CN103279952B - A kind of method for tracking target and device - Google Patents
A kind of method for tracking target and device Download PDFInfo
- Publication number
- CN103279952B CN103279952B CN201310183174.1A CN201310183174A CN103279952B CN 103279952 B CN103279952 B CN 103279952B CN 201310183174 A CN201310183174 A CN 201310183174A CN 103279952 B CN103279952 B CN 103279952B
- Authority
- CN
- China
- Prior art keywords
- matched
- image
- reference picture
- point set
- feature point
- 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
Landscapes
- Image Analysis (AREA)
Abstract
The embodiment of the invention discloses a kind of method for tracking target, the fisrt feature point set of the reference picture is obtained including extracting the Corner Feature of reference picture using Predistribution Algorithm, reject the abnormity point for exceeding the image range to be matched in the fisrt feature point set, obtain second feature point set, the second feature point set is subjected to images match with the image to be matched, the target following of the image to be matched is completed according to matching result.The embodiment of the invention also discloses a kind of target tracker.Using the present invention, the Corner Feature of extraction has good identification and stability, greatly improve the speed of matching, the set of characteristic points in reference picture is optimized simultaneously, reject the angle point of over range, the deficiency that characteristic point over range can not be matched in the prior art is solved, the reliability of matching is improved.
Description
Technical field
The present invention relates to image processing field, more particularly to a kind of method for tracking target and device.
Background technology
Target detection technique is also generally referred to as the feature extraction of image, is computer vision and Digital Image Processing
Concept, it refers to determine whether the pixel of each image is that the image information of characteristics of image was extracted using Computing
Journey.Target following technology is exactly that area-of-interest (Region Of are determined in video image or consecutive image sequence
Interest, ROI) template (Template) position is also referred to as, and the template of each two field picture is mapped.
Existing tracking technique scheme can be described as follows：
(1) in the image sequence of collection on determine to have recorded in a template, commonly referred to as reference picture, reference picture
Need tracked target；
(2) in order to improve the efficiency of computing, each pixel in reference picture is not converted, but from reference picture
Inside it is separated by and equidistantly pixel is uniformly extracted, the process is referred to alternatively as equidistant sampling, forms primitive character point set；
(3) computing is made to primitive character point set and its realm information, obtains new feature point set, according to new feature point set
Matching area is determined on image to be matched；
(4) half-tone information between matching area and reference template, such as SSD (Sum of Squared are calculated
Differences, squared difference and) method, using the method for minimizing error, pass through iteration so that matching area with reference
It can be matched between image, even if matching area needs tracked mesh with there is identical image information in reference picture
Mark.
(5) to image sequence repeat step (3)-(4) collected, by the template matches between picture frame and frame, most
Target is realized eventually continuously to track.
But, prior art has the following disadvantages：
(1) when obtaining the set of characteristic points of reference picture, use and the pixel in reference picture is carried out equidistantly
The method of sampling.The characteristic point so obtained is random big, and the image information generally comprised is less, it is impossible to good phenogram picture
Feature, reliability, stability is not high so that track algorithm does not possess good robustness.
(2) obtain after reference picture, just as standard form, no longer make any change.But in actual tracking
In system, with the motion of the image capture devices such as video camera, with this information it is possible to determine reference picture can part remove video camera figure
As acquisition range, it is impossible to be imaged, so that the subregion of reference picture is not among the image of successive image sequence.But it is existing
There is scheme to use initial reference picture, can to calculate gradation of image information algorithm between reference picture and matching area without
Method restrains, it is impossible to obtain correct result, so as to cause tracking to fail.
The content of the invention
Technical problem to be solved of the embodiment of the present invention is that there is provided a kind of method for tracking target.Existing skill can be solved
Characteristics of image stability is not high in art and reference picture over range causes the deficiency that tracking fails.
In order to solve the above-mentioned technical problem, first aspect present invention provides a kind of method for tracking target, including：
The Corner Feature for extracting reference picture using Predistribution Algorithm obtains the fisrt feature point set of the reference picture, institute
Stating reference picture is used to be tracked the target in current image to be matched；
The abnormity point for exceeding the image range to be matched in the fisrt feature point set is rejected, second feature point is obtained
Set；
The second feature point set is subjected to images match with the image to be matched, according to being completed matching result
The target following of image to be matched.
In the first possible implementation, the Predistribution Algorithm includes：Harris Corner Detection Algorithms, FAST accelerate
Segmentation detection feature Corner Detection Algorithm, KLT Corner Detection Algorithms or SUSAN most small nut value similar area Corner Detection Algorithms.
It is described to use in second of possible implementation with reference to first aspect or the first possible implementation
Before the step of Corner Feature that Predistribution Algorithm extracts reference picture obtains the fisrt feature point set of the reference picture, also wrap
Include：
The image sequence of camera acquisition is obtained, the image sequence is regard as image to be matched；
And the reference picture of the image to be matched is determined from the first two field picture of described image sequence.
It is described to reject in the third possible implementation with reference to second of possible implementation of first aspect
Exceed the abnormity point of the image range to be matched in the fisrt feature point set, the step of obtaining second feature point set is wrapped
Include：
Obtain the coordinate of each characteristic point in the fisrt feature point set；
The scope of the image to be matched is seen if fall out according to the coordinate of each characteristic point in the set of characteristic points,
If it has, then determining that this feature point is abnormity point；
By abnormity point elimination all in the fisrt feature point set, second feature point set is obtained.
