CN105931268A - Mean shift tracking method based on scale adaption in UUV underwater recovery process - Google Patents
Mean shift tracking method based on scale adaption in UUV underwater recovery process Download PDFInfo
- Publication number
- CN105931268A CN105931268A CN201610242997.0A CN201610242997A CN105931268A CN 105931268 A CN105931268 A CN 105931268A CN 201610242997 A CN201610242997 A CN 201610242997A CN 105931268 A CN105931268 A CN 105931268A
- Authority
- CN
- China
- Prior art keywords
- light source
- heart
- target
- shaped
- target light
- 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
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/10—Image acquisition modality
- G06T2207/10016—Video; Image sequence
-
- 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
- G06T2207/20076—Probabilistic image processing
Abstract
The invention provides a mean shift tracking method based on scale adaption in a UUV underwater recovery process. The invention relates to the mean shift tracking method based on scale adaption. The invention aims to solve the problem that the bandwidth of a kernel function is fixed in a traditional mean shift method and thus a target with a continuously changing scale in an image can not be tracked. The method comprises the steps of (1) determining y0, (2) obtaining y1, (3) establishing a heart-shaped target light source candidate target model at a current frame and obtaining a similarity coefficient, (4) obtaining y2 and rho(p(y2),q), (5) obtaining rho(p(y2),q), wherein when rho(p(y1),q) is larger than rho(p(y2),q), y2=(y1+y2)/2, ending tracking when ||y2-y1|| is smaller than epsilon, otherwise, allowing y1=y2, returning to the step (3) until ||y2-y1|| is smaller than epsilon, and ending the tracking. The mean shift tracking method is applied to the field of UUV underwater recovery.
Description
Technical field
The present invention relates to based on dimension self-adaption average drifting tracking.
Background technology
When UUV performs task in ocean, owing to the restriction of battery capacity makes its recovery particularly important, and the water of UUV
Lower recovery has is jolted height little, disguised, strategic value advantages of higher by stormy waves;Meanwhile, way of recycling of carrying a load on the back under water is by sea
UUV is reclaimed by cabin, the depressed place platform that midocean or submarine carry, and which does not require the operation of boom hoisting and personnel, has
Vast potential for future development.Video camera has been widely used on UUV, compares other sensor, and it has closely positioning precision
High, obtain contain much information, the advantage such as strong adaptability, resolution is high and cost is less.
In UUV reclaims under water, it is contemplated that common objects feature has the biggest decay in water, cabin, depressed place platform uses light
The guiding target that source is reclaimed as UUV, it is possible to be effective against interference and improve the effect that UUV guiding is reclaimed.The present invention uses one
Planting linear target light source array, light source arrangement signature, in cabin, the depressed place platform of motion occurs in the visual field of UUV video camera
Time, target light source is carried out visual tracking, and realizes the recovery under water of UUV further;Mean shift process is a kind of nonparametric
Multilayer networks method, method utilizes pixel characteristic point probability density function gradient optimization method, quick by iteration
Converge on the local maximum of probability density function, it is achieved fast target positions, and its amount of calculation is little, and strong adaptability is suitable for reality
Time follow the tracks of occasion;But, traditional mean shift process is owing to securing the bandwidth of kernel function, and kernel function affects each pixel
The weights of point also reflect the shapes and sizes of target, thus the target that can not preferably image mesoscale be continually changing carry out with
Track.Just because of this, traditional average drifting tracking range of application is restricted.
Summary of the invention
The invention aims to solve existing traditional mean shift process owing to securing the bandwidth of kernel function, therefore
The problem that the target well can not being continually changing image mesoscale is tracked, and propose in UUV removal process under water
Based on dimension self-adaption average drifting tracking.
