CN109064491A - A kind of nuclear phase pass filter tracking method of adaptive piecemeal - Google Patents
A kind of nuclear phase pass filter tracking method of adaptive piecemeal Download PDFInfo
- Publication number
- CN109064491A CN109064491A CN201810808535.XA CN201810808535A CN109064491A CN 109064491 A CN109064491 A CN 109064491A CN 201810808535 A CN201810808535 A CN 201810808535A CN 109064491 A CN109064491 A CN 109064491A
- Authority
- CN
- China
- Prior art keywords
- piecemeal
- target
- pixel
- value
- tracking
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T7/00—Image analysis
- G06T7/20—Analysis of motion
- G06T7/246—Analysis of motion using feature-based methods, e.g. the tracking of corners or segments
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/23—Clustering techniques
- G06F18/232—Non-hierarchical techniques
- G06F18/2321—Non-hierarchical techniques using statistics or function optimisation, e.g. modelling of probability density functions
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/24—Classification techniques
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/40—Extraction of image or video features
- G06V10/50—Extraction of image or video features by performing operations within image blocks; by using histograms, e.g. histogram of oriented gradients [HoG]; by summing image-intensity values; Projection analysis
- G06V10/507—Summing image-intensity values; Histogram projection analysis
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/10—Image acquisition modality
- G06T2207/10016—Video; Image sequence
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Data Mining & Analysis (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Evolutionary Biology (AREA)
- Evolutionary Computation (AREA)
- Bioinformatics & Computational Biology (AREA)
- General Engineering & Computer Science (AREA)
- Artificial Intelligence (AREA)
- Life Sciences & Earth Sciences (AREA)
- Multimedia (AREA)
- Probability & Statistics with Applications (AREA)
- Image Analysis (AREA)
Abstract
The present invention relates to a kind of nuclear phases of adaptive piecemeal to close filter tracking method, HOG feature, gray feature, textural characteristics training classifier are extracted as unit of super-pixel block, calculate tracking response figure, and figure calculates average peak correlation energy according to response, when any one block energy value is less than threshold value, it was demonstrated that tracking is lost, and reselects method of partition based on the aspect ratio of previous frame target area, otherwise according to the mutual alignment relation between each piecemeal, goal end position and size are obtained.When target aspect is than changing obvious, accurate tracking result just can not be obtained.Method of the invention can avoid tracking result mistake by adaptive piecemeal in the case where target aspect ratio changes, and HOG feature, gray feature and the textural characteristics based on super-pixel carry out classifier training, and tracking is carried out according to average peak correlation energy and loses judgement, it can be realized quick, accurate and effective target following.
Description
Technical field
The present invention relates to the nuclear phases of digital image processing techniques field more particularly to a kind of adaptive piecemeal to close filter tracking
Method.
Background technique
Target following is the research hotspot of computer vision field, in precise guidance, intelligent transportation, video monitoring, man-machine
The multiple fields extensive applications such as interaction.Although having obtained huge progress between in the past twenty years, but still face background
Change, block, the interference of illumination variation, the factors such as target deformation.
Existing plurality of target track algorithm shows excellent performance.In recent years, correlation filter be introduced into target with
Track, and achieve good tracking result.Bolme etc. proposes minimum output square error and (MOSSE), for the first time filters correlation
The thought of wave is introduced into target following.The it is proposeds such as Henriques are based on core loop structure detecting and tracking (CSK), using intensively adopting
Sample and kernel function, which simplify, calculates step, and this method is high-efficient, and average tracking speed is per second up to hundreds of frames;It is again public in 2015 afterwards
It has opened and the single channel in CSK method is extended to multichannel, using HOG feature replacement original gradation feature, proposed nuclear phase pass
Filter (KCF) tracking.Danelljan etc. has used color characteristic on the basis of CSK, proposes CN tracking.But
It is the above method is all fixed-size tracking.In order to solve the problems, such as dimension self-adaption, Danelljan and Yang are in KCF
On the basis of introduce dimension self-adaption mechanism, propose SAMF method and DSST method respectively, both methods is to solve scale
The representative algorithm of variation issue.DSST is similar to the thought that SAMF solves target scale variation: to present frame target according to default
Size amplification and diminution, find Optimum Matching value as new target scale from different scale.Duan Weiwei and Xu Yulong etc. exists
Method on the basis of core correlation filtering using piecemeal realizes multiscale tracing, but both methods is to image fixed block,
When target aspect is than changing obvious, accurate tracking result just can not be obtained.