It is described by institute in the 4th kind of possible implementation with reference to the third possible implementation of first aspect
State second feature point set and carry out images match with the image to be matched, the image to be matched is completed according to matching result
The step of target following, includes：
The subgraph to be matched of one and the reference picture formed objects are chosen in the image to be matched；
The feature for carrying the subgraph to be matched using the Predistribution Algorithm obtains third feature point set；
Calculate the distance between the second feature point set and described third feature point set；
Judge whether the distance is less than preset threshold value, if it has, then determining the reference picture and the figure to be matched
Subgraph match to be matched success as in, one and the ginseng are chosen if it has not, then re-executing in the image to be matched
The step of examining the subgraph to be matched of image formed objects.
It is described by institute in the 5th kind of possible implementation with reference to the 4th kind of possible implementation of first aspect
State second feature point set and carry out images match with the image to be matched, the image to be matched is completed according to matching result
After the step of target following, in addition to：
Delete the reference picture and using with the reference template subgraph to be matched that the match is successful as new with reference to figure
Picture.
Correspondingly, second aspect of the present invention additionally provides a kind of target tracker, including：
Characteristic extracting module, the Corner Feature for extracting reference picture using Predistribution Algorithm obtains the reference picture
Fisrt feature point set, the reference picture is used to be tracked the target in current image to be matched；
Characteristic optimization module, the exception of the image range to be matched is exceeded for rejecting in the fisrt feature point set
Point, obtains second feature point set；
Images match module, for the second feature point set to be carried out into images match, root with the image to be matched
The target following of the image to be matched is completed according to matching result.
In the first possible implementation, the characteristic extracting module is used to extract reference picture using Predistribution Algorithm
Corner Feature obtain the fisrt feature point set of the reference picture, the Predistribution Algorithm includes Harris Corner Detections and calculated
Method, FAST Accelerated fractionations detection feature Corner Detection Algorithm, KLT Corner Detection Algorithms or SUSAN most small nut value similar area angles
Point detection algorithm.
With reference to second aspect and the first possible implementation, in second of possible implementation, in addition to：
Template determining module, the image sequence for obtaining camera acquisition, regard the image sequence as image to be matched；
And the reference picture of the image to be matched is determined from the first two field picture of described image sequence.
With reference to second of possible implementation, in the third possible implementation, the characteristic optimization module bag
Include：
Acquiring unit, the coordinate for obtaining each characteristic point in the fisrt feature point set；
Judging unit, for treating described in being seen if fall out according to the coordinate of each characteristic point in the set of characteristic points
Scope with image, if it has, then determining that this feature point is abnormity point；
Culling unit, for by abnormity point elimination all in the fisrt feature point set, obtaining second feature point set
Close.
With reference to the third possible implementation, in the 4th kind of possible implementation, described image matching module bag
Include：
Unit is chosen, it is to be matched with the reference picture formed objects for choosing one in the image to be matched
Subgraph；
Extraction unit, the feature for extracting the subgraph to be matched using the Predistribution Algorithm obtains third feature point set
Close；
Computing unit, for calculating the distance between the second feature point set and described third feature point set；
With reference to the 4th kind of possible implementation, in the 5th kind of possible implementation, in addition to：
Template renewal module, for delete the reference picture and by with the reference template to be matched that the match is successful
Figure is used as new reference picture.
Implement the embodiment of the present invention, have the advantages that：
The set of characteristic points of the reference picture is obtained by extracting the Corner Feature of reference picture, is rejected in set of characteristic points
Beyond the abnormity point of image range to be matched, the set of characteristic points and image to be matched progress image of abnormity point will be rejected
Match somebody with somebody, to realize target following.Carrying out Graphic Pattern Matching using Corner Feature can subtract while image graphics key character is retained
Few data volume for participating in calculating, Corner Feature has good identification and stability, greatly improves the speed of matching.Originally simultaneously
Method is optimized to the set of characteristic points in reference picture, is rejected the angle point of over range, is solved characteristic point in the prior art
The deficiency that over range can not be matched, improves the reliability of matching.
Brief description of the drawings
In order to illustrate more clearly about the embodiment of the present invention or technical scheme of the prior art, below will be to embodiment or existing
There is the accompanying drawing used required in technology description to be briefly described, it should be apparent that, drawings in the following description are only this
Some embodiments of invention, for those of ordinary skill in the art, on the premise of not paying creative work, can be with
Other accompanying drawings are obtained according to these accompanying drawings.
Fig. 1 is a kind of schematic flow sheet of method for tracking target of the embodiment of the present invention；
Fig. 2 is a kind of another schematic flow sheet of method for tracking target of the embodiment of the present invention；
Fig. 3 is a kind of structural representation of target tracker of the embodiment of the present invention；
Fig. 4 is a kind of another structural representation of target tracker of the embodiment of the present invention；
Fig. 5 is the structural representation of characteristic optimization module in Fig. 4；
Fig. 6 is the structural representation of images match module in Fig. 4；
Fig. 7 is a kind of another structural representation of target tracker of the embodiment of the present invention.
Embodiment
Below in conjunction with the accompanying drawing in the embodiment of the present invention, the technical scheme in the embodiment of the present invention is carried out clear, complete
Site preparation is described, it is clear that described embodiment is only a part of embodiment of the invention, rather than whole embodiments.It is based on
Embodiment in the present invention, it is every other that those of ordinary skill in the art are obtained under the premise of creative work is not made
Embodiment, belongs to the scope of protection of the invention.
It is a kind of schematic flow sheet of method for tracking target of the embodiment of the present invention referring to Fig. 1, this method includes：
Step 101, the Corner Feature for extracting reference picture using Predistribution Algorithm obtain the fisrt feature of the reference picture
Point set, the reference picture is used to be tracked the target in current image to be matched.