Specifically follow the steps below:
Heart-shaped target light source center y in step one, establishment initial frame0;
Step 2, according to target light source center y heart-shaped in initial frame0Set up heart-shaped target light source modelAnd obtain the heart-shaped target light source center y of present frame1;
Wherein, δ represents Kronecker delta function, δ [b (xi)-u] as pixel value xiWhen belonging to u eigenvalue
It is 1, is 0 on the contrary;K () represents kernel function, and h represents kernel function bandwidth, and C is normaliztion constant;
Step 3, set up heart-shaped target light source candidate target model at present frame
Obtain Bhattacharyya likeness coefficient
Wherein, ChFor normaliztion constant;nhRepresent sum of all pixels in region of search;
Step 4, candidate target position, and obtain candidate's heart target area position y2And Similarity value
Step 5, set a width of h of kernel function band of previous framep, a width of h of kernel function band of present frame employingopt, use positive and negative
The mode of increment is to hoptIt is modified, i.e.
H=(1+ Δ h* λi)hp (8)
Wherein, λi=-1,0,1, Δ h takes 0.1hp;
And forward step 2 to, obtain 3 Similarity value according to different kernel function bandwidth h
To 3 Similarity value obtainedComparing, Similarity value is to the maximum
Now corresponding y2It is the new position of position, candidate's heart target area;
Relatively ρ (p (y1), q) with ρ (p (y2), q), as ρ (p (y1),q)>ρ(p(y2), time q),Obtain ρ
(p(y2),q);
When | | y2-y1| | < during ε, then iterative process terminates, and tracking terminates;Otherwise, y is made1=y2, return step 3,
Until | | y2-y1| | < ε, tracking terminates.
Invention effect
The present invention uses the guiding target that a kind of linear target light source array reclaims as UUV.At actual marine environment
In, common Vision Tracking is extremely difficult to tracking accuracy requirement;And mean shift algorithm amount of calculation is little, be suitable in real time with
Track occasion, but its target preferably can not being continually changing image mesoscale is tracked.The present invention is first to linear mesh
Heart-shaped target light source in mark array of source positions;Then use the adaptive updates strategy of target scale, devise base
In the average drifting tracking of dimension self-adaption, ensure the accuracy of visual target tracking, energy in UUV removal process under water
Enough targets being well continually changing image mesoscale are tracked.
By the tracking Comparative result of the present invention with tradition average drifting track algorithm, as shown in Figure 8, front 30 frame two kinds
The tracing deviation of method is roughly the same;But the deviation of the inventive method is 3 when 50 frame, traditional average drifting track algorithm
Deviation is 10;During 70 frame, the deviation of the inventive method is 2, and the deviation of traditional average drifting track algorithm is 12;92 frames with
Average drifting track algorithm traditional when of rear occurs following the tracks of the phenomenon lost, and the deviation of the inventive method is stable 2
Left and right;During whole tracking, the present invention can be accurately positioned tracking mesh based on dimension self-adaption average drifting tracking
Logo image, the accuracy of tracking is higher.
Accompanying drawing explanation
Fig. 1 is UUV video camera pictorial diagram;
Fig. 2 is target light source system structure schematic diagram;
Fig. 3 is heart-shaped light source schematic diagram;
Fig. 4 a is the centralized positioning result figure of heart-shaped target light source;
Fig. 4 b is the centralized positioning result schematic diagram of heart-shaped target light source;
Fig. 5 is present invention average drifting based on dimension self-adaption tracking flow chart;
Fig. 6 a is the tradition average drifting tracking tracking effect figure when 4 frame to heart-shaped target light source;
Fig. 6 b is the tradition average drifting tracking tracking effect figure when 32 frame to heart-shaped target light source;
Fig. 6 c is the tradition average drifting tracking tracking effect figure when 63 frame to heart-shaped target light source;
Fig. 6 d is the tradition average drifting tracking tracking effect figure when 96 frame to heart-shaped target light source;
Fig. 7 a is the inventive method tracking effect figure when 4 frame to heart-shaped light source;
Fig. 7 b is the inventive method tracking effect figure when 32 frame to heart-shaped light source;
Fig. 7 c is the inventive method tracking effect figure when 63 frame to heart-shaped light source;
Fig. 7 d is the inventive method tracking effect figure when 96 frame to heart-shaped light source;
Fig. 8 is the inventive method and the tracing positional deviation comparison diagram of tradition mean shift algorithm.
