CN105654516B - Satellite image based on target conspicuousness is to ground weak moving target detection method - Google Patents
Satellite image based on target conspicuousness is to ground weak moving target detection method Download PDFInfo
- Publication number
- CN105654516B CN105654516B CN201610091741.4A CN201610091741A CN105654516B CN 105654516 B CN105654516 B CN 105654516B CN 201610091741 A CN201610091741 A CN 201610091741A CN 105654516 B CN105654516 B CN 105654516B
- Authority
- CN
- China
- Prior art keywords
- track
- moving target
- target
- frame
- pixel
- 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.)
- Active
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
-
- 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
- G06T2207/10032—Satellite or aerial image; Remote sensing
-
- 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
- G06T2207/20004—Adaptive image processing
- G06T2207/20012—Locally adaptive
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 kind of satellite images based on target conspicuousness to ground weak moving target detection method, for solving the existing method technical problem low to weak moving target detection rate.Technical solution is to carry out background modeling to Saliency maps picture first, carries out significance analysis to gray level image, strengthens the moving target in image.By carrying out Gaussian Mixture background modeling to Saliency maps picture, solves weak moving target detection problem, generate track using classification, filter out false-alarm using trace information, with respect to the background art method, verification and measurement ratio improves 10%, and false alarm rate reduces 5%.
Description
Technical field
The present invention relates to a kind of satellite images to ground weak moving target detection method, is based on mesh more particularly to one kind
The satellite image of conspicuousness is marked to ground weak moving target detection method.
Background technique
Moving object detection is important subject in computer vision field to satellite image over the ground.Existing satellite mapping
As weak moving target detection algorithm verification and measurement ratio is lower over the ground.Document " Vehicle Detection and Roadside Tree
Shadow Removal in High Resolution Satellite Images, 2010 " disclose a kind of satellite image pair
Ground weak moving target detection method.This method utilizes the geometrical characteristic of vehicle in satellite image, using improved scale circle
Spot matching algorithm detects the elliptical motion vehicle in satellite image.For the vehicle exercised in road, Gauss is oval
Laplce's filtering, keeps road direction in image consistent with the direction of elliptic motion target geometry, when filtering equations analytic expression reaches
To local extremum, the Candidate Motion target in image can be obtained.But this method testing result quality is several dependent on moving target
Where to feature, the geometric direction of moving target, weak moving target detection rate are unable to get for small and weak moving target
Extremely low, false alarm rate is higher.
Summary of the invention
In order to overcome the shortcomings of that existing method is low to weak moving target detection rate, the present invention provides a kind of aobvious based on target
The satellite image of work property is to ground weak moving target detection method.This method carries out background modeling to Saliency maps picture first,
Significance analysis is carried out to gray level image, strengthens the moving target in image.By carrying out Gaussian Mixture back to Saliency maps picture
Scape modeling, solves weak moving target detection problem, generates track using classification, filters out false-alarm using trace information, relatively
In background technique method, verification and measurement ratio improves 10%, and false alarm rate reduces 5%.
The technical solution adopted by the present invention to solve the technical problems is: a kind of satellite image pair based on target conspicuousness
Ground weak moving target detection method, its main feature is that the following steps are included:
Step 1: calculating Saliency maps picture for image currently entered.Remember that certain pixel value is I in imagek, wherein k table
K-th of pixel in diagram picture.Remember that w is with pixel IkCentered on sliding window.| | | | indicate Gray homogeneity matrix between pixel,
diIndicate the distance of pixel k to pixel i.Then image pixel IkLocal significance value be expressed as
Wherein, f () is about diLinear decrease function.
When the adjacent pixel grey scale difference of certain pixel is bigger, a possibility that being moving target, is bigger, therefore with distance
Increase its conspicuousness multiple it is smaller.Each pixel I is obtained for passing through formula (1) in the gray level image of inputkRegional area is aobvious
Work property, ultimately forms local Saliency maps picture.