Specifically, angle point does not have clear and definite mathematical definition, it is that two dimensional image brightness changes violent point to generally believe angle point
Or in the curvature of image border curvature maximum point.In the various features of image, angle point be a kind of stabilization, invariable rotary and
The validity feature of gray inversion can be overcome.The algorithm of Corner Feature that target tracker is extracted in reference picture can be
Harris Corner Detection Algorithms, FAST Corner Detection Algorithms, KLT Corner Detection Algorithms or SUSAN Corner Detection Algorithms, can also
It is other algorithms, the present invention is not restricted, the Corner Feature that target tracker extracts reference picture obtains fisrt feature point set
Close, the fisrt feature point set is combined into a bianry image.Reference picture needs the image of tracked target to record, using this
Reference picture is matched to realize the tracking to target with image to be matched.
Step 102, the abnormity point for exceeding the image range to be matched in the fisrt feature point set is rejected, obtain the
Two set of characteristic points.
Specifically, when the shape of the target in image to be matched is imperfect, such as target is blocked or target is being treated
Exist when matching outside image, then in the fisrt feature point set extracted from reference picture and be not belonging within image range to be matched
Angle point, as abnormity point, if these abnormity points participate in images match calculating, matching primitives result can be caused not restrain
Fail to correct result, i.e. reference picture and images match to be matched.In order to avoid the generation of such case, target tracker
The coordinate of each angle point in fisrt feature point set is obtained, angle is judged according to the coordinate relation of the coordinate of angle point and image to be matched
Whether point exceeds image range to be matched, if it has, then obtaining the angle point for abnormity point, rejects all according to above-mentioned method
Abnormity point, obtains second feature point set.
Step 103, the second feature point set with the image to be matched is subjected to images match, according to matching result
Complete the target following of the image to be matched.
Specifically, choose the subgraph to be matched of one piece and reference picture formed objects in image to be matched, using with step
The Corner Feature that rapid 101 identical Robust Algorithm of Image Corner Extraction extracts the subgraph to be matched obtains third feature point set, based on angle point
The matching of feature has a two ways, it is a kind of be described with predetermined mode the second feature point set of reference picture extracted and
The third feature point set of subgraph to be matched, the matching to image translates into the matching of two vector sets.Another is to carry
The set of characteristic points got is represented with bianry image, and range conversion is carried out to bianry image, and two spies are measured with Distance conformability degree
Levy the similitude of point set.Other method can certainly be used, the present invention is not restricted.
Implement embodiments of the invention, the feature point set of the reference picture is obtained by extracting the Corner Feature of reference picture
Close, reject the abnormity point for exceeding image range to be matched in set of characteristic points, the set of characteristic points of abnormity point will be rejected with treating
Match image and carry out images match, to realize target following.Graphic Pattern Matching is carried out using Corner Feature and is retaining image graphics weight
Want that the data volume for participating in calculating can be reduced while feature, Corner Feature has good identification and stability, carried significantly
The speed of height matching.This method is optimized to the set of characteristic points in reference picture simultaneously, rejects the angle point of over range, is solved
The deficiency that characteristic point over range can not be matched in the prior art, improves the reliability of matching.
It is a kind of another embodiment schematic diagram of method for tracking target of the embodiment of the present invention, this method bag referring to Fig. 2
Include：Step 201, the image sequence for obtaining camera acquisition, regard the image sequence as image to be matched；And from described image sequence
The reference picture of the image to be matched is determined in first two field picture of row.
Specifically, the form of the image sequence of camera acquisition can be BMP（Bitmap, bitmap, abbreviation BMP）、TIFF
（Tagged Image File Format, TIF, abbreviation TIFF）Or JPEG（Joint Photographic
Experts Group, Joint Photographic Experts Group, abbreviation JPEG）Deng picture format or MPEG（Moving Pictures
Experts Group/Motin Pictures Experts Group, dynamic image expert group, abbreviation MPEG）Or AVI
（Audio Video Interleaved, Audio Video Interleaved form, abbreviation AVI）Deng video format, target tracker should
Image sequence determines from the first two field picture in image sequence to remember in reference picture, reference picture as image to be matched
Being loaded with needs tracked target, it is assumed that the size of reference picture is M*N（M, N are the number of pixel）, image to be matched it is big
Small is m*n（M, n are the number of pixel）, then have M ＜ m, N ＜ n.
Step 202, the Corner Feature extracted using Predistribution Algorithm in reference picture obtain the first spy of the reference picture
Point set is levied, the reference picture is used to be tracked the target in current image to be matched.
Specifically, the algorithm that target tracker extracts the Corner Feature in reference picture can be Harris Corner Detections
Algorithm, FAST Corner Detection Algorithms, KLT Corner Detection Algorithms or SUSAN Corner Detection Algorithms or other algorithms, this
Invention is not restricted, and the Corner Feature that target tracker extracts reference picture obtains fisrt feature point set, the fisrt feature
Point set is combined into a bianry image.Reference picture needs the image of tracked target to record, using the reference picture with treating
Matching image is matched to realize the tracking to target.
Step 203, reject the abnormity point for exceeding the image range to be matched in the fisrt feature point set, to the
Two set of characteristic points.
Specifically, target tracker obtains the coordinate of each angle point in fisrt feature point set, while obtaining to be matched
The coordinate of the pixel of each in image, judges whether angle point exceeds the scope of image to be matched according to the relation of both coordinates, if
It is yes, it is determined that the angle point is abnormity point, and all abnormity points in fisrt feature point set are rejected according to the method described above, obtains the
Two set of characteristic points.