Detailed description of the invention
Detailed description of the invention one: combine Fig. 1,2,5 explanation present embodiment, the UUV of present embodiment removal process under water
In based on dimension self-adaption average drifting tracking, specifically prepare according to following steps:
Heart-shaped target light source center y in step one, establishment initial frame0;
Step 2, according to target light source center y heart-shaped in initial frame0Set up heart-shaped target light source modelAnd obtain the heart-shaped target light source center y of present frame1;
Wherein, δ represents Kronecker delta function, δ [b (xi)-u] as pixel value xiWhen belonging to u eigenvalue
It is 1, is 0 on the contrary;K () represents kernel function, and h represents kernel function bandwidth, and C is normaliztion constant;
Step 3, set up heart-shaped target light source candidate target model at present frame
Obtain Bhattacharyya likeness coefficient
Wherein, ChFor normaliztion constant;nhRepresent sum of all pixels in region of search;
Step 4, candidate target position, and obtain candidate's heart target area position y2And Similarity value
Step 5, set a width of h of kernel function band of previous framep, a width of h of kernel function band of present frame employingopt, use positive and negative
The mode of increment is to hoptIt is modified, i.e.
H=(1+ Δ h* λi)hp (8)
Wherein, λi=-1,0,1, Δ h takes 0.1hp;
And forward step 2 to, obtain 3 Similarity value according to different kernel function bandwidth h
(by kernel function bandwidth h (λiWhen taking-1) it is brought into step 2, until obtaining Similarity value
At the kernel function bandwidth h (λ that step 5 is drawniWhen taking 0) it is brought into step 2, until obtaining Similarity valueAt the kernel function bandwidth h (λ that step 5 is drawniWhen taking 1) it is brought into step 2, until
Obtain Similarity value
To 3 Similarity value obtainedComparing, Similarity value is to the maximumNow corresponding y2It is (the definite position, new position of position, candidate's heart target area
Put);Corresponding bandwidth h is as optimum kernel function bandwidth hoptValue, in order to not make bandwidth tetchiness, formula (9) obtain finally
Kernel function bandwidth hnew, in order to not make bandwidth tetchiness, formula (9) obtain final kernel function bandwidth hnew
hnew=γ hopt+(1-γ)hp (9)
Wherein, γ is smoothing parameter, takes 0.1;Relatively ρ (p (y1), q) with ρ (p (y2), q), as ρ (p (y1),q)>ρ(p
(y2), time q),Obtain ρ (p (y2),q);
When | | y2-y1| | < during ε (ε is the threshold value set), then iterative process terminates, and tracking terminates;Otherwise, y is made1=
y2, return step 3, until | | y2-y1| | < ε, tracking terminates.
Detailed description of the invention two: present embodiment is unlike detailed description of the invention one: establish initial described in step one
Frame center's shape target light source center y0Centre coordinate be;Detailed process is:
Occur in during target light source system motion in the visual field of UUV (underwater unmanned vehicle) monocular-camera
Time, UUV monocular-camera gathers the sequence image of target light source system, to the sequence image of the target light source system collected
First two field picture chooses heart-shaped target light source (the first two field picture manually uses rectangle frame choose heart-shaped light source target), to first
The heart-shaped target light source (such as accompanying drawing 3) of frame carries out centralized positioning, i.e. 9 light sources to the heart-shaped target of the first frame and asks in each
Heart coordinate (utilizes centroid estimation), if their centre coordinate is
In formula, Pn(u, v) represents the pixel image coordinate of the n-th light source, and N represents the sum of all pixels of the n-th light source, (ui,
vi) represent ith pixel pixel image coordinate;
Owing to target light source system is about heart-shaped target light source centrosymmetry, thus the center obtaining heart-shaped target light source is sat
Mark (representing the center in cabin, depressed place):
In formula, PcU () represents the abscissa of centre coordinate, PcV () represents the vertical coordinate of centre coordinate.