Step 2: the Background of input picture, root are obtained using the modeling of ADAPTIVE MIXED Gaussian Background to Saliency maps picture
Background subtraction is done according to background image and the Saliency maps picture of input, obtains Candidate Motion target collection.
It is filtered out Step 3: being used to the false-alarm of Candidate Motion target collection based on space time information, it is continuous using target
Property information inhibit false-alarm.The association for generating path segment is layered based on movement, track and space time information.Track association is divided into just
Grade tracklets association is associated with advanced tracklets.
Primary Track association: assuming that t frame is there are m target, target and t of the t+1 frame there are n target, in t frame
Target in+1 is one-to-one relationship.Using the Euclidean distance between the target in present frame and each target of next frame as defeated
Enter matrix, the destination-related information of t frame and t+1 frame is calculated using Hungary Algorithm.Movement if m < n, i.e., in t+1 frame
Target is more, then will be not associated on moving target as new moving target.If m >=n, i.e., target is more in t frame, then will
Target retains two frames, then judges whether it is false-alarm according to subsequent association.
Advanced tracklets association: the test point extracted using Hungary Algorithm from every frame forms short track set.Benefit
Guarantee the accuracy of long track with kinematic similarity constraint, track similarity constraint and space-time restriction.
Kinematic similarity constraint: linear functionIt indicates the move distance function of target, indicates
Path segment viIn m-th of position coordinate, Δ t indicate time interval, v indicate in this time interval velocity constant.It enablesIndicate detectionWithBetween time interval.Then time upper adjacent track viAnd vjRespectively respectively backward and to
Preceding prediction only calculates τ frame, then track to reduce computation complexityWithBetween kinematic similarity are as follows:
Wherein, dis1And dis2It is respectively as follows:
Track similarity constraint: the vehicle moved on road, its track are straight line in shorter time-domain.
Therefore using RANSAC algorithm according to short track fitting straight line.Track viOn point should be distributed in track viThat estimates is straight
Line LiAbove or near this straight line.The track being associated with by false-alarm point, then abandon this track.Two tracks are such as
Fruit belongs to same target then two track viAnd vjBetween track similitude be expressed as
Wherein,It is the track midpoint jWith straight line LiThe distance of straight line.
Space-time restriction: since there can be no different in two different positions, same frame in the same time for vehicle
It is not same target that the moving target of position is inevitable.It must if the range for walking out satellite shooting for the vehicle moved on road
It cannot be so associated with the target in present viewing field, and path segment viAnd vjIf there is overlapping region or one of them
It is existing to be associated with track, is associated as S (v between two tracki,vj)=0.
According to three constraint conditions, the similarity constraint obtained between two path segments is as follows:
S(vi,vj)=w1Sm(vi,vj)+w2St(vi,vj) (6)
Wherein, Sm() and St() respectively indicates kinematic similarity and track similitude.w1And w2It is its weight factor.
After obtaining the similitude between different tracks, it is associated with to form candidate target track using Hungary Algorithm, due to false-alarm
The test point of generation flicker cause association after track and non-rectilinear, therefore to the track of generation carry out straight line fitting.Root
According to best matching degree, i.e., meets interior rate for obtaining linear equation and target trajectory collection in track and calculate of model
With degree, the number of point on straight line is counted, retains the parameter of the best linear equation of matching degree.When path length is more than
TlengthAnd interior rate is higher than TinnerWhen, then it is assumed that the moving target on this track is real target.
So far small dim moving target in satellite image is obtained, by the movement mesh after filtering out based on space time information false-alarm
It is denoted as exporting objective result in the picture for real moving target.
The beneficial effects of the present invention are: this method carries out background modeling to Saliency maps picture first, gray level image is carried out
Significance analysis strengthens the moving target in image.By carrying out Gaussian Mixture background modeling to Saliency maps picture, solve weak
Small moving object detection problem generates track using classification, filters out false-alarm using trace information, with respect to the background art method,
Verification and measurement ratio improves 10%, and false alarm rate reduces 5%.