Step 204, the subgraph to be matched for choosing in the image to be matched one and described image formed objects.
Specifically, with the example in step 201, it is assumed that the size M*N of reference picture, image to be matched is m*n, meets and closes
It is M ＜ m, N ＜ n, target tracker chooses the subgraph to be matched that a size is M*N, son to be matched in image to be matched
The search strategy of figure is typically using the method or genetic algorithm of traversal, and the present invention is not restricted.
Step 205, the Corner Feature for extracting the subgraph to be matched using the Predistribution Algorithm obtain third feature point set
Close.
Specifically, target tracker uses the Corner Feature for extracting subgraph to be matched with step 202 identical algorithm to obtain
To third feature point set.
The distance between step 206, the calculating second feature point set and described third feature point set.
Specifically, target tracker is obtained after carrying out range conversion using preset change scaling method to fisrt feature point set
Bianry image A, bianry image B is obtained to third feature point set using preset become after scaling method carries out range conversion of identical,
The preset change scaling method can be 3-4DT.The distance between bianry image A and bianry image B calculate can using it is European away from
From or non-Euclidean distance, this sentence Hausdorff distance exemplified by, Hausdorff distances are a kind of are defined on two point sets
Minimax（Max-Min）Distance, for example, calculate the Hausdorff distances between above-mentioned bianry image A and bianry image B,
Assuming that A={ a1 ..., ap }, B={ b1 ..., bq }, then the Hausdorff distance definitions between the two point sets be H (A, B)=
max(h(A,B),h(B,A)) (1)
h(A,B)=max（a∈A）min（b∈B）‖a-b‖ (2)
h(B,A)=max（b∈B）min（a∈A）‖b-a‖ (3)
‖ ‖ be between point set A and B point set apart from normal form (such as：L2 or Euclidean distance).
Here, formula (1) is referred to as two-way Hausdorff distances, is the most basic form of Hausdorff distances；In formula (2)
H (A, B) and h (B, A) are referred to as the unidirectional Hausdorff distances from set A to set B and from set B to set A.That is h
(A, B) is actually first to each point ai in point set A to the distance between the set B midpoint bj nearest apart from this point ai ‖ ai-
Bj ‖ are ranked up, and then take the maximum in the distance as h (A, B) value.H (B, A) can similarly be obtained, and be known by formula (1), double
To Hausdorff apart from H (A, B) be both one-way distance h (A, B) and h (B, A) in the greater, it has measured two point sets
Between maximum mismatch degree.
Step 207, judge the distance whether be less than preset threshold value.
Specifically, target tracker judgment step 206 calculates whether obtained distance is less than preset threshold value, if it is,
Step 206 is performed, step 204 is performed if it has not, then returning.
Step 208, the subgraph match to be matched success determined in the reference picture and the image to be matched.
Specifically, target tracker determine the image to be matched include needs tracking target, reference picture with
Images match success to be matched, and trace into the target in subgraph to be matched.
Step 209, delete the reference picture and using with the reference picture subgraph to be matched that the match is successful as new
Reference picture.
Specifically, using the subgraph to be matched that the match is successful in image to be matched as new reference picture, to next frame
When image to be matched carries out target following, images match is carried out using new reference picture.By the dynamic to reference picture more
Newly, the shapes and sizes of target in reference picture can be adjusted, are that images match is more accurate.
Implement embodiments of the invention, the feature point set of the reference picture is obtained by extracting the Corner Feature of reference picture
Close, reject the abnormity point for exceeding image range to be matched in set of characteristic points, the set of characteristic points of abnormity point will be rejected with treating
Match image and carry out images match, to realize target following.Graphic Pattern Matching is carried out using Corner Feature and is retaining image graphics weight
Want that the data volume for participating in calculating can be reduced while feature, Corner Feature has good identification and stability, carried significantly
The speed of height matching.This method is optimized to the set of characteristic points in reference picture simultaneously, rejects the angle point of over range, is solved
The deficiency that characteristic point over range can not be matched in the prior art, improves the reliability of matching.
Referring to Fig. 3, it is a kind of structural representation of target tracker of the embodiment of the present invention, hereinafter referred to as device 1, is somebody's turn to do
Device 1 includes：
Characteristic extracting module 11, the Corner Feature for extracting reference picture using Predistribution Algorithm obtains the reference picture
Fisrt feature point set, the reference picture be used for the target in current image to be matched is tracked.
Specifically, the algorithm that characteristic extracting module 11 extracts the Corner Feature in reference picture can be the inspection of Harris angle points
Method of determining and calculating, FAST Corner Detection Algorithms, KLT Corner Detection Algorithms or SUSAN Corner Detection Algorithms or other algorithms,
The present invention is not restricted, and the Corner Feature that target tracker extracts reference picture obtains fisrt feature point set, first spy
Levy point set and be combined into a bianry image.Reference picture needs the image of tracked target to record, using the reference picture with
Image to be matched is matched to realize the tracking to target.
Characteristic optimization module 12, for rejecting in the fisrt feature point set beyond the different of the image range to be matched
Chang Dian, obtains second feature point set.
Specifically, when the shape of the target in image to be matched is imperfect, such as target is blocked or target is being treated
Exist when matching outside image, then in the fisrt feature point set extracted from reference picture and be not belonging within image range to be matched
Angle point, as abnormity point, if these abnormity points participate in images match calculating, matching primitives result can be caused not restrain
Fail to correct result, i.e. reference picture and images match to be matched.In order to avoid the generation of such case, characteristic optimization module
12 obtain the coordinate of each angle point in fisrt feature point set, are judged according to the coordinate relation of the coordinate of angle point and image to be matched
Whether angle point exceeds image range to be matched, if it has, then obtaining the angle point for abnormity point, rejects all according to above-mentioned method
Abnormity point, obtain second feature point set.