Other step and parameter are identical with detailed description of the invention one.
Detailed description of the invention three: present embodiment is unlike detailed description of the invention one or two: basis described in step 2
Heart-shaped target light source center y in initial frame0Set up heart-shaped target light source modelTool
Body process is:
Show the center of target heart light source with '+' word table, to the centralized positioning result such as accompanying drawing 4a of heart-shaped target light source,
4b。
(target light source system: the practical situation of way of recycling of carrying a load on the back under water in view of UUV, the present invention use a kind of towards
Linear light source array to line traffic control position, " to line traffic control position " principle is that straight line is established by 2, controls UUV and makes horizontal, longitudinal direction, bow
To reaching position and posture.The feature of this target light source system is:
(1), the central point of 9 light sources arrange point-blank, adjacent light source spacing is equal and be fixed value;
(2), light-source system be centrosymmetric, heart-shaped light source is positioned at center, and left and right is respectively uniformly distributed 4 light sources;
(3), heart-shaped light source at the center of linear array of source and is positioned at cabin, depressed place Platform center;
(4), heart-shaped light source include 4 LED, wherein 1 the most luminous, remaining 3 the heart-shaped light sources of composition.
All light sources all do water-proofing treatment, the present invention realize during UUV reclaims under water to the vision of heart-shaped target light source with
Track.)
The present invention utilizes the adaptive updates strategy of target scale, improves traditional average drifting tracking, step
As follows:
In initial frame, formula (2) obtain the centre coordinate of heart-shaped target light source.
Mean shift process substantially belongs to feature based track algorithm, utilizes the rectangular histogram of target area to represent that it is special
Levy space, then by the Bhattacharyya similar value of target area template with candidate target region template, utilize average to float
The amount of shifting to iteration obtains best candidate target area.Specifically comprise the following steps that
1 height and length is utilized to be respectively hx、hyTrack window initialize heart-shaped target light source, the length and width of this track window
Also it is the bandwidth of kernel function simultaneously;Owing to target area can be described by rectangular histogram, if target area is by m in n interval
Level gray-scale pixels composition, each feature space is also designated as bin;
In present frame, the location of pixels of target area is { xi}I=1 ... n, b (xi) x is describediThe histogrammic index of pixel at place
Value, x*Representing heart-shaped target light source center of choosing, the model representation of heart-shaped target light source is q={qu}U=1 ... m, whereinPixel coordinate x and y in target zone is passed through hxAnd hyCarry out normalization, be 1 by the range set of k (), then
Heart-shaped target light source model (eigenvalue q in heart-shaped target light source modeluThe density fonction estimated) it is expressed as:
Wherein, δ represents Kronecker delta (Kronecker function) function, δ [b (xi)-u] as pixel value xiOwnership
It is 1 during u eigenvalue, is 0 on the contrary;K () represents kernel function, h represents kernel function bandwidth, usually the target area length of side
1/2, use and select Epanechnikov kernel function, i.e.
Wherein, C is normaliztion constant, in order to ensurex*Represent and choose the heart
Shape light source target center;Probability characteristics u=1,2 ..., m;quRepresent choose heart-shaped light source target color histogram grid feature to
The u component in amount;N is the pixel sum in rectangle frame, and m is grey level histogram grid sum.
Other step and parameter are identical with detailed description of the invention one or two.
Detailed description of the invention four: present embodiment is unlike one of detailed description of the invention one to three: described in step 3
Set up heart-shaped target light source candidate target modelObtain Bhattacharyya
Likeness coefficientDetailed process is:
Candidate target region refers to represent in ensuing picture frame mesh target area,
If heart-shaped target light source candidate target model center position is y,Represent the picture of target area in present frame
Element position, nhRepresent sum of all pixels in region of search;
Then heart-shaped target light source candidate target model is: (then eigenvalue p in heart-shaped target light source modeluY () is estimated close
Degree distribution function is expressed as;)
Wherein, in order to ensureNormaliztion constantObtain Bhattacharyya
Likeness coefficient
Other step and parameter are identical with one of detailed description of the invention one to three.