It elaborates With reference to embodiment to the present invention.
Specific embodiment
The present invention is based on the satellite images of target conspicuousness, and to ground weak moving target detection method, specific step is as follows:
(a) Saliency maps picture calculates: calculating Saliency maps picture for image currently entered.Certain pixel value is in note image
Ik, wherein k indicates k-th of pixel in image.Remember that w is with pixel IkCentered on sliding window.| | | | indicate gray scale between pixel
Distance matrix, diIndicate the distance of pixel k to pixel i.Then image pixel IkLocal significance value be expressed as
Wherein, f () is about diLinear decrease function.
The image sequence for inputting the shooting of one section of satellite calculates each pixel to current frame image first with local conspicuousness
Local conspicuousness.In order to guarantee computational efficiency and detection performance, it is 3 that the size of sliding window w is taken in this example.
When the adjacent pixel grey scale difference of certain pixel is bigger, a possibility that being moving target, is bigger, therefore with distance
Increase its conspicuousness multiple it is smaller.For passing through formula (1) available each pixel I in the gray level image of inputkPartial zones
Domain conspicuousness ultimately forms local Saliency maps picture.
(b) Gaussian Mixture background modeling: since mixed Gaussian background modeling has stronger adaptability for complex background,
Therefore to Saliency maps picture using ADAPTIVE MIXED Gaussian Background modeling obtain input picture Background, according to background image with it is defeated
The Saliency maps picture entered does background subtraction, obtains Candidate Motion target collection.
In view of computation complexity and accuracy, the local Saliency maps picture of input is carried on the back using ADAPTIVE MIXED Gauss
Scape modeling method calculates background image.The back of each pixel is approached using the Gaussian Profile of 3 different weights in this example
Scape distribution.Then initial background model is calculated using the ten frame consecutive images most started.The image that background subtraction obtains is carried out
Thresholding, threshold value are set as 15.So far Candidate Motion target collection is obtained.
(c) false-alarm based on space time information filters out: in view of the external appearance characteristic of moving target is unavailable, therefore to Candidate Motion
The false-alarm of target collection is used to be filtered out based on space time information, inhibits false-alarm using goal succession information.Based on movement, rail
Mark and space time information layering generate the association of path segment.Track association is divided into primary tracklets association and advanced
Tracklets association.
Primary Track association: assuming that t frame is there are m target, target and t of the t+1 frame there are n target, in t frame
Target in+1 is one-to-one relationship.Using the Euclidean distance between the target in present frame and each target of next frame as defeated
Enter matrix, the destination-related information of t frame and t+1 frame is calculated using Hungary Algorithm.Due to the presence of false-alarm, upper and lower two frame
Between the quantity of moving target may be different.If m < n, i.e., moving target in t+1 frame may be more, then will be not associated on
Moving target is as new moving target.If m >=n, i.e., target is more in t frame, then target is retained two frames, then after
Continuous association judges whether it is false-alarm.
Advanced tracklets association: the test point extracted using Hungary Algorithm from every frame forms short track set.Benefit
Guarantee the accuracy of long track with kinematic similarity constraint, track similarity constraint and space-time restriction.
Kinematic similarity constraint: linear functionIt indicates the move distance function of target, indicates
Path segment viIn m-th of position coordinate, Δ t indicate time interval, v indicate in this time interval velocity constant.It enablesIndicate detectionWithBetween time interval.Then time upper adjacent track viAnd vjRespectively respectively backward and to
Preceding prediction, to reduce computation complexity, we only calculate τ frame, then trackWithBetween kinematic similarity
Are as follows:
Wherein, dis1And dis2It is respectively as follows:
Track similarity constraint: the vehicle moved on road, its track are straight line in shorter time-domain.