Images match module 13, for the second feature point set to be carried out into images match with the image to be matched,
The target following of the image to be matched is completed according to matching result.
Specifically, images match module 13 chosen in image to be matched one piece it is to be matched with reference picture formed objects
Subgraph, is obtained using the Corner Feature that the subgraph to be matched is extracted with identical Robust Algorithm of Image Corner Extraction in step characteristic extracting module 11
Third feature point set, the matching based on Corner Feature has two ways, and a kind of is the ginseng for being described to extract with predetermined mode
Examine the second feature point set of image and the third feature point set of subgraph to be matched, the matching of image is translated into two to
The matching of quantity set.Another is that the set of characteristic points that will be extracted is represented with bianry image, and range conversion is carried out to bianry image,
The similitude of two set of characteristic points is measured with Distance conformability degree.Other method can certainly be used, the present invention is not restricted.
Implement embodiments of the invention, the feature point set of the reference picture is obtained by extracting the Corner Feature of reference picture
Close, reject the abnormity point for exceeding image range to be matched in set of characteristic points, the set of characteristic points of abnormity point will be rejected with treating
Match image and carry out images match, to realize target following.Graphic Pattern Matching is carried out using Corner Feature and is retaining image graphics weight
Want that the data volume for participating in calculating can be reduced while feature, Corner Feature has good identification and stability, carried significantly
The speed of height matching.This method is optimized to the set of characteristic points in reference picture simultaneously, rejects the angle point of over range, is solved
The deficiency that characteristic point over range can not be matched in the prior art, improves the reliability of matching.
Further, it is a kind of another structural representation of target tracker of the embodiment of the present invention referring to Fig. 4-Fig. 6
Figure, the device 1 in addition to including above-mentioned characteristic extracting module 11, characteristic optimization module 12 and images match module 13, in addition to：
Template determining module 14, the image sequence for obtaining camera acquisition, regard the image sequence as figure to be matched
Picture；And the reference picture of the image to be matched is determined from the first two field picture of described image sequence.
Specifically, template determining module 14 obtain camera acquisition image sequence form can be BMP, TIFF or
The video format such as the picture formats such as JPEG or MPEG or AVI, template determining module 14 is using the image sequence as to be matched
Image, and determine from the first two field picture in image sequence reference picture, tracked mesh in need described in reference picture
Mark, it is assumed that the size of reference picture is M*N（M, N are the number of pixel）, the size of image to be matched is m*n（M, n are pixel
Number）, then have M ＜ m, N ＜ n.
Template renewal module 15, for deleting the reference picture and by that the match is successful is to be matched with the reference template
Subgraph is used as new reference picture.
Specifically, template renewal module 15 using the subgraph to be matched that the match is successful in image to be matched as new with reference to figure
Picture, when carrying out target following to next frame image to be matched, images match is carried out using new reference picture.By to reference
The dynamic renewal of image, can adjust the shapes and sizes of target in reference picture, be that images match is more accurate.
Wherein, characteristic optimization module 12 includes：
Acquiring unit 121, the coordinate for obtaining each characteristic point in the fisrt feature point set；
Judging unit 122, described in being seen if fall out according to the coordinate of each characteristic point in the set of characteristic points
The scope of image to be matched, if it has, then determining that this feature point is abnormity point；
Culling unit 123, for by abnormity point elimination all in the fisrt feature point set, obtaining second feature point
Set.
Images match module 13 includes：
Unit 131 is chosen, is treated for choosing one in the image to be matched with the reference picture formed objects
Matching sub-image；
Extraction unit 132, the feature for extracting the subgraph to be matched using the Predistribution Algorithm obtains third feature
Point set；
Computing unit 133, for calculating the distance between the second feature point set and described third feature point set；
Matching unit 134, for judging whether the distance is less than preset threshold value, if it has, then determining described with reference to figure
Picture and the subgraph match to be matched success in the image to be matched, if it has not, then re-executing in the image to be matched
The step of one is chosen with the subgraphs to be matched of the reference picture formed objects.
Implement embodiments of the invention, the feature point set of the reference picture is obtained by extracting the Corner Feature of reference picture
Close, reject the abnormity point for exceeding image range to be matched in set of characteristic points, the set of characteristic points of abnormity point will be rejected with treating
Match image and carry out images match, to realize target following.Graphic Pattern Matching is carried out using Corner Feature and is retaining image graphics weight
Want that the data volume for participating in calculating can be reduced while feature, Corner Feature has good identification and stability, carried significantly
The speed of height matching.This method is optimized to the set of characteristic points in reference picture simultaneously, rejects the angle point of over range, is solved
The deficiency that characteristic point over range can not be matched in the prior art, improves the reliability of matching.
It is a kind of another structural representation of target tracker of the embodiment of the present invention, target following dress referring to Fig. 7
Putting 1 includes processor 61, memory 62, input unit 63 and output device 64, the number of the processor 61 in target tracker 1
Amount can be one or more, and Fig. 7 is by taking a processor as an example.In some embodiments of the present invention, processor 61, memory
62nd, input unit 63 and output device 64 can be connected by bus or other modes, in Fig. 7 so that bus is connected as an example.