Summary of the invention
The purpose of the present invention is to provide a kind of nuclear phase of adaptive piecemeal close filter tracking method, with obtain accurately with
Track result.
To realize the above-mentioned technical purpose, the present invention adopts the following technical scheme:
A kind of nuclear phase pass filter tracking method of adaptive piecemeal, selects target area in initial frame center, calculates aspect ratio,
Several piecemeals are divided an image into according to aspect ratio and calculate the super-pixel block of each piecemeal;It is extracted as unit of super-pixel block
HOG feature, gray feature, textural characteristics training classifier, calculate tracking response figure, and figure calculates average peak phase according to response
Energy is closed, when any one block energy value is less than threshold value, it was demonstrated that tracking is lost, based on the aspect ratio of previous frame target area
Method of partition is reselected, otherwise according to the mutual alignment relation between each piecemeal, obtains goal end position and size.
Further, the method specifically comprises the following steps:
S100 reads video sequence, obtains initial frame, the selected target region in initial frame;
S200 divides an image into several piecemeals according to the aspect ratio of target area, and determines every piece of region of search;
S300 calculates separately the super-pixel piecemeal of each piecemeal;
S400 extracts gray feature, HOG feature and the textural characteristics of each super-pixel piecemeal respectively, design factor matrix,
Training obtains following Nonlinear Classifier;
S500 obtains next frame, and the partitioned searching region of present frame is obtained according to the piecemeal target position of previous frame, calculates
Correlation filtering response diagram;
S600 is based on correlation filtering response diagram and calculates average peak correlation energy, is shown below:
F in formulamax、Fmin、Fw,hRespectively indicate response highest, minimum and w, the response on the position h;
When the average peak correlation energy value of each piecemeal is both greater than or is equal to history mean value, S700 is continued to execute;When
When wherein any one piece of value is less than history mean value, present frame uses the target position of previous frame, goes to S200;
Further, in the S200, target area aspect ratio is calculated, aspect ratio parameter is set as T1=1.4, T2=
0.6, when aspect ratio is greater than or equal to T1, target level is halved, aspect ratio is less than or equal to T2, target vertical is halved,
Aspect ratio is greater than T2Less than T1When, by the target quartering.
Further, in the S300, the calculation method of super-pixel piecemeal is as follows:
The selected number initialization cluster centre of S310, delimit the initial area of generic;
S320 calculates all pixels point gradient value, seed point is moved on to the smallest position of gradient in seed point 3*3 neighborhood
It sets;
S330 takes all pixels in the certain contiguous range of each cluster centre, calculates each pixel to cluster centre
Pixel is classified as the cluster centre of minimum range by distance;
S340 repeats S330 until error convergence;
S350 carries out connectivity processing to image.
Super-pixel is converted by image to calculate, and can greatly promote calculating speed.
Further, in the S400, textural characteristics, one-dimensional reality Gabor filter are obtained using one-dimensional Gabor filter
It indicates are as follows:
In formula, σ is the standard deviation of Gaussian function, x0For the center point coordinate of function, u0For the centre frequency of cosine wave;It will
The centre frequency of Gabor filtering is determined as 3/64,3/32,3/16, obtains three filter templates, sets 1 for template width,
Angle step is set as 15 °, that is, is provided with 24 directions;
Textural characteristics are obtained using above three filter, steps are as follows:
Gray value of image is mapped in [0,1] section by S410;
The datum mark of above three filter template is respectively aligned to pixel s by S420, along direction j, j ∈ 1,2,
3 ... 24 } filter response value is obtained, and normalize respectively;
S430 calculates the weighted average of the spectral value after equidirectional three normalization filter values and normalization as picture
The texture value in 24 directions of vegetarian refreshments;
24 texture values of S440 compared pixels point s, using minimum value as the final angle textural characteristics of pixel s;
S450 successively selects each pixel, repeats S410~S440, obtains complete angular texture signature value.
Every bit in tracking video has a textural characteristics along either direction, but along the line of a certain specific direction
Reason feature can be best described by object.Therefore, the present invention sets a kind of extracting rule, and the texture for choosing a certain specific direction is special
Sign, the textural characteristics so obtained are angular texture signature.