Therefore using RANSAC algorithm according to short track fitting straight line.Track viOn point should be distributed in track viThat estimates is straight
Line LiAbove or near this straight line.The track being associated with by false-alarm point, then abandon this track.Two tracks are such as
Fruit belongs to same target then two track viAnd vjBetween track similitude can be expressed as
Wherein,It is the track midpoint jWith straight line LiThe distance of straight line.
Space-time restriction: since there can be no different in two different positions, same frame in the same time for vehicle
It is not same target that the moving target of position is inevitable.It must if the range for walking out satellite shooting for the vehicle moved on road
It cannot be so associated with the target in present viewing field, and if path segment viAnd vjIf there is overlapping region or wherein it
One association track is existing, is associated as S (v between two tracki,vj)=0.
According to three constraint conditions, the similarity constraint between available two path segments is as follows:
S(vi,vj)=w1Sm(vi,vj)+w2St(vi,vj) (6)
Wherein, Sm() and St() respectively indicates kinematic similarity and track similitude.w1And w2It is its weight factor.
After obtaining the similitude between different tracks, it is associated with to form candidate target track using Hungary Algorithm, due to false-alarm
The test point of generation flicker cause association after track and non-rectilinear, therefore to the track of generation carry out straight line fitting.Root
According to best matching degree, i.e., meets interior rate for obtaining linear equation and target trajectory collection in track and calculate of model
With degree, the number of point on straight line is counted, retains the parameter of the best linear equation of matching degree.In the present invention, work as track
Length is more than TlengthAnd interior rate is higher than TinnerWhen, then it is assumed that the moving target on this track is real target.
To Candidate Motion target, layering Track association, the kinematic similarity constraint of advanced Track association, to reduce meter are used
Complexity is calculated, the frame of τ=5 is chosen.Kinematic similarity and track similarity weight w1And w2Respectively 0.8 and 0.2.According to acquisition
Track fitting straight line, as path length TlengthMore than or equal to 10, and point rate T in fitting a straight lineinnerWhen higher than 0.6, then it is assumed that should
Test point on track is realistic objective test point, exports the small dim moving target point detected in the picture.
So far small dim moving target in satellite image is obtained, by the movement mesh after filtering out based on space time information false-alarm
It is denoted as exporting objective result in the picture for real moving target.
Claims (1)
1. a kind of satellite image based on target conspicuousness is to ground weak moving target detection method, it is characterised in that including with
Lower step:
Step 1: calculating Saliency maps picture for current input image;Remember that certain pixel value is I in imagek, wherein k is indicated in image
K-th of pixel, i.e. pixel k;Remember that W is the sliding window centered on pixel k;| | | | indicate Gray homogeneity matrix between pixel,
doIndicate the distance of pixel k to pixel o;Then the local conspicuousness of image pixel k is expressed as
Wherein, f () is about doLinear decrease function, therefore its significance value is smaller with the increase of distance between pixels;When
A possibility that adjacent pixel grey scale difference of certain pixel is bigger, is moving target is bigger;The gray level image of input is passed through
Formula (1) obtains the local conspicuousness of each pixel k, ultimately forms Saliency maps picture;
Step 2: the background image of input picture is obtained using the modeling of ADAPTIVE MIXED Gaussian Background to Saliency maps picture, to back
Scape image and Saliency maps picture do background subtraction, obtain Candidate Motion target collection;
Step 3: the false-alarm to Candidate Motion target collection is filtered out using space time information, pressed down using goal succession information