Wherein, batch processing code is stored in memory 62, and processor 61 is used to call the journey stored in memory 62
Sequence code, for performing following operation：
The Corner Feature for extracting reference picture using Predistribution Algorithm obtains the fisrt feature point set of the reference picture, institute
Stating reference picture is used to be tracked the target in current image to be matched；
The abnormity point for exceeding the image range to be matched in the fisrt feature point set is rejected, second feature point is obtained
Set；
The second feature point set is subjected to images match with the image to be matched, according to being completed matching result
The target following of image to be matched.
Further, in some embodiments of the invention, processor 61 using include Harris Corner Detection Algorithms,
The Predistribution Algorithm of FAST Corner Detection Algorithms, KLT Corner Detection Algorithms or SUSAN Corner Detection Algorithms extracts the angle of reference picture
Point feature obtains the fisrt feature point set of the reference picture.
It is preferred that, in some embodiments of the invention, processor 61 is additionally operable to perform：
The image sequence of camera acquisition is obtained, the image sequence is regard as image to be matched；
And the reference picture of the image to be matched is determined from the first two field picture of described image sequence.
It is preferred that, in some embodiments of the invention, processor 61 performs the rejecting fisrt feature point set
In exceed the abnormity point of the image range to be matched, the step of obtaining second feature point set include：
Obtain the coordinate of each characteristic point in the fisrt feature point set；
The scope of the image to be matched is seen if fall out according to the coordinate of each characteristic point in the set of characteristic points,
If it has, then determining that this feature point is abnormity point；
By abnormity point elimination all in the fisrt feature point set, second feature point set is obtained.
It is preferred that, in some embodiments of the invention, processor 61 perform it is described by the second feature point set with
The image to be matched carries out images match, and the step of completing the target following of the image to be matched according to matching result is wrapped
Include：
The subgraph to be matched of one and the reference picture formed objects are chosen in the image to be matched；
The feature for carrying the subgraph to be matched using the Predistribution Algorithm obtains third feature point set；
Calculate the distance between the second feature point set and described third feature point set；
Judge whether the distance is less than preset threshold value, if it has, then determining the reference picture and the figure to be matched
Subgraph match to be matched success as in, one and the ginseng are chosen if it has not, then re-executing in the image to be matched
The step of examining the subgraph to be matched of image formed objects.
It is preferred that, in some embodiments of the invention, processor 61 is additionally operable to the execution deletion reference picture and will
New reference picture is used as with the reference template subgraph to be matched that the match is successful.
Implement embodiments of the invention, the feature point set of the reference picture is obtained by extracting the Corner Feature of reference picture
Close, reject the abnormity point for exceeding image range to be matched in set of characteristic points, the set of characteristic points of abnormity point will be rejected with treating
Match image and carry out images match, to realize target following.Graphic Pattern Matching is carried out using Corner Feature and is retaining image graphics weight
Want that the data volume for participating in calculating can be reduced while feature, Corner Feature has good identification and stability, carried significantly
The speed of height matching.This method is optimized to the set of characteristic points in reference picture simultaneously, rejects the angle point of over range, is solved
The deficiency that characteristic point over range can not be matched in the prior art, improves the reliability of matching.
One of ordinary skill in the art will appreciate that realize all or part of flow in above-described embodiment method, being can be with
The hardware of correlation is instructed to complete by computer program, described program can be stored in a computer read/write memory medium
In, the program is upon execution, it may include such as the flow of the embodiment of above-mentioned each method.Wherein, described storage medium can be magnetic
Dish, CD, read-only memory（Read-Only Memory, ROM）Or random access memory（Random Access
Memory, RAM）Deng.
Above disclosed is only a kind of preferred embodiment of the invention, can not limit the power of the present invention with this certainly
Sharp scope, one of ordinary skill in the art will appreciate that all or part of flow of above-described embodiment is realized, and according to present invention power
Profit requires made equivalent variations, still falls within and invents covered scope.
Claims (10)
1. a kind of method for tracking target, it is characterised in that including：
The Corner Feature for extracting reference picture using Predistribution Algorithm obtains the fisrt feature point set of the reference picture, the ginseng
Examining image is used to be tracked the target in current image to be matched；
Obtain the coordinate of each characteristic point in the fisrt feature point set；According to each feature in the fisrt feature point set
The coordinate of point sees if fall out the scope of the image to be matched, if it has, then determining that this feature point is abnormity point；Will be described
All abnormity point eliminations in fisrt feature point set, obtain second feature point set；
The second feature point set is subjected to images match with the image to be matched, treated according to being completed matching result
Target following with image.
2. the method as described in claim 1, it is characterised in that the Predistribution Algorithm includes：Harris Corner Detection Algorithms,
FAST Accelerated fractionations detection feature Corner Detection Algorithm, KLT Corner Detection Algorithms or the inspection of SUSAN most small nut value similar area angle points
Method of determining and calculating.
3. method as claimed in claim 1 or 2, it is characterised in that the use Predistribution Algorithm extracts the angle point of reference picture
Before the step of feature obtains the fisrt feature point set of the reference picture, in addition to：
The image sequence of camera acquisition is obtained, the image sequence is regard as image to be matched；
And the reference picture of the image to be matched is determined from the first two field picture of described image sequence.
4. the method as described in claim 1, it is characterised in that described by the second feature point set and the figure to be matched
As carrying out images match, the step of completing the target following of the image to be matched according to matching result includes：
The subgraph to be matched of one and the reference picture formed objects are chosen in the image to be matched；
The feature for carrying the subgraph to be matched using the Predistribution Algorithm obtains third feature point set；
Calculate the distance between the second feature point set and described third feature point set；
Judge whether the distance is less than preset threshold value, if it has, then determining in the reference picture and the image to be matched
Subgraph match to be matched success, if it has not, then re-execute in the image to be matched choose one with it is described with reference to figure
The step of subgraph to be matched of picture formed objects.