Method of the invention can avoid tracking result wrong by adaptive piecemeal in the case where target aspect ratio changes
Accidentally, and the HOG feature based on super-pixel, gray feature and textural characteristics carry out classifier training, in target and ambient background face
When color is consistent or height is similar, the microscopic feature and gross feature of image can be preferably taken into account, tracking accuracy is improved;And according to
Average peak correlation energy carries out tracking and loses judgement, realizes quick, accurate and effective target following.
Detailed description of the invention
Fig. 1 is the flow chart of the method for the present invention;
Fig. 2 is filter template setting figure of the present invention.
Specific embodiment
The technical scheme of the present invention will be further described with specific embodiment for explanation with reference to the accompanying drawing.
As shown in Figure 1, steps are as follows for tracking of the invention:
S100 reads video sequence, obtains initial frame, the selected target region in initial frame;
S200 divides an image into several piecemeals according to the aspect ratio of target area, and determines every piece of region of search;
Method of partition in the present embodiment is to calculate target area aspect ratio, and aspect ratio parameter is set as T1=1.4, T2=
0.6, when aspect ratio is greater than or equal to T1, target level is halved, aspect ratio is less than or equal to T2, target vertical is halved,
Aspect ratio is greater than T2Less than T1When, by the target quartering.
S300 calculates separately the super-pixel piecemeal of each piecemeal;
The calculation method of super-pixel piecemeal is as follows:
The selected number initialization cluster centre of S310, delimit the initial area of generic;
S320 calculates all pixels point gradient value, seed point is moved on to the smallest position of gradient in seed point 3*3 neighborhood
It sets;
S330 takes all pixels in the certain contiguous range of each cluster centre, calculates each pixel to cluster centre
Pixel is classified as the cluster centre of minimum range by distance;
S340 repeats S330 until error convergence;
S350 carries out connectivity processing to image.
S400 extracts gray feature, HOG feature and the textural characteristics of each super-pixel piecemeal respectively, design factor matrix,
Training obtains following Nonlinear Classifier;
Angular texture signature is obtained using one-dimensional Gabor filter, one-dimensional reality Gabor filter indicates are as follows:
In formula, σ is the standard deviation of Gaussian function, x0For the center point coordinate of function, u0For the centre frequency of cosine wave;It will
The centre frequency of Gabor filtering is determined as 3/64,3/32,3/16, obtains three filter templates, sets 1 for template width,
Angle step is set as 15 °, that is, is provided with 24 directions;It is as shown in Figure 1 filter template width, the schematic diagram that direction is arranged.
Textural characteristics are obtained using above three filter, steps are as follows:
Gray value of image is mapped in [0,1] section by S410;
The datum mark of three filter templates shown in FIG. 1 will be respectively aligned to pixel s by S420, along direction j, j ∈
1,2,3 ... and 24 } filter response value is obtained, and normalize respectively;
S430 calculates the weighted average of the spectral value after equidirectional three normalization filter values and normalization as picture
The texture value in 24 directions of vegetarian refreshments;
24 texture values of S440 compared pixels point s, using minimum value as the final angle textural characteristics of pixel s;
S450 successively selects each pixel, repeats S410~S440, obtains complete angular texture signature value.
S500 obtains next frame, and the partitioned searching region of present frame is obtained according to the piecemeal target position of previous frame, calculates
Correlation filtering response diagram;
S600 is based on correlation filtering response diagram and calculates average peak correlation energy, is shown below:
Fmax、Fmin、Fw,hRespectively indicate response highest, the response on minimum and position (w, h);
When the average peak correlation energy value of each piecemeal is both greater than or is equal to history mean value, S700 is continued to execute;When
When wherein any one piece of value is less than history mean value, present frame uses the target position of previous frame, goes to S200;
Concussion situation of the above formula to reflect response diagram illustrates that target occlusion occurs, loses when APCE reduces suddenly,
Next frame will not use the tracking result of this frame, and select piecemeal side based on the aspect ratio of previous frame target area again
Method carries out piecemeal tracking.
When target aspect ratio changes greatly, and has been unsatisfactory for the piecemeal situation of initial frame, if continuing to use initial frame
Partitioned mode, it will increase target's center position and size error;Or when piecemeal target is blocked or has lost, but not
Judged, all deviation accumulation is made to have arrived next frame, influences final tracking result.The response diagram of KCF is when tracking accurate
One close to ideal two-dimentional response diagram, and encounter block, target is lost, it is serious mismatch when response diagram shake
Acutely.Therefore whether the present invention judge target by calculating average peak correlation energy APCE and be blocked and need again
New piecemeal.
