization step of the method for the embodiment of the present invention is stressed with a specific application scenarios below.Referring to fig. 2, once
Specific image characteristic point extraction pair
It should be noted that the description and claims of this application and term " first " in above-mentioned attached drawing, "
Two " etc. be to be used to distinguish similar objects, without being used to describe a particular order or precedence order.It should be understood that using in this way
Object be interchangeable under appropriate circumstances, so as to the embodiment of the present invention described herein can in addition to illustrating herein or
Sequence other than those of description is implemented.
Fig. 4 is the block diagram of the camera of one embodiment of the invention, and referring to fig. 4, the camera 400 of the present embodiment includes: camera shooting
Head mould group 401 and picture processing chip 402,
The original image is sent to described image processing core for acquiring original image by the camera module 401
Piece 402；
Described image handles chip 402, for obtaining the original image of camera acquisition, is formed and is schemed according to the original image
As pyramid；The pyramidal bottom layer image of described image is divided into multiple block of pixels, determines corresponding with the block of pixels the
One characteristic point threshold value carries out characteristic point detection to the block of pixels according to the fisrt feature point threshold value, obtains fisrt feature point,
Wherein, the bottom layer image is the original image；Using the information of the next tomographic image adjacent with current tomographic image, obtain with
The corresponding second feature point threshold value of each block of pixels of the current tomographic image, according to the second feature point threshold value to described current
The respective pixel block of tomographic image carries out characteristic point detection, obtains second feature point, wherein the current layer is described image gold word
Any layer of the second layer of tower into top layer；The primitive character point is screened according to the location of pixels of the primitive character point, is obtained
Result is extracted to image characteristic point, wherein the primitive character point includes the fisrt feature point and by the second feature
Point projects to the characteristic point that original image obtains.
In one embodiment of the invention, described image processing chip 402 is specifically used for according to next tomographic image
With the scaling of the current tomographic image and the block of pixels number of next tomographic image, the current tomographic image is determined
The second central point where block of pixels multiple neighborhood blocks, wherein second central point is by the first of next tomographic image
Central point projects to obtain；According to the mean value of each neighborhood block, and using bilinear interpolation to picture where second central point
Plain block carries out interpolation, the second mean value of block of pixels where obtaining second central point, and by second mean value and each institute
The mean value for stating neighborhood block obtains second variance；Based on second mean value and the second variance, second feature point threshold value is obtained.
In one embodiment of the invention, described image processing chip 402 is specifically used for constructing the original image
The original image is divided into multiple block of pixels by Storage Structure of Tree, by the primitive character point in each block of pixels
It is correspondingly placed into the node of the Storage Structure of Tree；According to the number loop fission of the primitive character point in the node, obtain
To newly-increased node, the newly-increased node is added in the Storage Structure of Tree and obtains first node list；Obtain described
The response of primitive character point described in each node of one node listing, according to the response size to the primitive character point
Sequence, filters out the primitive character point to make number one.
In one embodiment of the invention, described image processing chip 402 is specifically used for, and the bottom layer image is divided
For the identical block of pixels of multiple sizes, judge whether the block of pixels is different according to the gray value of pixel in each block of pixels
Normal block；If the block of pixels is abnormal mass, the variance of the block of pixels is set as first constant, by the fisrt feature point
Threshold value is set as second constant, and detects to the abnormal mass without characteristic point；If the block of pixels is not abnormal mass, root
Mean value and variance are calculated according to the gray value of pixel in the block of pixels, and according to the mean value and the variance, described in acquisition
Fisrt feature point threshold value.
In one embodiment of the invention, described image handles chip 402 specifically to pixel in the block of pixels
Gray value is according to sorting from large to small, and the mean value of preset number gray value obtains the first mean value before calculating, and presets after calculating
The mean value of number gray value obtains the second mean value；It is pre- to judge whether the difference of first mean value and second mean value is less than
Gating limit value；It is then, to determine that the block of pixels is abnormal mass, otherwise, it determines the block of pixels is not abnormal mass.