5. method as claimed in claim 4, it is characterised in that described by the second feature point set and the figure to be matched
As carrying out images match, after the step of completing the target following of the image to be matched according to matching result, in addition to：
Delete the reference picture and new reference picture will be used as with the reference picture subgraph to be matched that the match is successful.
6. a kind of target tracker, it is characterised in that including：
Characteristic extracting module, the Corner Feature for extracting reference picture using Predistribution Algorithm obtains the first of the reference picture
Set of characteristic points, the reference picture is used to be tracked the target in current image to be matched；
Characteristic optimization module, the coordinate for obtaining each characteristic point in the fisrt feature point set；It is special according to described first
The coordinate for levying each characteristic point in point set sees if fall out the scope of the image to be matched, if it has, then determining the spy
Levy is a little abnormity point；By abnormity point elimination all in the fisrt feature point set, second feature point set is obtained；
Images match module, for the second feature point set to be carried out into images match with the image to be matched, according to
The target following of the image to be matched is completed with result.
7. device as claimed in claim 6, it is characterised in that the characteristic extracting module is used to extract ginseng using Predistribution Algorithm
The Corner Feature for examining image obtains the fisrt feature point set of the reference picture, and the Predistribution Algorithm is examined including Harris angle points
Method of determining and calculating, FAST Accelerated fractionations detection feature Corner Detection Algorithm, KLT Corner Detection Algorithms or the SUSAN most similar areas of small nut value
Domain Corner Detection Algorithm.
8. device as claimed in claims 6 or 7, it is characterised in that also include：
Template determining module, the image sequence for obtaining camera acquisition, regard the image sequence as image to be matched；And from
The reference picture of the image to be matched is determined in first two field picture of described image sequence.
9. device as claimed in claim 6, it is characterised in that described image matching module includes：
Unit is chosen, the son to be matched for choosing one and the reference picture formed objects in the image to be matched
Figure；
Extraction unit, the feature for extracting the subgraph to be matched using the Predistribution Algorithm obtains third feature point set；
Computing unit, for calculating the distance between the second feature point set and described third feature point set；
Matching unit, for judge the distance whether be less than preset threshold value, if it has, then determine the reference picture with it is described
Subgraph match to be matched success in image to be matched, one is chosen if it has not, then re-executing in the image to be matched
The step of with the subgraphs to be matched of the reference picture formed objects.
10. device as claimed in claim 9, it is characterised in that also include：
Template renewal module, for deleting the reference picture and making with the reference picture subgraph to be matched that the match is successful
For new reference picture.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201310183174.1A CN103279952B (en) | 2013-05-17 | 2013-05-17 | A kind of method for tracking target and device |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201310183174.1A CN103279952B (en) | 2013-05-17 | 2013-05-17 | A kind of method for tracking target and device |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN103279952A CN103279952A (en) | 2013-09-04 |
| CN103279952B true CN103279952B (en) | 2017-10-17 |
Family
ID=49062459
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201310183174.1A Active CN103279952B (en) | 2013-05-17 | 2013-05-17 | A kind of method for tracking target and device |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN103279952B (en) |
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109567600A (en) * | 2018-12-05 | 2019-04-05 | 江西书源科技有限公司 | The accessory automatic identifying method of household water-purifying machine |
Families Citing this family (13)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN104715470B (en) * | 2013-12-13 | 2017-09-22 | 南京理工大学 | A kind of klt Corner Detections device and method |
| CN105469427B (en) * | 2015-11-26 | 2018-06-19 | 河海大学 | One kind is for method for tracking target in video |
| CN107248169B (en) * | 2016-03-29 | 2021-01-22 | 中兴通讯股份有限公司 | Image positioning method and device |
| CN106250863B (en) * | 2016-08-09 | 2019-07-26 | 北京旷视科技有限公司 | Object tracking method and device |
| CN106960179B (en) * | 2017-02-24 | 2018-10-23 | 北京交通大学 | Rail line Environmental security intelligent monitoring method and device |
| CN107273801B (en) * | 2017-05-15 | 2021-11-30 | 南京邮电大学 | Method for detecting abnormal points by video multi-target tracking |
| CN107564033B (en) * | 2017-07-26 | 2020-03-03 | 北京臻迪科技股份有限公司 | Underwater target tracking method, underwater equipment and wearable equipment |
| CN108021921A (en) * | 2017-11-23 | 2018-05-11 | 塔普翊海（上海）智能科技有限公司 | Image characteristic point extraction system and its application |
| CN109919971B (en) * | 2017-12-13 | 2021-07-20 | 北京金山云网络技术有限公司 | Image processing method, image processing device, electronic equipment and computer readable storage medium |
| CN110148178B (en) | 2018-06-19 | 2022-02-22 | 腾讯科技（深圳）有限公司 | Camera positioning method, device, terminal and storage medium |
| CN109685830B (en) * | 2018-12-20 | 2021-06-15 | 浙江大华技术股份有限公司 | Target tracking method, device and equipment and computer storage medium |