S700 is using the maximum position of response as the new position of piecemeal target;
S800 determines target size and position according to the relative positional relationship between sub-block;
By taking two piecemeal of target level as an example, into next frame, distance between the center of two piecemeals is calculated, when the distance
Greater than between the center of initial two piecemeals apart from when, it is believed that target amplification, at this time calculate calculate two piecemeals centre bit
The midpoint for setting line, obtains the center of target, and target size is determined by piecemeal maximum magnitude obtained.Similarly, when working as
Between the center of two piecemeals of previous frame distance be less than initial two piecemeals center between apart from when, it is believed that shrinking of object,
After obtaining target's center position, target size is determined by the minimum zone of piecemeal.
S900 updates the display model of object module according to the following formulaAnd coefficient matrix
In formulaWithIt is illustrated respectively in the display model and coefficient matrix that present frame obtains;η is learning rate;
S1000 repeats S300~S900, and up to video, processing terminate.
Claims (5)
1. a kind of nuclear phase of adaptive piecemeal closes filter tracking method, which is characterized in that select target area in initial frame center, count
Aspect ratio is calculated, several piecemeals are divided an image into according to aspect ratio and calculate the super-pixel block of each piecemeal;With super-pixel block
HOG feature, gray feature, textural characteristics training classifier are extracted for unit, calculates tracking response figure, and figure calculates according to response
Average peak correlation energy, when any one block energy value is less than threshold value, it was demonstrated that tracking is lost, with the vertical of previous frame target area
Method of partition is reselected based on horizontal ratio, otherwise according to the mutual alignment relation between each piecemeal, obtains goal end position
With size.
2. the method according to claim 1, wherein including the following steps:
S100 reads video sequence, obtains initial frame, the selected target region in initial frame;
S200 divides an image into several piecemeals according to the aspect ratio of target area, and determines every piece of region of search;
S300 calculates separately the super-pixel piecemeal of each piecemeal;
S400 extracts gray feature, HOG feature and the textural characteristics of each super-pixel piecemeal, design factor matrix, training respectively
Obtain following Nonlinear Classifier;
S500 obtains next frame, and the partitioned searching region of present frame is obtained according to the piecemeal target position of previous frame, calculates related
Filter response figure;
S600 is based on correlation filtering response diagram and calculates average peak correlation energy, is shown below:
F in formulamax、Fmin、Fw,hRespectively indicate response highest, minimum and w, the response on the position h;
When the average peak correlation energy value of each piecemeal is both greater than or is equal to history mean value, S700 is continued to execute;When wherein
When any one piece of value is less than history mean value, present frame uses the target position of previous frame, goes to S200;
S700 is using the maximum position of response as the new position of piecemeal target;
S800 determines target size and position according to the relative positional relationship between sub-block;
S900 updates the display model of object module according to the following formulaAnd coefficient matrix
In formulaWithIt is illustrated respectively in the display model and coefficient matrix that present frame obtains;η is learning rate;
S1000 repeats S300~S900, and up to video, processing terminate.
3. method according to claim 1 or 2, which is characterized in that in the S200, calculate target area aspect ratio, indulge
It is horizontal to be set as T than parameter1=1.4, T2=0.6, when aspect ratio is greater than or equal to T1, target level is halved, aspect ratio is less than
Or it is equal to T2, target vertical is halved, aspect ratio is greater than T2Less than T1When, by the target quartering.
4. method according to claim 1 to 3, which is characterized in that in the S300, the calculation method of super-pixel piecemeal is such as
Under:
The selected number initialization cluster centre of S310, delimit the initial area of generic;
S320 calculates all pixels point gradient value, seed point is moved on to the smallest position of gradient in seed point 3*3 neighborhood;
S330 takes all pixels in the certain contiguous range of each cluster centre, calculate each pixel to cluster centre away from
From pixel to be classified as to the cluster centre of minimum range;
S340 repeats S330 until error convergence;
S350 carries out connectivity processing to image.