In one embodiment of the invention, described image processing chip 402 is by the fisrt feature point threshold value or described
Second feature point threshold value is compared to obtain comparison result with the upper limit value of preset threshold range and lower limit value respectively；If the ratio
Relatively result is the fisrt feature point threshold value or second feature point threshold value in the threshold range, then not to described first
Characteristic point threshold value or second feature point threshold value are modified；If the comparison result is the fisrt feature point threshold value or institute
Second feature point threshold value is stated greater than the upper limit value, then the upper limit value is assigned to the fisrt feature point threshold value or described the
Two characteristic point threshold values, if the comparison result be the fisrt feature point threshold value or second feature point threshold value be less than it is described under
The lower limit value is then assigned to the fisrt feature point threshold value or second feature point threshold value by limit value.
In one embodiment of the invention, node of the described image processing chip 402 with specific reference to start node list
Described in primitive character point number loop fission, specifically include described in the first node for judging the start node list
Whether primitive character point number is greater than 1, is then, to generate N number of second node, and will block of pixels corresponding with the first node etc.
It is divided into N number of subregion, the primitive character point in N number of subregion is correspondingly placed into N number of second node,
In, N is even number；Using the primitive character point number be not 0 the second node as newly-increased node, be added to described initial
In node listing, first node list length is obtained；Otherwise, the first node is deleted from the start node list, obtained
To first node list length；Judge whether the first node list length is greater than or equal to pre-set length threshold, is then, to stop
Only division obtains the first node list, otherwise returns and executes the point of the primitive character according to the node of start node list
Number loop fission the step of.
It should be noted that the illustration of each function performed by picture processing chip is said in the camera shown in Fig. 4
It is bright, it is consistent with the illustration explanation in preceding method embodiment, it no longer repeats one by one here.
A kind of computer readable storage medium is also provided in one embodiment of the present of invention, computer readable storage medium is deposited
Computer instruction is stored up, computer instruction makes the computer execute above-mentioned method.
It should be understood by those skilled in the art that, the embodiment of the present invention can provide as method, system or computer program
Product.Therefore, complete hardware embodiment, complete software embodiment or reality combining software and hardware aspects can be used in the present invention
Apply the form of example.Moreover, it wherein includes the computer of computer usable program code that the present invention, which can be used in one or more,
The computer program implemented in usable storage medium (including but not limited to magnetic disk storage, CD-ROM, optical memory etc.) produces
The form of product.
The present invention be referring to according to the method for the embodiment of the present invention, the process of equipment (system) and computer program product
Figure and/or block diagram describe.It should be understood that every one stream in flowchart and/or the block diagram can be realized by computer program instructions
The combination of process and/or box in journey and/or box and flowchart and/or the block diagram.It can provide these computer programs
Instruct the processor of general purpose computer, special purpose computer, Embedded Processor or other programmable data processing devices to produce
A raw machine, so that being generated by the instruction that computer or the processor of other programmable data processing devices execute for real
The dress for the function of being specified in one box or multiple boxes of one process or multiple processes and/or block diagrams of present flow chart
It sets.
It should be noted that the terms "include", "comprise" or its any other variant are intended to the packet of nonexcludability
Contain, so that the process, method, article or equipment for including a series of elements not only includes those elements, but also including
Other elements that are not explicitly listed, or further include for elements inherent to such a process, method, article, or device.
In the absence of more restrictions, the element limited by sentence "including a ...", it is not excluded that including the element
Process, method, article or equipment in there is also other identical elements.
In specification of the invention, numerous specific details are set forth.Although it is understood that the embodiment of the present invention can
To practice without these specific details.In some instances, well known method, structure and skill is not been shown in detail
Art, so as not to obscure the understanding of this specification.Similarly, it should be understood that disclose in order to simplify the present invention and helps to understand respectively
One or more of a inventive aspect, in the above description of the exemplary embodiment of the present invention, each spy of the invention
Sign is grouped together into a single embodiment, figure, or description thereof sometimes.However, should not be by the method solution of the disclosure
It is interpreted into and reflects an intention that i.e. the claimed invention requires more than feature expressly recited in each claim
More features.More precisely, just as the following claims reflect, inventive aspect is single less than disclosed above
All features of embodiment.Therefore, it then follows thus claims of specific embodiment are expressly incorporated in the specific embodiment party
Formula, wherein each, the claims themselves are regarded as separate embodiments of the invention.