| CN113409373B (en) * | 2021-06-25 | 2023-04-07 | 浙江商汤科技开发有限公司 | Image processing method, related terminal, device and storage medium |
| CN114926508B (en) * | 2022-07-21 | 2022-11-25 | 深圳市海清视讯科技有限公司 | Visual field boundary determining method, device, equipment and storage medium |
Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN101226592A (en) * | 2008-02-21 | 2008-07-23 | 上海交通大学 | Method for tracing object based on component |
| CN101399969A (en) * | 2007-09-28 | 2009-04-01 | 三星电子株式会社 | System, device and method for moving target detection and tracking based on moving camera |
| CN101840507A (en) * | 2010-04-09 | 2010-09-22 | 江苏东大金智建筑智能化系统工程有限公司 | Target tracking method based on character feature invariant and graph theory clustering |
Family Cites Families (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| JP5027741B2 (en) * | 2008-06-18 | 2012-09-19 | セコム株式会社 | Image monitoring device |
| CN102750691B (en) * | 2012-05-29 | 2014-08-06 | 重庆大学 | Corner pair-based image registration method for Cauchy-Schwarz (CS) divergence matching |
-
2013
- 2013-05-17 CN CN201310183174.1A patent/CN103279952B/en active Active
Patent Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN101399969A (en) * | 2007-09-28 | 2009-04-01 | 三星电子株式会社 | System, device and method for moving target detection and tracking based on moving camera |
| CN101226592A (en) * | 2008-02-21 | 2008-07-23 | 上海交通大学 | Method for tracing object based on component |
| CN101840507A (en) * | 2010-04-09 | 2010-09-22 | 江苏东大金智建筑智能化系统工程有限公司 | Target tracking method based on character feature invariant and graph theory clustering |
Non-Patent Citations (3)
| Title |
|---|
| 基于角点检测的实时目标跟踪方法;冯增光，张炯，宁纪锋，颜永丰;《计算机工程与设计》;20121031;第33卷(第10期);3892-3897页 * |
| 应用角点匹配实现目标跟踪;罗刚，张云峰;《中国光学与应用光学》;20091231;第2卷(第6期);479-480页 * |
| 目标跟踪过程中的遮挡问题研究;汪颖进;《中国优秀博硕士学位论文全文数据库 (硕士) 信息科技辑》;20050615(第02期);摘要以及正文第20-22,24-25,51,54页 * |
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109567600A (en) * | 2018-12-05 | 2019-04-05 | 江西书源科技有限公司 | The accessory automatic identifying method of household water-purifying machine |
Also Published As
| Publication number | Publication date |
|---|---|
| CN103279952A (en) | 2013-09-04 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN103279952B (en) | A kind of method for tracking target and device | |
| CN109543606B (en) | Human face recognition method with attention mechanism | |
| JP7490141B2 (en) | IMAGE DETECTION METHOD, MODEL TRAINING METHOD, IMAGE DETECTION APPARATUS, TRAINING APPARATUS, DEVICE, AND PROGRAM | |
| CN110414507B (en) | License plate recognition method and device, computer equipment and storage medium | |
| CN108960211B (en) | Multi-target human body posture detection method and system | |
| WO2022000862A1 (en) | Method and apparatus for detecting object in fisheye image, and storage medium | |
| Zhang et al. | A dense u-net with cross-layer intersection for detection and localization of image forgery | |
| CN108986152B (en) | Foreign matter detection method and device based on difference image | |
| CN106408597A (en) | Neighborhood entropy and consistency detection-based SAR (synthetic aperture radar) image registration method | |
| CN109886223B (en) | Face recognition method, bottom library input method and device and electronic equipment | |
| CN110033514B (en) | Reconstruction method based on point-line characteristic rapid fusion | |
| JP2011511340A (en) | Feature-based signature for image identification | |
| CN112418165B (en) | Small-size target detection method and device based on improved cascade neural network | |
| CN111783629A (en) | Human face in-vivo detection method and device for resisting sample attack | |
| CN113228105A (en) | Image processing method and device and electronic equipment | |
| CN105631860B (en) | Image point extracting method of the same name based on partial ordering's direction histogram description | |
| Yu et al. | Multibranch feature difference learning network for cross-spectral image patch matching | |
| CN107229935B (en) | Binary description method of triangle features | |
| US20220207261A1 (en) | Method and apparatus for detecting associated objects | |
| Kang et al. | An adaptive fusion panoramic image mosaic algorithm based on circular LBP feature and HSV color system | |
| CN110223319B (en) | Dynamic target real-time tracking method and system based on improved geometric particle filtering | |
| Wang et al. | Gabor filtering-based scale and rotation invariance feature for 2D barcode region detection | |
| JP2011170554A (en) | Object recognition device, object recognition method, and object recognition program | |
| CN116109891B (en) | Image data amplification method, device, computing equipment and storage medium | |
| Shen et al. | Unsupervised human skeleton extraction from Kinect depth images |
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 | ||
| TR01 | Transfer of patent right | ||
| TR01 | Transfer of patent right |
Effective date of registration: 20191218 Address after: No.1, floor 3, No.319, zhanggongshan Road, Yuhui District, Bengbu City, Anhui Province Patentee after: Bengbu guijiu Intellectual Property Service Co., Ltd Address before: 518129 Bantian HUAWEI headquarters office building, Longgang District, Guangdong, Shenzhen Patentee before: Huawei Technologies Co., Ltd. |
|
| TR01 | Transfer of patent right | ||
| TR01 | Transfer of patent right |
Effective date of registration: 20201021 Address after: C 013, C 015, C 016, C 020, C 021, C 022, 3 / F, e-commerce Industrial Park, Nantong home textile city, Jinchuan Avenue, Chuanjiang Town, Tongzhou District, Nantong City, Jiangsu Province 226000 Patentee after: Ruide Yinfang (Nantong) Information Technology Co., Ltd Address before: No.1, floor 3, No.319, zhanggongshan Road, Yuhui District, Bengbu City, Anhui Province Patentee before: Bengbu guijiu Intellectual Property Service Co.,Ltd. |