CN104715476A - Salient object detection method based on histogram power function fitting - Google Patents
Salient object detection method based on histogram power function fitting Download PDFInfo
- Publication number
- CN104715476A CN104715476A CN201510078176.3A CN201510078176A CN104715476A CN 104715476 A CN104715476 A CN 104715476A CN 201510078176 A CN201510078176 A CN 201510078176A CN 104715476 A CN104715476 A CN 104715476A
- Authority
- CN
- China
- Prior art keywords
- pixel
- super
- remarkable
- power function
- histogram
- 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
Landscapes
- Image Analysis (AREA)
- Facsimile Image Signal Circuits (AREA)
Abstract
A salient object detection method based on histogram power function fitting comprises the four steps of the histogram power function fitting, a superpixel classification, a salient region problem and salient object detection. The salient object detection method based on the histogram power function fitting has the advantages that gray threshold is obtained through using an FT saliency map, a figure manifold sorting method, an SLIC method superpixel classification and the histogram power function fitting, under the situations of multiple images, multiple targets and complex scenes, the detection efficiency is high, the performance is good, the precision is good, the problem in salient object detection fields is solved, by means of the salient object detection method based on the histogram power function fitting, the execution speed is high, and the complexity of algorithm is low; meanwhile, higher detection precision can be guaranteed.
Description
Technical field
The present invention relates to image well-marked target detection field, specifically a kind of well-marked target detection method based on histogram power function fitting.
Background technology
As everyone knows, the develop rapidly of computing power and function is that machine intelligence provides reliable feasible condition, and along with going deep into of the subject such as machine learning, pattern-recognition, people more and more wish that computing machine can more autonomous more intelligent finishing the work.Realize this target, need computing machine can understand the environment of surrounding.The main mode of human perception external information is by vision, so the key of computer understanding surrounding environment has visually-perceptible processing power.
Well-marked target is the target that in image, people pay close attention to the most, generally comprises the information that more people are interested, more useful.Therefore, well-marked target detects and is widely used in the fields such as target identification, Iamge Segmentation, image retrieval.Conventional well-marked target detection technique mainly contains the marking area detection technique based on local contrast, as: based on local contrast and fuzzy growth technology, multiple dimensioned center-surrounding histogram and Color-spatial distribution correlation technique etc.; And based on the marking area detection technique of global contrast, as: Achanta is from frequency field angle, method (the Frequency-tuned salient region detection that a kind of marking area based on global contrast detects is proposed, be called for short FT method), the method is using the saliency value of the Euclidean distance between each pixel value in Gassian low-pass filter image and the average pixel value of entire image as this point.But can lose efficacy in the following two cases:
(1) color of marking area accounts for the major part in image, and after being calculated by the method, background can have higher saliency value;
(2) contain a small amount of outstanding color in background, the saliency value of this part color in such background also can be very high.
Pertinent literature: ACHANTA R, HEMAMI S, ESTRADA F, et al. Frequency-tuned salient region detection [C] // IEEE Conference on Computer Vision and Pattern Recognition, 2009:1597 – 1604.
In addition, although the performance of current many well-marked target detection models under single well-marked target and simple background scene is close to the standard of test set, under multiple goal and complex background, under the background of especially melting mutually in target, good performance can not be obtained.
Figure manifold ranking (Graph Based Manifold Ranking) is a kind of clustering method occurred in the recent period, Laplacian regularization or non-regularization matrix is obtained, under different variants can be applicable to different environment with degree matrix by the adjacency matrix of calculating chart.Figure manifold ranking is applied to well-marked target and detects by the people such as Chuan Yang, image is carried out SLIC segmentation, and the super-pixel after segmentation is as figure node, and using image border node as the seed of dependence query with detection background, then difference of negating obtains marking area.The Detection results of this method under single goal and simple background is relatively good, but, when well-marked target be positioned at image border, multiple goal scene, background is complicated or prospect is melted mutually with background when, Detection results is not ideal enough.
Pertinent literature: Chuan Yang, Lihe Zhang, Huchuan Lu, Minghsuan Yang, Saliency Detection via Graph-Based Manifold Ranking, CVPR2013, P3166-3173.