The above description is merely a specific embodiment, under above-mentioned introduction of the invention, those skilled in the art
Other improvement or deformation can be carried out on the basis of the above embodiments.It will be understood by those skilled in the art that above-mentioned tool
Body description only preferably explains that the purpose of the present invention, protection scope of the present invention are subject to the protection scope in claims.
Claims (10)
1. a kind of characteristics of image point extracting method characterized by comprising
The original image for obtaining camera acquisition forms image pyramid according to the original image；
The pyramidal bottom layer image of described image is divided into multiple block of pixels, determines fisrt feature corresponding with the block of pixels
Point threshold value carries out characteristic point detection to the block of pixels according to the fisrt feature point threshold value, obtains fisrt feature point, wherein
The bottom layer image is the original image；
According to the information of the next tomographic image adjacent with current tomographic image, obtain corresponding with each block of pixels of the current tomographic image
Second feature point threshold value, characteristic point is carried out to the respective pixel block of the current tomographic image according to the second feature point threshold value
Detection, obtains second feature point, wherein the current layer is any layer of the pyramidal second layer of described image into top layer；
The primitive character point is screened according to the location of pixels of primitive character point, image characteristic point is obtained and extracts result, wherein institute
Primitive character point is stated to include the fisrt feature point and project to the characteristic point that original image obtains by the second feature point.
2. the method according to claim 1, wherein described be divided into the pyramidal bottom layer image of described image
Multiple block of pixels determine that fisrt feature point threshold value corresponding with the block of pixels includes:
The bottom layer image is divided into the identical block of pixels of multiple sizes, according to the gray value of pixel in each block of pixels
Judge whether the block of pixels is abnormal mass；
If the block of pixels is abnormal mass, the variance of the block of pixels is set as first constant, by the fisrt feature point
Threshold value is set as second constant, and detects to the abnormal mass without characteristic point；
If the block of pixels is not abnormal mass, mean value and variance are calculated according to the gray value of pixel in the block of pixels,
And according to the mean value and the variance, the fisrt feature point threshold value is obtained.
3. according to the method described in claim 2, it is characterized in that, the gray value according to pixel in each block of pixels
Judge whether the block of pixels is that abnormal mass includes:
To the gray value of pixel in the block of pixels according to sorting from large to small, the mean value of preset number gray value before calculating
The first mean value is obtained, and the mean value of preset number gray value obtains the second mean value after calculating；
Judge whether first mean value and the difference of second mean value are less than predetermined threshold value；
It is then, to determine that the block of pixels is abnormal mass, otherwise, it determines the block of pixels is not abnormal mass.
4. the method according to claim 1, wherein next tomographic image that the basis is adjacent with current tomographic image
Information, obtaining corresponding with each block of pixels of current tomographic image second feature point threshold value includes:
According to the block of pixels of the scaling and next tomographic image of next tomographic image and the current tomographic image
Number, multiple neighborhood blocks of block of pixels where determining the second central point of the current tomographic image, wherein second central point by
The first nodal point of next tomographic image projects to obtain；
According to the mean value of each neighborhood block, and using bilinear interpolation block of pixels where second central point is carried out slotting
Value, the second mean value of block of pixels where obtaining second central point, and by second mean value and each neighborhood block
Mean value obtains second variance；
Based on second mean value and the second variance, second feature point threshold value is obtained.
5. the method according to claim 1, wherein this method further include:
By the fisrt feature point threshold value or second feature point threshold value respectively with the upper limit value of preset threshold range and lower limit
Value is compared to obtain comparison result；
If the comparison result is the fisrt feature point threshold value or second feature point threshold value in the threshold range,
It does not modify to the fisrt feature point threshold value or second feature point threshold value；
If the comparison result is that the fisrt feature point threshold value or second feature point threshold value are greater than the upper limit value, will
The upper limit value is assigned to the fisrt feature point threshold value or second feature point threshold value,
If the comparison result is that the fisrt feature point threshold value or second feature point threshold value are less than the lower limit value, will
The lower limit value is assigned to the fisrt feature point threshold value or second feature point threshold value.