False-alarm processed;The association for generating path segment is layered based on movement, track and space time information;Track association is divided into primary track to close
Connection and advanced Track association;
Primary Track association: assuming that t frame, there are a moving target of m ', there are a moving targets of n ' for t+1 frame;It will be in present frame
Moving target and each moving target of next frame between Euclidean distance as input matrix, calculate the using Hungary Algorithm
The moving target related information of t frame and t+1 frame;If m ' < n ', i.e., the moving target in t+1 frame is more, then will be not associated on
Moving target as new moving target;If m ' >=n ', i.e., moving target is more in t frame, then moving target is retained two
Then frame judges whether it is false-alarm according to subsequent association;
Advanced Track association: extracting test point from every frame using Hungary Algorithm, forms short track set;Utilize kinematic similarity
Constraint, track similarity constraint and space-time restriction guarantee the accuracy of long track;
Kinematic similarity constraint: linear functionIndicate the move distance function of target,Indicate rail
Mark viIn the corresponding position of m frame coordinate, Δ t indicates time interval, and v indicates velocity constant in this time interval;
It enablesIndicate track viInWithBetween time interval;Adjacent track v on timeiWith track vjRespectively backward and
Forward prediction only calculates τ frame, then track v to reduce computation complexityiAnd vjBetween kinematic similarity are as follows:
Wherein, dis1And dis2It is respectively as follows:
Track similarity constraint: the vehicle moved on road, its track are straight line in shorter time-domain;Therefore
Using RANSAC algorithm according to short track fitting straight line;Track viOn point should be distributed in track viThe straight line L of estimationi
Above or near this straight line;If certain track is obtained by false-alarm point, this track is abandoned;Two track vi
And vjBetween track similitude be expressed as
Wherein,It is track viMidpointWith straight line LjLinear distance;
Space-time restriction: since there can be no the different locations in two different positions, same frame in the same time for vehicle
Moving target it is inevitable be not same target;It is inevitable not if the range for walking out satellite shooting for the vehicle moved on road
It can be associated with the target in present viewing field, and track viAnd vjIf there is overlapping region or the association track of one of them
It is existing, S (v is associated as between two tracki,vj)=0;
According to kinematic similarity constraint, track similarity constraint and space-time restriction, the similarity constraint between two tracks is obtained
It is as follows:
S(vi,vj)=w1Smotion(vi,vj)+w2Strack(vi,vj) (6)
Wherein, Smotion() and Strack() respectively indicates kinematic similarity and track similitude;w1And w2It is weight factor;
It after obtaining the similitude between different tracks, is associated with to form candidate target track using Hungary Algorithm, since false-alarm generates
Test point flicker cause association after track and non-rectilinear, therefore to the track of generation carry out straight line fitting;According to rail
The matching degree for the straight line that the interior rate and candidate target track for meeting gained linear equation in mark are obtained with fitting, reservation
With the parameter for spending best linear equation;When path length is more than TlengthAnd interior rate is higher than TinnerWhen, then it is assumed that in this rail
Moving target on mark is real moving target;
So far small dim moving target in satellite image is obtained, the moving target after being filtered out based on space time information false-alarm is made
For real moving target, moving target result is exported in the picture.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201610091741.4A CN105654516B (en) | 2016-02-18 | 2016-02-18 | Satellite image based on target conspicuousness is to ground weak moving target detection method |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201610091741.4A CN105654516B (en) | 2016-02-18 | 2016-02-18 | Satellite image based on target conspicuousness is to ground weak moving target detection method |