5. method described in -4 according to claim 1, which is characterized in that in the S400, obtained using one-dimensional Gabor filter
Angular texture signature, one-dimensional reality Gabor filter indicate are as follows:
In formula, σ is the standard deviation of Gaussian function, x0For the center point coordinate of function, u0For the centre frequency of cosine wave;By Gabor
The centre frequency of filtering is determined as 3/64,3/32,3/16, obtains three filter templates, sets 1 for template width, angle
Step-length is set as 15 °, that is, is provided with 24 directions;
Textural characteristics are obtained using above three filter, steps are as follows:
Gray value of image is mapped in [0,1] section by S410;
The datum mark of above three filter template is respectively aligned to pixel s by S420, along direction j, j ∈ 1,2,3,
... 24 } filter response value is obtained, and normalize respectively;
S430 calculates the weighted average of the spectral value after equidirectional three normalization filter values and normalization as pixel
24 directions texture value;
24 texture values of S440 compared pixels point s, using minimum value as the final angle textural characteristics of pixel s;
S450 successively selects each pixel, repeats S410~S440, obtains complete angular texture signature value.
Applications Claiming Priority (2)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN2018103296509 | 2018-04-12 | ||
CN201810329650 | 2018-04-12 |
Publications (1)
Publication Number | Publication Date |
---|---|
CN109064491A true CN109064491A (en) | 2018-12-21 |
Family
ID=64835318
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201810808535.XA Pending CN109064491A (en) | 2018-04-12 | 2018-07-20 | A kind of nuclear phase pass filter tracking method of adaptive piecemeal |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN109064491A (en) |
Cited By (6)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN110349183A (en) * | 2019-05-30 | 2019-10-18 | 西安电子科技大学 | A kind of tracking based on KCF, device, electronic equipment and storage medium |
CN110414439A (en) * | 2019-07-30 | 2019-11-05 | 武汉理工大学 | Anti- based on multi-peak detection blocks pedestrian tracting method |
CN110706252A (en) * | 2019-09-09 | 2020-01-17 | 西安理工大学 | Robot nuclear correlation filtering tracking algorithm under guidance of motion model |
CN110942472A (en) * | 2019-11-28 | 2020-03-31 | 广西师范大学 | Nuclear correlation filtering tracking method based on feature fusion and self-adaptive blocking |
CN112348847A (en) * | 2020-10-26 | 2021-02-09 | 南京邮电大学 | Target scale self-adaptive tracking method |
CN116823737A (en) * | 2023-06-05 | 2023-09-29 | 中铁九局集团电务工程有限公司 | Tunnel wall abnormity detection method and system in low-texture environment |
Citations (4)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US6012458A (en) * | 1998-03-20 | 2000-01-11 | Mo; Larry Y. L. | Method and apparatus for tracking scan plane motion in free-hand three-dimensional ultrasound scanning using adaptive speckle correlation |
CN107123130A (en) * | 2017-03-06 | 2017-09-01 | 华南理工大学 | Kernel correlation filtering target tracking method based on superpixel and hybrid hash |
CN107154024A (en) * | 2017-05-19 | 2017-09-12 | 南京理工大学 | Dimension self-adaption method for tracking target based on depth characteristic core correlation filter |
CN108090919A (en) * | 2018-01-02 | 2018-05-29 | 华南理工大学 | Improved kernel correlation filtering tracking method based on super-pixel optical flow and adaptive learning factor |
-
2018
- 2018-07-20 CN CN201810808535.XA patent/CN109064491A/en active Pending
Patent Citations (4)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US6012458A (en) * | 1998-03-20 | 2000-01-11 | Mo; Larry Y. L. | Method and apparatus for tracking scan plane motion in free-hand three-dimensional ultrasound scanning using adaptive speckle correlation |
CN107123130A (en) * | 2017-03-06 | 2017-09-01 | 华南理工大学 | Kernel correlation filtering target tracking method based on superpixel and hybrid hash |
CN107154024A (en) * | 2017-05-19 | 2017-09-12 | 南京理工大学 | Dimension self-adaption method for tracking target based on depth characteristic core correlation filter |
CN108090919A (en) * | 2018-01-02 | 2018-05-29 | 华南理工大学 | Improved kernel correlation filtering tracking method based on super-pixel optical flow and adaptive learning factor |
Non-Patent Citations (2)
Title |
---|
孙智华: "基于自适应分块和在线判别式分类器的单目标跟踪技术研究", 《中国优秀硕士学位论文全文数据库信息科技编辑》 * |
栾悉道: "《多媒体情报处理技术》", 30 May 2016 * |