6. the method according to claim 1, wherein described in the location of pixels screening according to primitive character point
Primitive character point includes:
The original image is divided into multiple block of pixels by the Storage Structure of Tree for constructing the original image, by each picture
The primitive character point in plain block is correspondingly placed into the node of the Storage Structure of Tree；
According to the number loop fission of the primitive character point in the node, newly-increased node is obtained, the newly-increased node is added
First node list is obtained into the Storage Structure of Tree；
The response for obtaining primitive character point described in each node of the first node list, according to the response size pair
The primitive character point sequence, filters out the primitive character point to make number one.
7. according to the method described in claim 6, it is characterized in that, the number according to the primitive character point in the node
Loop fission obtains newly-increased node, the newly-increased node is added in the Storage Structure of Tree and obtains first node list
Include:
According to the number loop fission of the point of primitive character described in the node of start node list, specifically include judge it is described initial
Whether the primitive character point number in the first node of node listing is greater than 1, is then, to generate N number of second node, and will be with
The corresponding block of pixels of the first node is divided into N number of subregion, and the primitive character point in N number of subregion is corresponding
It is put into N number of second node, wherein N is even number；The second node that the primitive character point number is not 0 is made
To increase node newly, it is added in the start node list, obtains first node list length；Otherwise, from the start node
The first node is deleted in list, obtains first node list length；
Judge whether the first node list length is greater than or equal to pre-set length threshold, is that then, stopping division obtaining described
Otherwise first node list returns to the number loop fission for executing the point of the primitive character according to the node of start node list
The step of.
8. a kind of camera, which is characterized in that including camera module and picture processing chip,
The original image is sent to described image processing chip for acquiring original image by the camera module；
Described image handles chip, for obtaining the original image of camera acquisition, forms image gold word according to the original image
Tower；The pyramidal bottom layer image of described image is divided into multiple block of pixels, determines fisrt feature corresponding with the block of pixels
Point threshold value carries out characteristic point detection to the block of pixels according to the fisrt feature point threshold value, obtains fisrt feature point, wherein
The bottom layer image is the original image；Using the information of the next tomographic image adjacent with current tomographic image, obtain with it is described
The corresponding second feature point threshold value of each block of pixels of current tomographic image, according to the second feature point threshold value to the current layer figure
The respective pixel block of picture carries out characteristic point detection, obtains second feature point, wherein the current layer is that described image is pyramidal
Any layer of the second layer into top layer；The primitive character point is screened according to the location of pixels of the primitive character point, obtains figure
As feature point extraction result, wherein the primitive character point includes the fisrt feature point and thrown by the second feature point
The characteristic point that shadow is obtained to original image.
9. camera according to claim 8, which is characterized in that
Described image handles chip, specifically for the scaling according to next tomographic image and the current tomographic image, with
And the block of pixels number of next tomographic image, multiple neighbours of block of pixels where determining the second central point of the current tomographic image
Domain block, wherein second central point is projected to obtain by the first nodal point of next tomographic image；According to each neighborhood block
Mean value, and interpolation is carried out to block of pixels where second central point using bilinear interpolation, obtains second central point
Second mean value of place block of pixels, and second variance is obtained by the mean value of second mean value and each neighborhood block；It is based on
Second mean value and the second variance obtain second feature point threshold value.
10. camera according to claim 8, which is characterized in that
Described image handles chip, specifically for constructing the Storage Structure of Tree of the original image, by described original image etc.