Detailed description of the invention five: present embodiment is unlike one of detailed description of the invention one to four: described in step 4
Candidate target positions, and obtains candidate's heart target area position y2And Similarity valueSpecifically
Process is:
Candidate target location is to find the region maximum with heart-shaped target light source model similarity in the picture, i.e. makes
Bhattacharyya coefficient is maximum, it is possible to obtain candidate heart target light source center y2;
Candidate heart target light source center y2For:
In formula, g (x) is to kernel function derivation;
Wherein, weight coefficient wiFor
{ p is calculated according to formula (3)u(y2)}U=1 ..., m, and obtain Similarity value
Traditional mean shift process is owing to securing the bandwidth of kernel function so that it can not change continuous to yardstick well
The target changed effectively is followed the tracks of.The present invention utilizes the adaptive updates strategy of target scale, improves algorithm.
Other step and parameter are identical with one of detailed description of the invention one to four.
Employing following example checking beneficial effects of the present invention:
The underwater picture sequence of one group of target scale change is carried out heart-shaped target light source centralized positioning, and it is used biography
System average drifting tracking is tested with average drifting tracking based on dimension self-adaption, experimental result such as Fig. 6 a,
Shown in 6b, 6c, 6d, Fig. 7 a, 7b, 7c, 7d.Owing to the kernel function bandwidth of tradition average drifting tracking immobilizes, it is impossible to
Adaptive follow heart-shaped light source dimensional variation, and correct optimum Bhattacharyya Similarity value cannot be obtained and occur
Deviation, when 63 frame, rectangle tracking window has deviateed target light source center, and during 96 frame, drift conditions is serious and follows the tracks of target
Lose.And this underwater picture sequence is preferably followed the tracks of in present invention average drifting based on dimension self-adaption track side's rule
Effect, carries out adaptive renewal, in target travel due to kernel function bandwidth (tracking window size) along with light source dimensional variation
During be satisfied by requirement.Two kinds of methods tracing positional aberration curve in physical image coordinate system is as shown in Figure 8, it is possible to
Find out that very large deviation has occurred when 43 frame in traditional average drifting tracking, after 90 frames, follow the tracks of track rejection, this
Bright tracing positional deviation is the most more stable and less, and tracking accuracy has reached actual requirement.
Claims (5)
- Based on dimension self-adaption average drifting tracking in 1.UUV removal process under water, it is characterised in that UUV reclaimed under water Journey specifically follows the steps below based on dimension self-adaption average drifting tracking:Heart-shaped target light source center y in step one, establishment initial frame0;Step 2, according to target light source center y heart-shaped in initial frame0Set up heart-shaped target light source model And obtain the heart-shaped target light source center y of present frame1;Wherein, δ represents Kronecker delta function, δ [b (xi)-u] as pixel value xiIt is 1 when belonging to u eigenvalue, It is 0 on the contrary;K () represents kernel function, and h represents kernel function bandwidth, and C is normaliztion constant;Step 3, set up heart-shaped target light source candidate target model at present frame Obtain Bhattacharyya likeness coefficientWherein, ChFor normaliztion constant;nhRepresent sum of all pixels in region of search;Step 4, candidate target position, and obtain candidate's heart target area position y2And Similarity valueStep 5, set a width of h of kernel function band of previous framep, a width of h of kernel function band of present frame employingopt, use positive negative increment Mode is to hoptIt is modified, i.e.H=(1+ △ h* λi)hp (8)Wherein, λi=-1,0,1, △ h takes 0.1hp;And forward step 2 to, obtain 3 Similarity value according to different kernel function bandwidth hTo acquirement 3 Similarity valueComparing, Similarity value is to the maximum Now corresponding y2It is the new position of position, candidate's heart target area;Relatively ρ (p (y1), q) with ρ (p (y2), q), as ρ (p (y1),q)>ρ(p(y2), time q),Obtain ρ (p (y2),q);When | | y2-y1| | < during ε, then iterative process terminates, and tracking terminates;Otherwise, y is made1=y2, return step 3, until | | y2-y1| | < ε, tracking terminates.