Publications (2)
Publication Number | Publication Date |
---|---|
CN105654516A CN105654516A (en) | 2016-06-08 |
CN105654516B true CN105654516B (en) | 2019-03-26 |
Family
ID=56488552
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201610091741.4A Active CN105654516B (en) | 2016-02-18 | 2016-02-18 | Satellite image based on target conspicuousness is to ground weak moving target detection method |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN105654516B (en) |
Families Citing this family (8)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US10002109B2 (en) * | 2016-06-24 | 2018-06-19 | Freeport-Mcmoran Inc. | Systems and methods of correlating satellite position data with terrestrial features |
CN107895384A (en) * | 2017-12-01 | 2018-04-10 | 中国科学院长春光学精密机械与物理研究所 | Target extraction method and device |
CN109712169A (en) * | 2018-11-15 | 2019-05-03 | 上海卫星工程研究所 | Moving-target motion profile prediction technique and method for autonomous tracking based on EO-1 hyperion |
CN110084231A (en) * | 2019-04-29 | 2019-08-02 | 北京富吉瑞光电科技有限公司 | A kind of method and apparatus of infrared panoramic system multi-target detection |
CN110057449B (en) * | 2019-04-29 | 2021-06-18 | 上海卫星工程研究所 | Air target height and speed calculation method based on field-of-view spectroscopic hyperspectral detection |
CN110502968B (en) * | 2019-07-01 | 2022-03-25 | 西安理工大学 | Method for detecting infrared small and weak moving target based on track point space-time consistency |
CN111862006A (en) * | 2020-06-30 | 2020-10-30 | 北京北方智图信息技术有限公司 | Detection method and device for small aircraft |
CN112818905B (en) * | 2020-12-10 | 2022-08-26 | 北京航空航天大学 | Finite pixel vehicle target detection method based on attention and spatio-temporal information |
Citations (10)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN101344967A (en) * | 2008-09-02 | 2009-01-14 | 西北工业大学 | Detection method for small mobile objective in astronomical image |
CN101520892A (en) * | 2009-03-17 | 2009-09-02 | 西北工业大学 | Detection method of small objects in visible light image |
CN101546429A (en) * | 2009-05-05 | 2009-09-30 | 西北工业大学 | Detection method of small target of feature space |
CN101567087A (en) * | 2009-05-25 | 2009-10-28 | 北京航空航天大学 | Method for detecting and tracking small and weak target of infrared sequence image under complex sky background |
CN102081800A (en) * | 2011-01-06 | 2011-06-01 | 西北工业大学 | Method for detecting spatial weak moving target |
CN102096829A (en) * | 2011-01-06 | 2011-06-15 | 西北工业大学 | Iterative optimization distance categorization-based space weak and small target detection method |
CN102103748A (en) * | 2010-12-14 | 2011-06-22 | 西北工业大学 | Method for detecting and tracking infrared small target in complex background |
CN104143197A (en) * | 2014-07-31 | 2014-11-12 | 中国科学院自动化研究所 | Detection method for moving vehicles in aerial photography scene |
CN104200492A (en) * | 2014-08-25 | 2014-12-10 | 西北工业大学 | Automatic detecting and tracking method for aerial video target based on trajectory constraint |
CN105261043A (en) * | 2015-11-20 | 2016-01-20 | 天津大学 | Video motion object detection method on the basis of significance detection |
Family Cites Families (1)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
TW201001338A (en) * | 2008-06-16 | 2010-01-01 | Huper Lab Co Ltd | Method of detecting moving objects |
-
2016
- 2016-02-18 CN CN201610091741.4A patent/CN105654516B/en active Active
Patent Citations (10)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN101344967A (en) * | 2008-09-02 | 2009-01-14 | 西北工业大学 | Detection method for small mobile objective in astronomical image |
CN101520892A (en) * | 2009-03-17 | 2009-09-02 | 西北工业大学 | Detection method of small objects in visible light image |
CN101546429A (en) * | 2009-05-05 | 2009-09-30 | 西北工业大学 | Detection method of small target of feature space |
CN101567087A (en) * | 2009-05-25 | 2009-10-28 | 北京航空航天大学 | Method for detecting and tracking small and weak target of infrared sequence image under complex sky background |