It is divided into multiple block of pixels, the primitive character point in each block of pixels is correspondingly placed into the node of the Storage Structure of Tree
In；According to the number loop fission of the primitive character point in the node, newly-increased node is obtained, the newly-increased node is added to
First node list is obtained in the Storage Structure of Tree；Obtain primitive character described in each node of the first node list
Point response, sort according to the response size to the primitive character point, filter out make number one it is described original
Characteristic point.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910161155.6A CN110046623B (en) | 2019-03-04 | 2019-03-04 | Image feature point extraction method and camera |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910161155.6A CN110046623B (en) | 2019-03-04 | 2019-03-04 | Image feature point extraction method and camera |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN110046623A true CN110046623A (en) | 2019-07-23 |
| CN110046623B CN110046623B (en) | 2021-09-10 |
Family
ID=67274490
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201910161155.6A Active CN110046623B (en) | 2019-03-04 | 2019-03-04 | Image feature point extraction method and camera |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN110046623B (en) |
Cited By (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110619338A (en) * | 2019-09-18 | 2019-12-27 | 成都信息工程大学 | Image feature extraction method capable of long-time dependence |
| CN111915527A (en) * | 2020-08-05 | 2020-11-10 | 成都圭目机器人有限公司 | Self-adaptive illumination compensation method based on multiple templates |
| CN113283442A (en) * | 2020-02-19 | 2021-08-20 | 北京四维图新科技股份有限公司 | Feature point extraction method and device |
Citations (11)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN102254144A (en) * | 2011-07-12 | 2011-11-23 | 四川大学 | Robust method for extracting two-dimensional code area in image |
| CN103077527A (en) * | 2013-02-05 | 2013-05-01 | 湖北工业大学 | Robust multi-source satellite remote sensing image registration method |
| CN103886338A (en) * | 2014-02-13 | 2014-06-25 | 中国科学院光电研究院 | Multispectral image matching method and system thereof |
| US20140226913A1 (en) * | 2009-01-14 | 2014-08-14 | A9.Com, Inc. | Method and system for matching an image using image patches |
| CN105844663A (en) * | 2016-03-21 | 2016-08-10 | 中国地质大学（武汉） | Adaptive ORB object tracking method |
| CN106960451A (en) * | 2017-03-13 | 2017-07-18 | 西安电子科技大学 | A kind of method for lifting the weak texture region characteristic point quantity of image |
| CN107240126A (en) * | 2016-03-28 | 2017-10-10 | 华天科技（昆山）电子有限公司 | The calibration method of array image |
| CN108010045A (en) * | 2017-12-08 | 2018-05-08 | 福州大学 | Visual pattern characteristic point error hiding method of purification based on ORB |
| CN108573269A (en) * | 2017-10-24 | 2018-09-25 | 北京金山云网络技术有限公司 | Image characteristic point matching method, coalignment, electronic equipment and storage medium |
| CN108830831A (en) * | 2018-05-11 | 2018-11-16 | 中南大学 | One kind is based on the improvement matched zinc flotation froth nature velocity characteristic extracting method of SURF |
| CN109284781A (en) * | 2018-09-10 | 2019-01-29 | 湖北工业大学 | Image classification algorithms and system based on manifold learning |
-
2019
- 2019-03-04 CN CN201910161155.6A patent/CN110046623B/en active Active
Patent Citations (11)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20140226913A1 (en) * | 2009-01-14 | 2014-08-14 | A9.Com, Inc. | Method and system for matching an image using image patches |
| CN102254144A (en) * | 2011-07-12 | 2011-11-23 | 四川大学 | Robust method for extracting two-dimensional code area in image |
| CN103077527A (en) * | 2013-02-05 | 2013-05-01 | 湖北工业大学 | Robust multi-source satellite remote sensing image registration method |
| CN103886338A (en) * | 2014-02-13 | 2014-06-25 | 中国科学院光电研究院 | Multispectral image matching method and system thereof |
| CN105844663A (en) * | 2016-03-21 | 2016-08-10 | 中国地质大学（武汉） | Adaptive ORB object tracking method |
| CN107240126A (en) * | 2016-03-28 | 2017-10-10 | 华天科技（昆山）电子有限公司 | The calibration method of array image |
| CN106960451A (en) * | 2017-03-13 | 2017-07-18 | 西安电子科技大学 | A kind of method for lifting the weak texture region characteristic point quantity of image |