- The most according to claim 1 based on dimension self-adaption average drifting tracking, its feature in UUV removal process under water It is: described in step one, establish heart-shaped target light source center y in initial frame0Centre coordinate be;Detailed process is:When occurring in during target light source system motion in the visual field of UUV monocular-camera, UUV monocular-camera gathers mesh The sequence image of mark light-source system, chooses heart-shaped target to the first two field picture of the sequence image of the target light source system collected Light source, carries out centralized positioning to the heart-shaped target light source of the first frame, i.e. asks for 9 light sources of the heart-shaped target of the first frame each Centre coordinate, centre coordinate isIn formula, Pn(u, v) represents the pixel image coordinate of the n-th light source, and N represents the sum of all pixels of the n-th light source, (ui,vi) table Show the pixel image coordinate of ith pixel;Owing to target light source system is about heart-shaped target light source centrosymmetry, thus obtain the centre coordinate of heart-shaped target light source:In formula, PcU () represents the abscissa of centre coordinate, PcV () represents the vertical coordinate of centre coordinate.
- The most according to claim 2 based on dimension self-adaption average drifting tracking, its feature in UUV removal process under water It is: according to target light source center y heart-shaped in initial frame described in step 20Set up heart-shaped target light source modelDetailed process is:In present frame, the location of pixels of target area is { xi}I=1 ... n, b (xi) x is describediThe histogrammic index value of pixel at place, the heart The model representation of shape target light source is q={qu}U=1 ... m, whereinBe 1 by the range set of k (), then heart-shaped mesh Mark source model is expressed as:Wherein, δ represents Kronecker delta function, δ [b (xi)-u] as pixel value xiIt is 1 when belonging to u eigenvalue, It is 0 on the contrary;K () represents kernel function, and h represents kernel function bandwidth, for the 1/2 of the target area length of side, uses and selects Epanechnikov kernel function, i.e.Wherein, C is normaliztion constant, in order to ensurex*Represent and choose heart-shaped light Source target's center;Probability characteristics u=1,2 ..., m;quRepresent and choose in heart-shaped light source target color histogram grid characteristic vector The u component;N is the pixel sum in rectangle frame, and m is grey level histogram grid sum.
- The most according to claim 3 based on dimension self-adaption average drifting tracking, its feature in UUV removal process under water It is: described in step 3, set up heart-shaped target light source candidate target model? To Bhattacharyya likeness coefficientDetailed process is:If heart-shaped target light source candidate target model center position is y,Represent the pixel position of target area in present frame Put, nhRepresent sum of all pixels in region of search;Then heart-shaped target light source candidate target model is:Wherein, in order to ensureNormaliztion constantObtain Bhattacharyya phase Like property coefficient
- The most according to claim 4 based on dimension self-adaption average drifting tracking, its feature in UUV removal process under water It is: candidate target location described in step 4, obtains position, candidate's heart target areaAnd phase Like angle valueDetailed process is:Candidate target location is to find the region maximum with heart-shaped target light source model similarity in the picture, i.e. makes Bhattacharyya coefficient is maximum, it is possible to obtain candidate heart target light source center y2;Candidate heart target light source center y2For:In formula, g (x) is to kernel function derivation;Wherein, weight coefficient wiFor{ p is calculated according to formula (3)u(y2)}U=1 ..., m, and obtain Similarity value
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201610242997.0A CN105931268A (en) | 2016-04-18 | 2016-04-18 | Mean shift tracking method based on scale adaption in UUV underwater recovery process |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201610242997.0A CN105931268A (en) | 2016-04-18 | 2016-04-18 | Mean shift tracking method based on scale adaption in UUV underwater recovery process |
Publications (1)
Publication Number | Publication Date |
---|---|
CN105931268A true CN105931268A (en) | 2016-09-07 |
Family
ID=56838431
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201610242997.0A Pending CN105931268A (en) | 2016-04-18 | 2016-04-18 | Mean shift tracking method based on scale adaption in UUV underwater recovery process |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN105931268A (en) |
Cited By (2)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN110068331A (en) * | 2018-01-24 | 2019-07-30 | 北京致感致联科技有限公司 | Underwater navigation positioning device and system |
CN111473787A (en) * | 2019-01-23 | 2020-07-31 | 北京致感致联科技有限公司 | Underwater navigation positioning equipment and system |
-
2016
- 2016-04-18 CN CN201610242997.0A patent/CN105931268A/en active Pending
Non-Patent Citations (1)
Title |
---|
陈胜蓝等: "核窗宽自适应的均值偏移跟踪算法", 《湖南工业大学学报》 * |
Cited By (2)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN110068331A (en) * | 2018-01-24 | 2019-07-30 | 北京致感致联科技有限公司 | Underwater navigation positioning device and system |
CN111473787A (en) * | 2019-01-23 | 2020-07-31 | 北京致感致联科技有限公司 | Underwater navigation positioning equipment and system |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN105787962B (en) | A kind of monocular vision tracking recycled under water based on UUV | |
CN109784333A (en) | Based on an objective detection method and system for cloud bar power channel characteristics | |
CN105719352B (en) | Face three-dimensional point cloud super-resolution fusion method and apply its data processing equipment | |
CN109740665A (en) | Shielded image ship object detection method and system based on expertise constraint | |
CN106683118B (en) | Unmanned aerial vehicle target tracking method based on hierarchical model | |
CN104899590A (en) | Visual target tracking method and system for unmanned aerial vehicle | |
CN103268616A (en) | Multi-feature multi-sensor method for mobile robot to track moving body | |
CN103413352A (en) | Scene three-dimensional reconstruction method based on RGBD multi-sensor fusion | |
Chen et al. | Transforming a 3-d lidar point cloud into a 2-d dense depth map through a parameter self-adaptive framework | |
CN110544269A (en) | twin network infrared target tracking method based on characteristic pyramid | |
CN108876799A (en) | A kind of real-time step detection method based on binocular camera | |
CN110796741B (en) | Airborne laser sounding point cloud filtering method based on bidirectional cloth simulation | |
Aykin et al. | On feature extraction and region matching for forward scan sonar imaging | |
CN102156995A (en) | Video movement foreground dividing method in moving camera | |
CN109859202B (en) | Deep learning detection method based on USV water surface optical target tracking | |
CN110070565A (en) | A kind of ship trajectory predictions method based on image superposition | |
CN109583509A (en) | Data creation method, device and electronic equipment | |
CN108364305A (en) | Vehicle-mounted pick-up video target tracking method based on modified DSST | |
CN102063707B (en) | Mean shift based grey relation infrared imaging target segmentation method | |
CN103578083A (en) | Single image defogging method based on joint mean shift | |
CN111241970A (en) | SAR image sea surface ship detection method based on yolov3 algorithm and sliding window strategy | |
CN105100546A (en) | Movement estimation method and device | |
CN110400274B (en) | Infrared image enhancement method for vehicle-mounted infrared pedestrian detection | |
CN107730539B (en) | Autonomous underwater robot control system and sonar target tracking method | |
CN110097047A (en) | A kind of vehicle checking method using single line laser radar based on deep learning |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
C06 | Publication | ||
PB01 | Publication | ||
C10 | Entry into substantive examination | ||
SE01 | Entry into force of request for substantive examination | ||
WD01 | Invention patent application deemed withdrawn after publication | ||
WD01 | Invention patent application deemed withdrawn after publication |
Application publication date: 20160907 |