CN102103748A (en) * | 2010-12-14 | 2011-06-22 | 西北工业大学 | Method for detecting and tracking infrared small target in complex background |
CN102081800A (en) * | 2011-01-06 | 2011-06-01 | 西北工业大学 | Method for detecting spatial weak moving target |
CN102096829A (en) * | 2011-01-06 | 2011-06-15 | 西北工业大学 | Iterative optimization distance categorization-based space weak and small target detection method |
CN104143197A (en) * | 2014-07-31 | 2014-11-12 | 中国科学院自动化研究所 | Detection method for moving vehicles in aerial photography scene |
CN104200492A (en) * | 2014-08-25 | 2014-12-10 | 西北工业大学 | Automatic detecting and tracking method for aerial video target based on trajectory constraint |
CN105261043A (en) * | 2015-11-20 | 2016-01-20 | 天津大学 | Video motion object detection method on the basis of significance detection |
Non-Patent Citations (5)
Title |
---|
AN EFFECTIVE DETECTION ALGORITHM FOR MOVING OBJECT WITH COMPLEX BACKGROUND;FU-YUAN HU et al.;《Proceedings of the Fourth International Conference on Machine Learning and Cybernetics》;20050831;5011-5015 |
Small and Dim Moving Target Detection in Deep Space Background;Jinqiu Sun et al.;《MMSP 2008》;20081231;902-905 |
Vehicle Detection in Very High Resolution Satellite Images of City Areas;Jens Leitloff et al.;《IEEE TRANSACTIONS ON GEOSCIENCE AND REMOTE SENSING》;20100731;第48卷(第7期);2795-2806 |
Visual Attention Detection in Video Sequences Using Spatiotemporal Cues;Yun Zhai et al.;《Proceedings of the 14th ACM international conference on Multimedia》;20061231;815-824 |
基于迭代距离分类与轨迹关联检测空间弱小目标;姚睿 等;《光学精密工程》;20120131;第20卷(第1期);179-189 |
Also Published As
Publication number | Publication date |
---|---|
CN105654516A (en) | 2016-06-08 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN105654516B (en) | Satellite image based on target conspicuousness is to ground weak moving target detection method | |
CN108986064B (en) | People flow statistical method, equipment and system | |
CN109919974B (en) | Online multi-target tracking method based on R-FCN frame multi-candidate association | |
CN109242884B (en) | Remote sensing video target tracking method based on JCFNet network | |
CN102542289B (en) | Pedestrian volume statistical method based on plurality of Gaussian counting models | |
CN104574445B (en) | A kind of method for tracking target | |
CN104392468B (en) | Based on the moving target detecting method for improving visual background extraction | |
CN103400157B (en) | Road pedestrian and non-motor vehicle detection method based on video analysis | |
CN101339655B (en) | Visual sense tracking method based on target characteristic and bayesian filtering | |
CN106203513B (en) | A kind of statistical method based on pedestrian's head and shoulder multi-target detection and tracking | |
CN104680559B (en) | The indoor pedestrian tracting method of various visual angles based on motor behavior pattern | |
CN103426179B (en) | A kind of method for tracking target based on mean shift multiple features fusion and device | |
CN103310444B (en) | A kind of method of the monitoring people counting based on overhead camera head | |
CN102609720B (en) | Pedestrian detection method based on position correction model | |
CN107424171A (en) | A kind of anti-shelter target tracking based on piecemeal | |
CN103886325B (en) | Cyclic matrix video tracking method with partition | |
CN106600625A (en) | Image processing method and device for detecting small-sized living thing | |
CN104574439A (en) | Kalman filtering and TLD (tracking-learning-detection) algorithm integrated target tracking method | |
CN104517095B (en) | A kind of number of people dividing method based on depth image | |
CN104794463A (en) | System and method for achieving indoor human body falling detection based on Kinect | |
CN103729854A (en) | Tensor-model-based infrared dim target detecting method | |
CN107689052A (en) | Visual target tracking method based on multi-model fusion and structuring depth characteristic | |
CN104835182A (en) | Method for realizing dynamic object real-time tracking by using camera | |
CN103854292A (en) | Method and device for calculating number of people and population motion direction | |
CN110853353B (en) | Vision-based density traffic vehicle counting and traffic flow calculating method and system |
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 |