Cited By (12)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN110349183A (en) * | 2019-05-30 | 2019-10-18 | 西安电子科技大学 | A kind of tracking based on KCF, device, electronic equipment and storage medium |
CN110349183B (en) * | 2019-05-30 | 2022-12-09 | 西安电子科技大学 | Tracking method and device based on KCF, electronic equipment and storage medium |
CN110414439A (en) * | 2019-07-30 | 2019-11-05 | 武汉理工大学 | Anti- based on multi-peak detection blocks pedestrian tracting method |
CN110414439B (en) * | 2019-07-30 | 2022-03-15 | 武汉理工大学 | Anti-blocking pedestrian tracking method based on multi-peak detection |
CN110706252A (en) * | 2019-09-09 | 2020-01-17 | 西安理工大学 | Robot nuclear correlation filtering tracking algorithm under guidance of motion model |
CN110706252B (en) * | 2019-09-09 | 2020-10-23 | 西安理工大学 | Robot nuclear correlation filtering tracking algorithm under guidance of motion model |
CN110942472A (en) * | 2019-11-28 | 2020-03-31 | 广西师范大学 | Nuclear correlation filtering tracking method based on feature fusion and self-adaptive blocking |
CN110942472B (en) * | 2019-11-28 | 2023-10-13 | 江苏砺行能源科技有限公司 | Nuclear correlation filtering tracking method based on feature fusion and self-adaptive blocking |
CN112348847A (en) * | 2020-10-26 | 2021-02-09 | 南京邮电大学 | Target scale self-adaptive tracking method |
CN112348847B (en) * | 2020-10-26 | 2023-08-15 | 南京邮电大学 | Target scale self-adaptive tracking method |
CN116823737A (en) * | 2023-06-05 | 2023-09-29 | 中铁九局集团电务工程有限公司 | Tunnel wall abnormity detection method and system in low-texture environment |
CN116823737B (en) * | 2023-06-05 | 2024-05-07 | 中铁九局集团电务工程有限公司 | Tunnel wall abnormity detection method and system in low-texture environment |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN109064491A (en) | A kind of nuclear phase pass filter tracking method of adaptive piecemeal | |
CN102324030B (en) | Target tracking method and system based on image block characteristics | |
CN110210360B (en) | Rope skipping counting method based on video image target recognition | |
CN106997597A (en) | It is a kind of based on have supervision conspicuousness detection method for tracking target | |
CN108682017A (en) | Super-pixel method for detecting image edge based on Node2Vec algorithms | |
CN108596951A (en) | A kind of method for tracking target of fusion feature | |
CN107886507B (en) | A kind of salient region detecting method based on image background and spatial position | |
CN103632137B (en) | A kind of human eye iris segmentation method | |
CN107392968A (en) | The image significance detection method of Fusion of Color comparison diagram and Color-spatial distribution figure | |
CN112837344A (en) | Target tracking method for generating twin network based on conditional confrontation | |
CN106384363B (en) | A kind of quick self-adapted weight solid matching method | |
CN107944437B (en) | A kind of Face detection method based on neural network and integral image | |
CN108615229B (en) | Collision detection optimization method based on curvature point clustering and decision tree | |
CN101872112B (en) | Three-dimensional camera shooting automatic collecting system | |
CN105957107A (en) | Pedestrian detecting and tracking method and device | |
CN109886267A (en) | A kind of soft image conspicuousness detection method based on optimal feature selection | |
CN104376334A (en) | Pedestrian comparison method based on multi-scale feature fusion | |
CN104599288A (en) | Skin color template based feature tracking method and device | |
Zhang et al. | A prior-based graph for salient object detection | |
CN106447662A (en) | Combined distance based FCM image segmentation algorithm | |
CN110334581A (en) | A kind of multi-source Remote Sensing Images change detecting method | |
CN106874843A (en) | A kind of method for tracking target and equipment | |
CN114511803B (en) | Target shielding detection method for visual tracking task | |
CN110910417A (en) | Weak and small moving target detection method based on super-pixel adjacent frame feature comparison | |
CN107506400B (en) | A kind of image search method based on cognitive characteristics and manifold ranking |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
PB01 | Publication | ||
PB01 | Publication | ||
SE01 | Entry into force of request for substantive examination | ||
SE01 | Entry into force of request for substantive examination | ||
RJ01 | Rejection of invention patent application after publication |
Application publication date: 20181221 |
|
RJ01 | Rejection of invention patent application after publication |