Summary of the invention
Technical matters to be solved by this invention is to provide a kind of well-marked target detection method based on histogram power function fitting, a gray threshold is found by the histogram data calculating the remarkable figure that FT algorithm obtains, this threshold value can extract the super-pixel belonging to well-marked target region, and using the inquiry seed of these super-pixel as figure manifold ranking, the super-pixel that likely there is remarkable pixel is extracted again by self-adaption binaryzation method, as supplementing of inquiry seed, realize the remarkable figure close to test set standard, thus realize being positioned at image border, multiple goal scene, the detection of the well-marked target of background complexity or the more difficult detection such as prospect and background Xiang Rong.
The present invention for solving the problems of the technologies described above adopted technical scheme is: a kind of well-marked target detection method based on histogram power function fitting, is characterized in that: described object detection method comprises the following steps:
Step one: histogram power function fitting: original image FT algorithm is generated FT and significantly schemes and calculate the intensity histogram diagram data of this remarkable figure, obtain the gray threshold x-for super-pixel classification in the remarkable figure of FT according to intensity histogram diagram data least square fitting power function curve Solving Equations
0;
Step 2: super-pixel is classified: original image SLIC algorithm is divided into n super-pixel, according to the gray threshold that step one obtains, super-pixel is divided into remarkable super-pixel and background super-pixel;
Step 3: marking area is located: find out the super-pixel that there is remarkable pixel;
Step 4: well-marked target detects: the relevance ranking value calculating super-pixel correlation matrix and each super-pixel by figure manifold ranking method, and by the relevance ranking value normalization of each super-pixel being obtained the significance of each super-pixel, the significance assignment of each super-pixel is generated final remarkable figure to all pixels that it comprises.
In step one of the present invention by the method for least square fitting power function curve Solving Equations gray threshold be: according to the intensity histogram diagram data least square fitting power function curve equation of the remarkable figure of FT; By the power function curve equation differentiate obtained, be the point (x-of-1 by derivative
0, y-
0) flex point of gray scale and remarkable gray scale as a setting; X-
0as the gray threshold of separating background and well-marked target in the remarkable figure of FT;
The method of remarkable super-pixel and background super-pixel super-pixel is divided into be in step 2 of the present invention:
One, all pixels average gray mean_gray (i) belonging to same super-pixel i in the remarkable figure of FT are calculated;
Two, all super-pixel are generated by number one instruction vectorial Y1=[y-1, y--2 ... yn] T; what average gray mean_gray (i) in all super-pixel is greater than gray threshold is classified as remarkable super-pixel; its value yi (i=1,2 ... n) 1 is set to; otherwise be classified as background super-pixel, its value yi (i=1,2;, n) be set to 0;
In step 3 of the present invention, the method for marking area location is:
One, by remarkable for FT figure binaryzation: adopt self-adaption binaryzation method, gray scale is set to 255 higher than the pixel grey scale of the remarkable figure average gray of FT 2 times of values, and gray scale is set to 0 lower than the pixel grey scale of this gray scale;
Two, all super-pixel are generated by number a vectorial Y of instruction
2=[y-
1, y
--2... y
n]
t, the super-pixel numbering of gray scale belonging to the pixel of 255 in the remarkable binary map of statistics FT, the indicated value yi of the super-pixel i that there is remarkable pixel is set to 1, and all the other are set to 0.
In step 4 of the present invention, the computing method of super-pixel correlation matrix are: the super-pixel composition diagram G=(V after being split by original image, E), wherein V represents all super-pixel set of figure G, E represents the full fillet set of all nodes, calculates super-pixel correlation matrix C=(D-α W) by figure manifold ranking method
-1, wherein, D is the degree matrix of figure G, and W is the adjacency matrix of super-pixel, and α is related coefficient.
In step 4 of the present invention, the computing method of the relevance ranking value of each super-pixel are:
One, all super-pixel are generated by number a vectorial Y=[y-of instruction
1, y
--
2... y
n]
t, make Y=Y
1.| Y
2, namely Y gets Y
1and Y
2the value of step-by-step or computing;
Two, according to formula f*=(D-α W)
-1y tries to achieve the relevance ranking value of each super-pixel.
Step 2 original image of the present invention is 180-230 by the super-pixel number n that SLIC algorithm is split.
The invention has the beneficial effects as follows: (1) uses the remarkable figure of FT to obtain remarkable super-pixel by the method for histogram data power function fitting calculating target, background segment threshold value, with all regions that FT remarkable binary map location well-marked target may exist, improve the precision that well-marked target detects; (2) well-marked target detection is carried out by figure manifold ranking method, can realize fast detecting and the binary map segmentation of marking area at the well-marked target of the simple scenario of single goal, simple background and multiple goal, complex scene, further ensure the precision of target detection, compensate for the failure conditions that FT algorithm may exist; (3) detection of FT remarkable figure composition graphs manifold ranking is faster than the execution speed of most of well-marked target detection method, algorithm complex is low, can ensure higher accuracy of detection simultaneously; (4) super-pixel classification involved in the present invention adopts the good pixel clustering technique of forward position performance---SLIC method, and the super-pixel after cluster can effectively preserve well-marked target edge, ensure that the last remarkable figure generated can more clearly display-object profile; (5) marking area location involved in the present invention have employed FT significantly the binary map that obtains of figure self-adaption binaryzation method in conjunction with the method for super-pixel information, can ensure that remarkable pixel region is located out to greatest extent, effectively improve the precision of detection; (6) well-marked target involved in the present invention detects the figure manifold ranking method that have employed similar Spectral Clustering, non-regular laplacian matrix can search the node relevant to oneself effectively from inquiry seed in figure node, and sort according to correlativity size, such that remarkable figure formation speed is fast, precision is high; (7) histogram power function fitting involved in the present invention asks grey relevant dynamic matrix to be a kind of theory of errors based on least square method, belongs to numerical analysis category, is not simple digital image processing techniques.(8) the present invention is in image multiple goal and under scene complicated situation, detection efficiency is high, and performance is good, and precision is high, solves a great problem of well-marked target detection field.
Accompanying drawing explanation
Fig. 1 is general flow chart of the present invention;
Fig. 2 is that power function fitting of the present invention asks gray threshold process flow diagram;
Fig. 3 is super-pixel classification process figure of the present invention;
Fig. 4 is marking area positioning flow figure of the present invention;
Fig. 5 is the well-marked target overhaul flow chart that the present invention is based on figure manifold ranking;
Fig. 6 is the present invention's remarkable figure binaryzation process flow diagram;
Fig. 7 is basic procedure instance graph of the present invention.
Embodiment
A kind of well-marked target detection method based on histogram power function fitting of the present invention, comprising: the steps such as the classification of histogram power function fitting, super-pixel, marking area location, well-marked target detection.
The present invention is based on the specific implementation of the well-marked target detection method of histogram power function fitting for illustrating, existing in conjunction with the embodiments and accompanying drawing be described below:
Fig. 1 is general flow chart of the present invention, and it realizes a complete procedure based on the well-marked target detection method of histogram power function fitting by following steps, comprising:
Step one: the FT using FT algorithm to generate original image significantly schemes, and the remarkable diagram data FTimname of FT is according to formula
its histogram data hist (i) is calculated, wherein i=0,1,2 after normalization ..., 255, hist (i) represents that gray scale is the number of pixels of i;
Step 2: for eliminating low gray scale high frequency time to the adverse effect asking flex point to bring, before matching by remarkable for FT figure histogram pixel logarithmic data hist (i) all divided by 100 (0≤i≤255), then least square method is passed through, with data set (i, hist (i)) | 0≤i≤255} matching power function curve y=Ax
b, obtain coefficient A and B;
Step 3: to power function y=Ax
bdifferentiate, obtains dy/dx=A*Bx
b-1, make dy/dx=-1 solve x
0, by x
0as the gray threshold of super-pixel classification;
Step 4: use SLIC algorithm original image to be divided into n (value of n is 180-230) super-pixel, obtain the attaching information matrix superpixels of original image element ownership super-pixel, superpixels (i, j) denotation coordination is the numbering of the super-pixel belonging to pixel of (i, j);
Step 5: define the one-dimensional vector Y that a length is super-pixel number n
1=[y
1, y
2.., y
n]
t, calculate all pixel grey scale mean value mean_gray belonging to same super-pixel in the remarkable figure of FT, and and x
0relatively, mean value is greater than x
0super-pixel i be regarded as remarkable super-pixel, y
ivalue be set to 1, otherwise be set to 0;
Step 6: by remarkable for FT figure binaryzation, the pixel grey scale that gray scale is greater than the remarkable Fig. 2 of FT times average gray is set to 255, and all the other are set to 0.Define the one-dimensional vector Y that a length is super-pixel number n
2=[y
1, y
2.., y
n]
t, by the indicated value y of the super-pixel i belonging to the pixel of 255 of gray scale in all binary map
ibe set to 1, other are set to 0;
Step 7: super-pixel is considered as node composition diagram G=(V, E), wherein V represents the node set of figure G and all super-pixel set, and E represents the full fillet set of all nodes.First ask the adjacency matrix W of super-pixel, each element w, ci and cj of W represent the color average of super-pixel i and j at CIELAB color space, and σ is weights coefficient, the constant between desirable 0 to 1, and the present invention gets 0.1, according to formula
calculate the adjacent weights of each node, obtain the adjacency matrix of figure G, according to formula
try to achieve the degree matrix D=diag{d of figure G
11.., d
nn, if α is related coefficient, the constant between desirable 0 to 1, the present invention gets α=0.99, according to formula (D-α W)
-1try to achieve the correlation matrix of figure G;
Step 8: the remarkable instruction vector sum location instruction vector obtained in step 5 and step 6 is step-by-step or computing merging Y=Y
1.|Y
2, obtain indicating vectorial Y as inquiry, according to formula
try to achieve the significance of each super-pixel, according to formula
normalization significance, by all pixels that the significance assignment after normalization comprises to super-pixel, generates final remarkable figure;
Step 9: according to formula
remarkable figure binaryzation is obtained marking area binarization segmentation figure.
Described in step 2, the mode of least square method power function fitting is as follows:
Known histogram data point (i, hist (i)), wherein i=1,2 ..., m distribution is roughly a power function curve.Make the power function curve that matching two is joined
, for solving conveniently, will
take the logarithm in both sides, order
;
This curve is not by all data points, but makes sum of square of deviations
for minimum, the deviation wherein often organizing data and matched curve is
;
According to the principle of least square, A and B should be got and make
have minimal value, therefore A and B should meet following condition:
Obtain following normal equation group
Solve this system of equations, solve A and B, substitute into
obtain matching power function curve.
Claims (5)
1. based on a well-marked target detection method for histogram power function fitting, it is characterized in that: described object detection method comprises the following steps:
Step one: histogram power function fitting: original image FT algorithm is generated FT and significantly schemes and calculate the intensity histogram diagram data of this remarkable figure, obtain the gray threshold x-for super-pixel classification in the remarkable figure of FT according to intensity histogram diagram data least square fitting power function curve Solving Equations
0;
Step 2: super-pixel is classified: original image SLIC algorithm is divided into n super-pixel, according to the gray threshold that step one obtains, super-pixel is divided into remarkable super-pixel and background super-pixel;
Step 3: marking area is located: find out the super-pixel that there is remarkable pixel;
Step 4: well-marked target detects: the relevance ranking value calculating super-pixel correlation matrix and each super-pixel by figure manifold ranking method, and by the relevance ranking value normalization of each super-pixel being obtained the significance of each super-pixel, the significance assignment of each super-pixel is generated final remarkable figure to all pixels that it comprises.
2. a kind of well-marked target detection method based on histogram power function fitting according to claim 1, it is characterized in that: in described step one by the method for least square fitting power function curve Solving Equations gray threshold be: according to intensity histogram diagram data least square fitting power function curve equation, by the power function curve equation differentiate obtained, be the point (x-of-1 by derivative
0, y-
0) flex point of gray scale and remarkable gray scale as a setting, x-
0as the gray threshold of separating background and well-marked target in the remarkable figure of FT;
A kind of well-marked target detection method based on histogram power function fitting according to claim 1, is characterized in that: super-pixel be divided into the method for remarkable super-pixel and background super-pixel to be in described step 2:
One, all pixels average gray mean_gray (i) belonging to same super-pixel i in the remarkable figure of FT are calculated;
Two, all super-pixel are generated by number a vectorial Y of instruction
1=[y-
1, y
--
2... y
n]
t, what average gray mean_gray (i) in all super-pixel is greater than gray threshold is classified as remarkable super-pixel, its value y
i(i=1,2 ..., n) be set to 1, otherwise be classified as background super-pixel, its value y
i(i=1,2 ..., n) be set to 0;
A kind of well-marked target detection method based on histogram power function fitting according to claim 1, is characterized in that: in described step 3, the method for marking area location is:
One, by remarkable for FT figure binaryzation: adopt self-adaption binaryzation method that gray scale is set to 255 higher than the pixel grey scale of the remarkable figure average gray of FT 2 times of values, otherwise be set to 0;
Two, all super-pixel are generated by number a vectorial Y of instruction
2=[y-
1, y
--2... y
n]
t, in the remarkable binary map of statistics FT, the super-pixel numbering of gray scale belonging to the pixel of 255, is set to 1 the indicated value yi of the super-pixel i that there is remarkable pixel, otherwise is set to 0.
3. a kind of well-marked target detection method based on histogram power function fitting according to claim 1, it is characterized in that: in described step 4, the computing method of super-pixel correlation matrix are: the super-pixel composition diagram G=(V after original image is split, E), wherein V represents all super-pixel set of figure G, E represents the full fillet set of all nodes, calculates super-pixel correlation matrix C=(D-α W) by figure manifold ranking method
-1, wherein, D is the degree matrix of figure G, and W is the adjacency matrix of super-pixel, and α is related coefficient.
4. a kind of well-marked target detection method based on histogram power function fitting according to claim 1, is characterized in that: in described step 4, the computing method of the relevance ranking value of each super-pixel are:
One, all super-pixel are generated by number a vectorial Y=[y-of instruction
1, y
--
2... y
n]
t, make Y=Y
1.| Y
2, namely Y gets Y
1and Y
2the value of step-by-step or computing;
Two, according to formula f*=(D-α W)
-1y tries to achieve the relevance ranking value of each super-pixel.
5. a kind of well-marked target detection method based on histogram power function fitting according to claim 1, is characterized in that: described step 2 original image is 180-230 by the super-pixel number n that SLIC algorithm is split.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201510078176.3A CN104715476B (en) | 2015-02-13 | 2015-02-13 | A kind of well-marked target detection method based on histogram power function fitting |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201510078176.3A CN104715476B (en) | 2015-02-13 | 2015-02-13 | A kind of well-marked target detection method based on histogram power function fitting |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN104715476A true CN104715476A (en) | 2015-06-17 |
| CN104715476B CN104715476B (en) | 2017-06-16 |
Family
ID=53414770
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201510078176.3A Expired - Fee Related CN104715476B (en) | 2015-02-13 | 2015-02-13 | A kind of well-marked target detection method based on histogram power function fitting |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN104715476B (en) |
Cited By (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN105761238A (en) * | 2015-12-30 | 2016-07-13 | 河南科技大学 | Method of extracting saliency target through gray statistical data depth information |
| CN105913064A (en) * | 2016-04-12 | 2016-08-31 | 福州大学 | Image visual saliency detection fitting optimization method |
| CN106056590A (en) * | 2016-05-26 | 2016-10-26 | 重庆大学 | Manifold Ranking-based foreground- and background-characteristic combined saliency detection method |
| CN106296695A (en) * | 2016-08-12 | 2017-01-04 | 西安理工大学 | Adaptive threshold natural target image based on significance segmentation extraction algorithm |
| CN106447699A (en) * | 2016-10-14 | 2017-02-22 | 中国科学院自动化研究所 | High-speed rail overhead contact line equipment object detection and tracking method based on Kalman filtering |
Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20110229025A1 (en) * | 2010-02-10 | 2011-09-22 | Qi Zhao | Methods and systems for generating saliency models through linear and/or nonlinear integration |
| CN102831402A (en) * | 2012-08-09 | 2012-12-19 | 西北工业大学 | Sparse coding and visual saliency-based method for detecting airport through infrared remote sensing image |
| CN104240244A (en) * | 2014-09-10 | 2014-12-24 | 上海交通大学 | Significant object detection method based on propagation modes and manifold ranking |
-
2015
- 2015-02-13 CN CN201510078176.3A patent/CN104715476B/en not_active Expired - Fee Related
Patent Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20110229025A1 (en) * | 2010-02-10 | 2011-09-22 | Qi Zhao | Methods and systems for generating saliency models through linear and/or nonlinear integration |
| CN102831402A (en) * | 2012-08-09 | 2012-12-19 | 西北工业大学 | Sparse coding and visual saliency-based method for detecting airport through infrared remote sensing image |
| CN104240244A (en) * | 2014-09-10 | 2014-12-24 | 上海交通大学 | Significant object detection method based on propagation modes and manifold ranking |
Non-Patent Citations (1)
| Title |
|---|
| 杨川等: ""基于先验融合和流形排序的显著目标检测"", 《中国优秀硕士学位论文全文数据库信息科技辑》 * |
Cited By (10)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN105761238A (en) * | 2015-12-30 | 2016-07-13 | 河南科技大学 | Method of extracting saliency target through gray statistical data depth information |
| CN105761238B (en) * | 2015-12-30 | 2018-11-06 | 河南科技大学 | A method of passing through gray-scale statistical data depth information extraction well-marked target |
| CN105913064A (en) * | 2016-04-12 | 2016-08-31 | 福州大学 | Image visual saliency detection fitting optimization method |
| CN105913064B (en) * | 2016-04-12 | 2017-03-08 | 福州大学 | A kind of image vision conspicuousness detection fitting optimization method |
| CN106056590A (en) * | 2016-05-26 | 2016-10-26 | 重庆大学 | Manifold Ranking-based foreground- and background-characteristic combined saliency detection method |
| CN106056590B (en) * | 2016-05-26 | 2019-02-22 | 重庆大学 | Conspicuousness detection method based on Manifold Ranking and combination prospect background feature |
| CN106296695A (en) * | 2016-08-12 | 2017-01-04 | 西安理工大学 | Adaptive threshold natural target image based on significance segmentation extraction algorithm |
| CN106296695B (en) * | 2016-08-12 | 2019-05-24 | 西安理工大学 | Adaptive threshold natural target image segmentation extraction algorithm based on conspicuousness |
| CN106447699A (en) * | 2016-10-14 | 2017-02-22 | 中国科学院自动化研究所 | High-speed rail overhead contact line equipment object detection and tracking method based on Kalman filtering |
| CN106447699B (en) * | 2016-10-14 | 2019-07-19 | 中国科学院自动化研究所 | High iron catenary object detecting and tracking method based on Kalman filtering |
Also Published As
| Publication number | Publication date |
|---|---|
| CN104715476B (en) | 2017-06-16 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN104715251B (en) | A kind of well-marked target detection method based on histogram linear fit | |
| CN107633226B (en) | Human body motion tracking feature processing method | |
| CN109636784B (en) | Image saliency target detection method based on maximum neighborhood and super-pixel segmentation | |
| CN109086724B (en) | Accelerated human face detection method and storage medium | |
| CN108629783B (en) | Image segmentation method, system and medium based on image feature density peak search | |
| CN110610150B (en) | Tracking method, device, computing equipment and medium of target moving object | |
| CN105740945A (en) | People counting method based on video analysis | |
| CN104715476A (en) | Salient object detection method based on histogram power function fitting | |
| CN110992378B (en) | Dynamic updating vision tracking aerial photographing method and system based on rotor flying robot | |
| CN104077605A (en) | Pedestrian search and recognition method based on color topological structure | |
| CN104463870A (en) | Image salient region detection method | |
| CN113763427B (en) | Multi-target tracking method based on coarse-to-fine shielding processing | |
| CN111274964B (en) | Detection method for analyzing water surface pollutants based on visual saliency of unmanned aerial vehicle | |
| CN114529583B (en) | Power equipment tracking method and tracking system based on residual regression network | |
| CN107194917B (en) | DAP and ARE L M-based on-orbit SAR image change detection method | |
| CN109448024B (en) | Visual tracking method and system for constructing constraint correlation filter by using depth data | |
| Zhao | Motion track enhancement method of sports video image based on otsu algorithm | |
| Dornaika et al. | A comparative study of image segmentation algorithms and descriptors for building detection | |
| CN108241837B (en) | Method and device for detecting remnants | |
| Han et al. | Accurate and robust vanishing point detection method in unstructured road scenes | |
| CN104899875A (en) | Rapid image cooperation salient region monitoring method based on integration matching | |
| Liu | Research on intelligent visual image feature region acquisition algorithm in Internet of Things framework | |
| CN110532892B (en) | Method for detecting road vanishing point of single image of unstructured road | |
| CN114511803A (en) | Target occlusion detection method for visual tracking task | |
| CN108389219B (en) | Weak and small target tracking loss re-detection method based on multi-peak judgment |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| C06 | Publication | ||
| PB01 | Publication | ||
| C10 | Entry into substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| TR01 | Transfer of patent right | ||
| TR01 | Transfer of patent right |
Effective date of registration: 20191106 Address after: Room E601, innovation building, No. 68, Wenhua Road, Jinshui District, Zhengzhou City, Henan Province Patentee after: Zhengzhou Xinma Technology Co., Ltd Address before: 471000 Xiyuan Road, Jianxi District, Henan, No. 48, No. Patentee before: Henan University of Science and Technology |
|
| CF01 | Termination of patent right due to non-payment of annual fee | ||
| CF01 | Termination of patent right due to non-payment of annual fee |
Granted publication date: 20170616 Termination date: 20200213 |