| CN108573269A (en) * | 2017-10-24 | 2018-09-25 | 北京金山云网络技术有限公司 | Image characteristic point matching method, coalignment, electronic equipment and storage medium |
| CN108010045A (en) * | 2017-12-08 | 2018-05-08 | 福州大学 | Visual pattern characteristic point error hiding method of purification based on ORB |
| CN108830831A (en) * | 2018-05-11 | 2018-11-16 | 中南大学 | One kind is based on the improvement matched zinc flotation froth nature velocity characteristic extracting method of SURF |
| CN109284781A (en) * | 2018-09-10 | 2019-01-29 | 湖北工业大学 | Image classification algorithms and system based on manifold learning |
Non-Patent Citations (3)
| Title |
|---|
| PRATHAP NAIR 等: "REGION SEGMENTATION AND FEATURE POINT EXTRACTION ON 3D FACES USING A POINT DISTRIBUTION MODEL", 《2007IEEE INTERNATIONAL CONFERENCE ON IMAGE PROCESSING》 * |
| 屈玉福 等: "自适应变尺度特征点提取方法", 《光学精密工程》 * |
| 杨晓晖 等: "图像特征点提取角点量阈值自适应选取方法", 《光学与光电技术》 * |
Cited By (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110619338A (en) * | 2019-09-18 | 2019-12-27 | 成都信息工程大学 | Image feature extraction method capable of long-time dependence |
| CN110619338B (en) * | 2019-09-18 | 2022-02-08 | 成都信息工程大学 | Image feature extraction method capable of long-time dependence |
| CN113283442A (en) * | 2020-02-19 | 2021-08-20 | 北京四维图新科技股份有限公司 | Feature point extraction method and device |
| CN111915527A (en) * | 2020-08-05 | 2020-11-10 | 成都圭目机器人有限公司 | Self-adaptive illumination compensation method based on multiple templates |
| CN111915527B (en) * | 2020-08-05 | 2023-04-07 | 成都圭目机器人有限公司 | Self-adaptive illumination compensation method based on multiple templates |
Also Published As
| Publication number | Publication date |
|---|---|
| CN110046623B (en) | 2021-09-10 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN107301402B (en) | Method, device, medium and equipment for determining key frame of real scene | |
| CN109522967A (en) | A kind of commodity attribute recognition methods, device, equipment and storage medium | |
| CN110046623A (en) | A kind of characteristics of image point extracting method and camera | |
| CN110990516B (en) | Map data processing method, device and server | |
| CN111898617B (en) | Target detection method and system based on attention mechanism and parallel void convolution network | |
| CN110334762B (en) | Feature matching method based on quad tree combined with ORB and SIFT | |
| US20230169677A1 (en) | Pose Estimation Method and Apparatus | |
| CN111027372A (en) | Pedestrian target detection and identification method based on monocular vision and deep learning | |
| CN108985443A (en) | Action identification method and its neural network generation method, device and electronic equipment | |
| CN111161387A (en) | Method and system for synthesizing image in stacked scene, storage medium and terminal equipment | |
| WO2022048468A1 (en) | Planar contour recognition method and apparatus, computer device, and storage medium | |
| CN115329111B (en) | Image feature library construction method and system based on point cloud and image matching | |
| CN111583378B (en) | Virtual asset processing method and device, electronic equipment and storage medium | |
| CN111583381A (en) | Rendering method and device of game resource map and electronic equipment | |
| CN110930503A (en) | Method and system for establishing three-dimensional model of clothing, storage medium and electronic equipment | |
| CN109344846B (en) | Image feature extraction method and device | |
| CN106033613B (en) | Method for tracking target and device | |
| CN112734747A (en) | Target detection method and device, electronic equipment and storage medium | |
| CN113570615A (en) | Image processing method based on deep learning, electronic equipment and storage medium | |
| CN109815975A (en) | A kind of objective classification method and relevant apparatus based on robot | |
| CN116228850A (en) | Object posture estimation method, device, electronic equipment and readable storage medium | |
| CN110689609B (en) | Image processing method, image processing device, electronic equipment and storage medium | |
| CN114372931A (en) | Target object blurring method and device, storage medium and electronic equipment | |
| CN114022721A (en) | Image feature point selection method, related device, equipment and storage medium | |
| Uskenbayeva et al. | Contour analysis of external images |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |