CN104537690A - Motor point target detection method based on maximum-time index union - Google Patents
Motor point target detection method based on maximum-time index union Download PDFInfo
- Publication number
- CN104537690A CN104537690A CN201410830954.5A CN201410830954A CN104537690A CN 104537690 A CN104537690 A CN 104537690A CN 201410830954 A CN201410830954 A CN 201410830954A CN 104537690 A CN104537690 A CN 104537690A
- Authority
- CN
- China
- Prior art keywords
- time index
- image
- candidate
- field picture
- target
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Granted
Links
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T7/00—Image analysis
- G06T7/20—Analysis of motion
- G06T7/215—Motion-based segmentation
-
- 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)
- Multimedia (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- Image Analysis (AREA)
Abstract
The invention discloses a motor point target detection method based on maximum-time index union. The method comprises the following steps: 1) a maximum gray value projected image and a time index frame image are obtained; 2) noise filtering process is performed on the time index frame image; 3)a motor point target in the time index frame image is detected. The motor point target detection method based on the maximum-time index union can effectively detect the motion trail of dispersed punctiform targets in a sequence image and need no prior information of the targets. The calculated amount is small and the instantaneity is high.
Description
Technical field
The invention belongs to technical field of image processing, be specifically related to a kind of moving spot targets detection method based on maximal value-time index associating.
Background technology
The detection of moving spot targets, recognition and tracking technology are one of important contents of sequence video image motion target identification, the research of sequence image tracker was mainly concentrated on closely in the past, general objective, strong signal to noise ratio (S/N ratio) when.And the general objective under this kind of condition contains more rich texture and profile information, can various features be chosen to its detection, single frames basis is carried out the detection and indentification of target.
Point moving target due to image-forming range comparatively far away, target imaging on photodetector is only the imaging area of a several even pixel, shows as weakness shape in systems in which, and lacks shape and structural information; Target is more weak relative to background gray scale, and signal to noise ratio (S/N ratio) is very low, single be difficult to same noise range from gray scale and separate, and must process by the movable information that comprises in the image sequence by sensor imaging of combining target.During analysis except gray scale and movable information can utilize, other information that can utilize for segmentation and detection algorithm is little, and the moving object detection and tracking method of general based target structure cannot use.Its difficult point of detection and tracking in short for the faint Moving point target in sequence image is mainly:
(1) target image is point-like, and target support region size is tending towards a little, and the information such as amorphism size texture can utilize.
(2) brightness of image motion point target and the contrast of its neighborhood background lower.
(3) dot image target noise clutter in its neighborhood is lower than very, just lower to the noise clutter ratio of image area whole in gamut.
(4) must search in whole time-space domain appears in target at random.
(5) target travel rule makes the searching and detecting of target image become more difficult without fixed pattern.
(6) in object tracking process, because target signal is lower, usually cause algorithm to detect multiple false target thus affect the performance of tracking filter.
At present, in image sequence, the detection method of moving spot targets mainly can be divided into Detect before Track and root-first search two class methods.The advanced row space filter preprocessing of detection method of Detect before Track, single-frame images realize target is strengthened and background suppress, improve the signal to noise ratio (S/N ratio) of image, the detection of target is carried out again on this basis by the method for Threshold detection, then in image sequence after singulation according to the continuity of the movement locus of target, confirmed by data fusion, multiframe, the target trajectory that detecting and tracking is possible, i.e. " single frame detection and multiframe confirm " strategy.Mainly comprise: (1) transform domain detection-wavelet transformation, Curvlet conversion etc.; (2) array configurations are improved based on the method-median filter of background estimating and suppression, least means square, high-pass filtering, spatial matched filtering, Largest Mean filtering and they some and adaptive background is estimated; (3) morphologic filtering-expansion, corrosion, Top-Hat conversion etc.
Root-first search algorithm does not judge there being driftlessness in single-frame images, but first possible track more in multiple image data is followed the tracks of simultaneously, the authenticity of certain criterion to every bar track is adopted to make soft judgement, progressively reject the false track be made up of noise, maintain real trace.When soft judgement exceedes a certain thresholding, just make the hard judgement that this track is target trajectory, avoid the track caused because signal to noise ratio (S/N ratio) is low undetected, improve detection probability.
Root-first search method comprises the method based on dynamic programming, the method based on maximum likelihood, the method based on projective transformation, three-dimensional matching matrix, Multistage hypothesis method, and high-order is correlated with, particle filter scheduling algorithm.
Detect before Track algorithm relies on single-frame images to detect target, and when target signal is lower, target detection rate can lower.And root-first search method can make full use of multiple image enhancing moving spot targets information, thus improve target detection rate.But conventional root-first search method exists needs moving spot targets prior imformation (speed, direction etc.) or algorithm complexity, and calculated amount is large, does not utilize the shortcomings such as detection in real time.
Summary of the invention
For the above-mentioned defect existing for prior art, the invention provides a kind of moving spot targets detection method based on maximal value-time index associating, the method effectively can detect the movement locus of discrete point target in sequence image, without the need to the prior imformation of target, calculated amount is little, real-time.
Concrete technical scheme of the present invention is:
Based on a moving spot targets detection method for maximal value-time index associating, it is characterized in that, comprise the following steps:
1) maximum gradation value projected image and time index two field picture is obtained;
1.1) choosing image size is that the K frame sequence image of M × N is as data sample f (i, j, k); Described data sample is separate Gaussian random variable, and noise is white Gaussian noise;
1.2) by data sample, obtain maximum gradation value projected image z (i, j), its expression formula is as follows:
z(i,j)=max[f(i,j,1),f(i,j,2),...,f(i,j,K)]
Wherein i=1 ..., M, j=1 ..., N;
1.3) acquisition time index frame image is expressed as maxposi (i, j); The temporal information that when described time index two field picture is each pixel position generation maximum gradation value of record maximum gradation value projected image, each pixel is corresponding;
2) noise filtering process is carried out to time index two field picture;
3) moving spot targets in time index two field picture is detected;
3.1) candidate tracks is obtained for the first time;
3.1.1) in time index two field picture, choose the connected domain alternatively target of L frame; If candidate target is P
l; Wherein, 1≤L≤K
3.1.2) with candidate target P
lstart search; Search radius is R, R is moving target maximum movement speed threshold value; Candidate target point P is found in hunting zone
l-1and P
l+1after, calculate P
l-P
l-1and P
l+1-P
lthe inner product of two vectors, if two vector P
l-P
l-1and P
l+1-P
lmould be respectively D
l-1Land D
lL+1, the angle between two vectors is θ;
If the mould of two vectors meets | D
l-1L-D
lL+1|≤D
th, and | cos θ-1|≤σ
th, then P is judged
l-1, P
l, P
l+1for candidate tracks, otherwise be judged to be pseudo-track; Described candidate tracks is straight path;
Wherein, D
thbe expressed as vector length difference limen value, span is 1 ~ 3 pixel; σ
thbe the threshold value of angle between two vectors, span is 0 ~ 0.1;
3.2) secondary obtains candidate tracks;
3.2.1) according to step 3.1.2) candidate tracks that obtains, obtain the direction of candidate tracks and the velocity estimation value V of moving target, wherein: V=(D
l-1L+ D
lL+1)/2;
3.2.2) carry out searching for forward and backward along candidate tracks direction, set up straight line tracking window and search and track window backward respectively;
3.2.3) by the candidate target point P in straight line tracking window
l-2with step 3.1) in candidate target carry out the calculating of inner product of vector, if two vector P
l-1-P
l-2and P
l-P
l-1mould be respectively D
l-2L-1and D
l-1L, the angle between two vectors is θ;
If the mould of vector meets | D
l-2L-1-D
l-1L|≤D
th, and | cos θ-1|≤σ
th, then the candidate target point judging in straight line tracking window is the point on candidate tracks, otherwise is judged to be pseudo-tracing point;
Wherein, D
thbe expressed as vector length difference limen value, span is 1 ~ 3 pixel.σ
thbe the threshold value of angle between two vectors, span is 0 ~ 0.1.
3.2.4) step 3.2.3 is repeated), judge that whether candidate target point in search and track window is backward the point on candidate tracks;
3.3) to step 3.2) in all candidate tracks points carry out the time numbering judge, finally determine the track of moving spot targets;
If all candidate tracks points are the relation increased progressively continuously, then think that this candidate tracks is the track of final moving spot targets; Otherwise be pseudo-track.
Above-mentioned steps 2) time index two field picture carries out noise filtering and comprises the following steps:
2.1) binary image is obtained
Adopt adaptive threshold fuzziness to carry out binaryzation to maximum gradation value projected image, obtain binary image;
Its expression formula is: V
th=μ+α σ;
In formula: μ represents the weighted mean of all pixel values of view picture maximal value Gray Projection image; σ is the variance of image view picture maximal value Gray Projection image, and α is coefficient, and α gets 3 ~ 5; V
thfor segmentation threshold, BWz (i, j) represents the bianry image obtained;
2.2) area performance filtering; Described area performance filtering by setting area threshold filter, or passes through morphologic filtering;
2.3) time index two field picture is associated with the temporal information of carrying out of binary image, the time index two field picture after acquisition time information association;
timetag(i,j)=maxposi(i,j)×BWz;
Wherein, timetag (i, j) represent temporal information association after time index two field picture;
Above-mentioned steps 2) to carry out noise filtering be time index information filter to time index two field picture; Described time index information filter carries out noise filtering by arranging difform temporal information template.
Above-mentioned temporal information template is rectangle or rhombus or cross or upper triangle or lower triangle.
The invention has the advantages that:
1, method of the present invention is by obtaining maximal value gray-scale value projected image and the time index two field picture of sequence image, the characteristics of motion of a moving target is converted to the time index information of maximal value gray-scale value projection, achieve the patten transformation of three-dimensional series image to maximum gradation value projected image and time index two field picture, greatly reduce data calculated amount.
2, method of the present invention is in time index two field picture, utilize not coplanar characteristic and the time information characteristic of ground unrest and point target, carry out noise filtering, then utilize distance, angle information and movement locus monotonicity in time to extract target, have that calculated amount is little, real-time good, be easy to hard-wired feature.
3, present invention employs Adaptive Thresholding to split maximum gradation value projected image, generate binary image, area performance filtering and time index information filter mode is adopted to carry out noise filtering process to time index two field picture to binary image, can in real time neatly filtering affect the ground unrest of target detection, calculate simple.
Accompanying drawing explanation
Fig. 1 is employing method 1 workflow schematic diagram of the present invention when carrying out noise filtering.
Fig. 2 is employing method 2 workflow schematic diagram of the present invention when carrying out noise filtering.
Fig. 3 is specific embodiments of the invention schematic diagrams.
Embodiment
The present invention proposes a kind of moving spot targets detection method based on maximal value-time index associating, the overall procedure of the method comprises: obtain maximum gradation value projected image, noise filtering, determine the track of final moving spot targets.
Below in conjunction with accompanying drawing, concrete steps of the present invention are described:
Step 1) obtain maximum gradation value projected image and time index two field picture;
Step 1.1) (M, N are picture altitude and width, are 128 ~ 2048 according to the common span of actual conditions M and N to choose the K that image size is M × N; The span of K is generally 3 ~ 15) frame sequence image is as data sample f (i, j, k); Described data sample is separate Gaussian random variable, and noise is white Gaussian noise;
Step 1.2) by data sample, obtain maximum gradation value projected image z (i, j), its expression formula is as follows:
z(i,j)=max[f(i,j,1),f(i,j,2),...,f(i,j,K)]
Wherein i=1 ..., M, j=1 ..., N;
Step 1.3) acquisition time index frame image is expressed as maxposi (i, j); The temporal information that when described time index two field picture is record maximum gradation value projected image each pixel position generation maximum gradation value, each pixel is corresponding;
Divide into groups to the sequence image containing moving spot targets in this case, image size is 1024 × 1024, and image figure place is 16bit, and every 5 frames are that a frame group carries out target detection;
Step 2) noise filtering process is carried out to time index two field picture;
It should be noted that: noise filtering process comprises two kinds of methods, respectively as shown in Fig. 1, Fig. 2:
The concrete steps of method 1 are:
2.1) binary image is obtained
Adaptive threshold fuzziness is carried out to maximum value projection image, the binary image after splitting can be obtained; Threshold value calculates as follows:
V
th=μ+α·σ (2)
In formula, μ is the weighted mean of all pixel values of view picture maximal value gray-scale value projected image, and calculating formula is:
σ is the variance of view picture maximal value gray-scale value projected image, and calculating formula is:
α is a coefficient relevant with noise, generally gets 3 ~ 5.Along with α value improves, threshold value can improve thereupon, and false-alarm probability can reduce, but detection probability also can reduce.M, N are picture altitude and width, all get 1024;
Step 2.2) area performance filtering; Area performance filtering by setting area threshold filter, or passes through morphologic filtering; (note: because noise generally shows as discrete isolated single pixel, and to put moving target be discrete Gaussian distribution, occupy certain pixel number, by setting area threshold value filtering noise, or by morphologic filtering (burn into expands, opens operation, closed operation etc.).
Step 2.3) time index two field picture is associated with the temporal information of carrying out of binary image, the time index two field picture after acquisition time information association;
timetag(i,j)=maxposi(i,j)×BWz;
Wherein, timetag (i, j) represent temporal information association after time index two field picture;
Method 2 is time index information filter; Described time index information filter carries out noise filtering by arranging difform temporal information template.(note: due to the random character of noise, it shows as random temporal information in the time index image of maximal value, and moving spot targets shows as the temporal information in continuous print a certain moment, utilize this characteristic, a certain size difform temporal information template (rectangle, rhombus, cross can be set, upper triangle, lower triangle etc.) carry out noise filtering.
Step 3) moving spot targets in time index two field picture is detected;
Step 3.1) choose candidate tracks for the first time;
Step 3.1.1) in time index two field picture, choose the connected domain alternatively target of L frame; If candidate target is P
3;
Step 3.1.2) with candidate target P
lstart search; Search radius is R, R is moving target maximum movement speed threshold value; P is found in hunting zone
l-1and P
l+1candidate target point after, calculate P
l-P
l-1and P
l+1-P
lthe inner product of two vectors, if two vector P
l-P
l-1and P
l+1-P
lmould be respectively D
l-1Land D
lL+1, the angle between two vectors is θ;
If the mould of two vectors meets | D
l-1L-D
lL+1|≤D
th, and | cos θ-1|≤σ
th, then P is judged
l-1, P
l, P
l+1for candidate tracks, otherwise be judged to be pseudo-track; Candidate tracks is straight path;
Wherein, D
thbe expressed as vector length difference limen value, span is 1 ~ 3 pixel.σ
thbe the threshold value of angle between two vectors, span is 0 ~ 0.1.
Step 3.2) secondary chooses candidate tracks;
Step 3.2.1) according to step 3.1.2) candidate tracks that obtains, obtain the direction of candidate tracks and the velocity estimation value V of moving target, wherein: V=(D
l-1L+ D
lL+1)/2;
3.2.2) carry out searching for forward and backward along candidate tracks direction, set up straight line tracking window and search and track window backward respectively;
3.2.3) by the candidate target point P in straight line tracking window
l-2with step 3.1) in candidate target carry out the calculating of inner product of vector, if two vector P
l-1-P
l-2and P
l-P
l-1mould be respectively D
l-2L-1and D
l-1L, the angle between two vectors is θ;
If the mould of vector meets | D
l-2L-1-D
l-1L|≤D
th, and | cos θ-1|≤σ
th, then the candidate target point judging in straight line tracking window is the point on candidate tracks, otherwise is judged to be pseudo-tracing point;
Wherein, D
thbe expressed as vector length difference limen value, span is 1 ~ 3 pixel.σ
thbe the threshold value of angle between two vectors, span is 0 ~ 0.1;
3.2.4) step 3.2.3 is repeated), judge that whether candidate target point in search and track window is backward the point on candidate tracks;
3.3) to step 3.2) in all candidate tracks points carry out time numbering (each two field picture to there being time numbering, and time numbering continuously) and judge, finally determine the track of moving spot targets;
If all candidate tracks points are the relation increased progressively continuously, then think that this candidate tracks is the track of final moving spot targets; Otherwise be pseudo-track.
As shown in Figure 3, the specific embodiment of this case is:
Choose candidate tracks for the first time;
In time index two field picture access time be numbered 3 connected domain alternatively target P 3 start search, search radius is set to R (R is moving target maximum movement speed threshold value), after finding the time to be numbered candidate target point P2 and P4 of 2 and 4 in hunting zone, calculate the inner product of P2-P3 and P3-P4 two vectors, if the mould of two vectors meets | D
23-D
34|≤D
th, (wherein D
thfor vector length difference limen value, span is 1 ~ 3 pixel), and | cos θ-1|≤σ
th(θ is the angle between two vectors, σ
thbe the threshold value of angle between two vectors, span is 0 ~ 0.1), then judge P2, P3, P4 are a candidate tracks; Otherwise be judged to be pseudo-track.
Secondary chooses candidate tracks;
According to the first candidate tracks obtained, obtain the direction of candidate tracks and calculate the velocity estimation value V of moving target, concrete formula is: V=(D
23+ D
34)/2;
Carry out along candidate tracks direction searching for forward and backward estimated position P1e and P5e obtaining P1 and P5 respectively, centered by estimated position, set up a certain size tracking window forward and tracking window backward respectively, carry out the detection of window internal object.
Candidate target point P1 in tracking window is forward calculated to the inner product of P1-P2 and P2-P3 two vectors.If the mould of two vectors meets | D
12-D
23|≤D
th, (wherein D
thfor vector length difference limen value, span is 1 ~ 3 pixel), and | cos θ-1|≤σ
th(θ is the angle between two vectors, σ
thbe the threshold value of angle between two vectors, span is 0 ~ 0.1), then judge that P1 is candidate tracks point; Otherwise be judged to be pseudo-tracing point.The search in like manner can carrying out P5 candidate tracks point judges.
Carry out time numbering to all candidate tracks points to judge, finally determine the track of moving spot targets;
If all candidate tracks point P1, P2, P3, P4, P5 are the relation increased progressively continuously, this candidate tracks is the track of final moving spot targets; Otherwise be pseudo-track.
After the movement objective orbit utilizing said method to extract in frame group, the calculating of intercepting to movement destination image and sub-pixed mapping precision centroid position can be realized easily in conjunction with original image.
Claims (4)
1., based on a moving spot targets detection method for maximal value-time index associating, it is characterized in that, comprise the following steps:
1) maximum gradation value projected image and time index two field picture is obtained;
1.1) choosing image size is that the K frame sequence image of M × N is as data sample f (i, j, k); Described data sample is separate Gaussian random variable, and noise is white Gaussian noise;
1.2) by data sample, obtain maximum gradation value projected image z (i, j), its expression formula is as follows:
z(i,j)=max[f(i,j,1),f(i,j,2),...,f(i,j,K)]
Wherein i=1 ..., M, j=1 ..., N;
1.3) acquisition time index frame image is expressed as maxposi (i, j); The temporal information that when described time index two field picture is each pixel position generation maximum gradation value of record maximum gradation value projected image, each pixel is corresponding;
2) noise filtering process is carried out to time index two field picture;
3) moving spot targets in time index two field picture is detected;
3.1) candidate tracks is obtained for the first time;
3.1.1) in time index two field picture, choose the connected domain alternatively target of L frame; If candidate target is P
l; Wherein, 1≤L≤K;
3.1.2) with candidate target P
lstart search; Search radius is R, R is moving target maximum movement speed threshold value; Candidate target point P is found in hunting zone
l-1and P
l+1after, calculate P
l-P
l-1and P
l+1-P
lthe inner product of two vectors, if two vector P
l-P
l-1and P
l+1-P
lmould be respectively D
l-1Land D
lL+1, the angle between two vectors is θ;
If the mould of two vectors meets | D
l-1L-D
lL+1|≤D
th, and | cos θ-1|≤σ
th, then P is judged
l-1, P
l, P
l+1for candidate tracks, otherwise be judged to be pseudo-track; Described candidate tracks is straight path;
Wherein, D
thbe expressed as vector length difference limen value, span is 1 ~ 3 pixel; σ
thbe the threshold value of angle between two vectors, span is 0 ~ 0.1;
3.2) secondary obtains candidate tracks;
3.2.1) according to step 3.1.2) candidate tracks that obtains, obtain the direction of candidate tracks and the velocity estimation value V of moving target, wherein: V=(D
l-1L+ D
lL+1)/2;
3.2.2) carry out searching for forward and backward along candidate tracks direction, set up straight line tracking window and search and track window backward respectively;
3.2.3) by the candidate target point P in straight line tracking window
l-2with step 3.1) in candidate target carry out the calculating of inner product of vector; If two vector P
l-1-P
l-2and P
l-P
l-1mould be respectively D
l-2L-1and D
l-1L, the angle between two vectors is θ;
If the mould of vector meets | D
l-2L-1-D
l-1L|≤D
th, and | cos θ-1|≤σ
th, then the candidate target point judging in straight line tracking window is the point on candidate tracks, otherwise is judged to be pseudo-tracing point;
Wherein, D
thbe expressed as vector length difference limen value, span is 1 ~ 3 pixel.σ
thbe the threshold value of angle between two vectors, span is 0 ~ 0.1;
3.2.4) step 3.2.3 is repeated), judge that whether candidate target point in search and track window is backward the point on candidate tracks;
3.3) to step 3.2) in all candidate tracks points carry out the time numbering judge, finally determine the track of moving spot targets;
If all candidate tracks points are the relation increased progressively continuously, then think that this candidate tracks is the track of final moving spot targets; Otherwise be pseudo-track.
2. the moving spot targets detection method based on the associating of maximal value-time index according to claim 1, is characterized in that: described step 2) time index two field picture carries out noise filtering and comprises the following steps:
2.1) binary image is obtained
Adopt adaptive threshold fuzziness to carry out binaryzation to maximum gradation value projected image, obtain binary image;
Its expression formula is: V
th=μ+α σ;
In formula: μ represents the weighted mean of all pixel values of view picture maximal value Gray Projection image; σ is the variance of view picture maximal value Gray Projection image, and α is coefficient, and α gets 3 ~ 5; V
thfor segmentation threshold, BWz (i, j) represents the bianry image obtained;
2.2) area performance filtering; Described area performance filtering by setting area threshold filter, or passes through morphologic filtering;
2.3) time index two field picture is associated with the temporal information of carrying out of binary image, the time index two field picture after acquisition time information association;
timetag(i,j)=maxposi(i,j)×BWz;
Wherein, timetag (i, j) represent temporal information association after time index two field picture.
3. the moving spot targets detection method based on the associating of maximal value-time index according to claim 1, is characterized in that: described step 2) to carry out noise filtering be time index information filter to time index two field picture; Described time index information filter carries out noise filtering by arranging difform temporal information template.
4. the moving spot targets detection method based on the associating of maximal value-time index according to claim 3, is characterized in that: described temporal information template is rectangle or rhombus or cross or upper triangle or lower triangle.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201410830954.5A CN104537690B (en) | 2014-12-26 | 2014-12-26 | One kind is based on the united moving spot targets detection method of maximum time index |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201410830954.5A CN104537690B (en) | 2014-12-26 | 2014-12-26 | One kind is based on the united moving spot targets detection method of maximum time index |
Publications (2)
Publication Number | Publication Date |
---|---|
CN104537690A true CN104537690A (en) | 2015-04-22 |
CN104537690B CN104537690B (en) | 2017-06-27 |
Family
ID=52853209
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201410830954.5A Active CN104537690B (en) | 2014-12-26 | 2014-12-26 | One kind is based on the united moving spot targets detection method of maximum time index |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN104537690B (en) |
Cited By (3)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN106780556A (en) * | 2016-12-19 | 2017-05-31 | 广州视源电子科技股份有限公司 | Motion image detection method and device |
CN107194896A (en) * | 2017-06-05 | 2017-09-22 | 华中科技大学 | A kind of background suppression method and system based on neighbour structure |
CN109919969A (en) * | 2019-01-22 | 2019-06-21 | 广东工业大学 | A method of realizing that visual movement controls using depth convolutional neural networks |
Citations (2)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US20060187356A1 (en) * | 2005-02-24 | 2006-08-24 | Yohzoh Ogura | Video playback apparatus and video playback method |
CN101604448A (en) * | 2009-03-16 | 2009-12-16 | 北京中星微电子有限公司 | A kind of speed-measuring method of moving target and system |
-
2014
- 2014-12-26 CN CN201410830954.5A patent/CN104537690B/en active Active
Patent Citations (2)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US20060187356A1 (en) * | 2005-02-24 | 2006-08-24 | Yohzoh Ogura | Video playback apparatus and video playback method |
CN101604448A (en) * | 2009-03-16 | 2009-12-16 | 北京中星微电子有限公司 | A kind of speed-measuring method of moving target and system |
Non-Patent Citations (7)
Title |
---|
丛明煜等: "《复杂背景成像条件下运动点目标的轨迹提取》", 《光学精密工程》 * |
刘鑫: "《基于投影的运动目标跟踪研究》", 《电子测试》 * |
屈有山等: "《双目凝视系统中运动点目标实时检测算法》", 《光电工程》 * |
胡礼勇等: "《基于投影变换与轨迹关联的目标检测算法》", 《新无线电工程》 * |
陈尚锋等: "《基于加权动态规划和航迹关联的小目标检测技术》", 《国防科技大学学报》 * |
马亚奇等: "《基于最大值投影的运动点目标检测算法及其性能》", 《上海航天》 * |
高珊等: "《基于方向累积的运动点目标检测新算法》", 《激光与红外》 * |
Cited By (5)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN106780556A (en) * | 2016-12-19 | 2017-05-31 | 广州视源电子科技股份有限公司 | Motion image detection method and device |
CN107194896A (en) * | 2017-06-05 | 2017-09-22 | 华中科技大学 | A kind of background suppression method and system based on neighbour structure |
CN107194896B (en) * | 2017-06-05 | 2019-12-17 | 华中科技大学 | Background suppression method and system based on neighborhood structure |
CN109919969A (en) * | 2019-01-22 | 2019-06-21 | 广东工业大学 | A method of realizing that visual movement controls using depth convolutional neural networks |
CN109919969B (en) * | 2019-01-22 | 2022-11-22 | 广东工业大学 | Method for realizing visual motion control by utilizing deep convolutional neural network |
Also Published As
Publication number | Publication date |
---|---|
CN104537690B (en) | 2017-06-27 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
Chadwick et al. | Distant vehicle detection using radar and vision | |
CN114202696B (en) | SAR target detection method and device based on context vision and storage medium | |
Lookingbill et al. | Reverse optical flow for self-supervised adaptive autonomous robot navigation | |
CN105513349B (en) | Mountainous area highway vehicular events detection method based on double-visual angle study | |
CN109598684B (en) | Correlation filtering tracking method combined with twin network | |
CN102609720B (en) | Pedestrian detection method based on position correction model | |
CN104851097B (en) | The multichannel SAR GMTI methods aided in based on target shape and shade | |
CN103886325B (en) | Cyclic matrix video tracking method with partition | |
CN109871763A (en) | A kind of specific objective tracking based on YOLO | |
CN102496016B (en) | Infrared target detection method based on space-time cooperation framework | |
CN104574445A (en) | Target tracking method and device | |
CN104408482A (en) | Detecting method for high-resolution SAR (Synthetic Aperture Radar) image object | |
CN104992453A (en) | Target tracking method under complicated background based on extreme learning machine | |
CN101996401A (en) | Target analysis method and device based on intensity image and range image | |
CN111208479B (en) | Method for reducing false alarm probability in deep network detection | |
Dimitrievski et al. | Weakly supervised deep learning method for vulnerable road user detection in FMCW radar | |
CN104301585A (en) | Method for detecting specific kind objective in movement scene in real time | |
CN101344967B (en) | Detection method for small mobile objective in astronomical image | |
CN104537690A (en) | Motor point target detection method based on maximum-time index union | |
Chen et al. | A novel AMS-DAT algorithm for moving vehicle detection in a satellite video | |
Arróspide et al. | On-board robust vehicle detection and tracking using adaptive quality evaluation | |
CN102509308A (en) | Motion segmentation method based on mixtures-of-dynamic-textures-based spatiotemporal saliency detection | |
CN103996207A (en) | Object tracking method | |
CN116665097A (en) | Self-adaptive target tracking method combining context awareness | |
CN105160287A (en) | Method for detecting characteristics of space-time interest point under condition of no camera movement |
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 | ||
GR01 | Patent grant |